/* * Licensed to the Apache Software Foundation (ASF) under one or more * contributor license agreements. See the NOTICE file distributed with * this work for additional information regarding copyright ownership. * The ASF licenses this file to You under the Apache License, Version 2.0 * (the "License"); you may not use this file except in compliance with * the License. You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ package org.apache.commons.lang3; import java.io.UnsupportedEncodingException; import java.nio.charset.Charset; import java.text.Normalizer; import java.util.ArrayList; import java.util.Arrays; import java.util.HashSet; import java.util.Iterator; import java.util.List; import java.util.Locale; import java.util.Objects; import java.util.Set; import java.util.StringJoiner; import java.util.function.Supplier; import java.util.regex.Pattern; import net.lax1dude.eaglercraft.v1_8.HString; import org.apache.commons.lang3.function.ToBooleanBiFunction; /** *

* Operations on {@link java.lang.String} that are {@code null} safe. *

* * * *

* The {@code StringUtils} class defines certain words related to String * handling. *

* * * *

* {@code StringUtils} handles {@code null} input Strings quietly. That is to * say that a {@code null} input will return {@code null}. Where a * {@code boolean} or {@code int} is being returned details vary by method. *

* *

* A side effect of the {@code null} handling is that a * {@code NullPointerException} should be considered a bug in * {@code StringUtils}. *

* *

* Methods in this class include sample code in their Javadoc comments to * explain their operation. The symbol {@code *} is used to indicate any input * including {@code null}. *

* *

* #ThreadSafe# *

* * @see java.lang.String * @since 1.0 */ //@Immutable public class StringUtils { private static final int STRING_BUILDER_SIZE = 256; // Performance testing notes (JDK 1.4, Jul03, scolebourne) // Whitespace: // Character.isWhitespace() is faster than WHITESPACE.indexOf() // where WHITESPACE is a string of all whitespace characters // // Character access: // String.charAt(n) versus toCharArray(), then array[n] // String.charAt(n) is about 15% worse for a 10K string // They are about equal for a length 50 string // String.charAt(n) is about 4 times better for a length 3 string // String.charAt(n) is best bet overall // // Append: // String.concat about twice as fast as StringBuffer.append // (not sure who tested this) /** * A String for a space character. * * @since 3.2 */ public static final String SPACE = " "; /** * The empty String {@code ""}. * * @since 2.0 */ public static final String EMPTY = ""; /** * A String for linefeed LF ("\n"). * * @see JLF: * Escape Sequences for Character and String Literals * @since 3.2 */ public static final String LF = "\n"; /** * A String for carriage return CR ("\r"). * * @see JLF: * Escape Sequences for Character and String Literals * @since 3.2 */ public static final String CR = "\r"; /** * Represents a failed index search. * * @since 2.1 */ public static final int INDEX_NOT_FOUND = -1; /** *

* The maximum size to which the padding constant(s) can expand. *

*/ private static final int PAD_LIMIT = 8192; /** * Pattern used in {@link #stripAccents(String)}. */ private static final Pattern STRIP_ACCENTS_PATTERN = Pattern.compile("\\p{InCombiningDiacriticalMarks}+"); //$NON-NLS-1$ /** *

* Abbreviates a String using ellipses. This will turn "Now is the time for all * good men" into "Now is the time for..." *

* *

* Specifically: *

* * *
	 * StringUtils.abbreviate(null, *)      = null
	 * StringUtils.abbreviate("", 4)        = ""
	 * StringUtils.abbreviate("abcdefg", 6) = "abc..."
	 * StringUtils.abbreviate("abcdefg", 7) = "abcdefg"
	 * StringUtils.abbreviate("abcdefg", 8) = "abcdefg"
	 * StringUtils.abbreviate("abcdefg", 4) = "a..."
	 * StringUtils.abbreviate("abcdefg", 3) = IllegalArgumentException
	 * 
* * @param str the String to check, may be null * @param maxWidth maximum length of result String, must be at least 4 * @return abbreviated String, {@code null} if null String input * @throws IllegalArgumentException if the width is too small * @since 2.0 */ public static String abbreviate(final String str, final int maxWidth) { return abbreviate(str, "...", 0, maxWidth); } /** *

* Abbreviates a String using ellipses. This will turn "Now is the time for all * good men" into "...is the time for..." *

* *

* Works like {@code abbreviate(String, int)}, but allows you to specify a "left * edge" offset. Note that this left edge is not necessarily going to be the * leftmost character in the result, or the first character following the * ellipses, but it will appear somewhere in the result. * *

* In no case will it return a String of length greater than {@code maxWidth}. *

* *
	 * StringUtils.abbreviate(null, *, *)                = null
	 * StringUtils.abbreviate("", 0, 4)                  = ""
	 * StringUtils.abbreviate("abcdefghijklmno", -1, 10) = "abcdefg..."
	 * StringUtils.abbreviate("abcdefghijklmno", 0, 10)  = "abcdefg..."
	 * StringUtils.abbreviate("abcdefghijklmno", 1, 10)  = "abcdefg..."
	 * StringUtils.abbreviate("abcdefghijklmno", 4, 10)  = "abcdefg..."
	 * StringUtils.abbreviate("abcdefghijklmno", 5, 10)  = "...fghi..."
	 * StringUtils.abbreviate("abcdefghijklmno", 6, 10)  = "...ghij..."
	 * StringUtils.abbreviate("abcdefghijklmno", 8, 10)  = "...ijklmno"
	 * StringUtils.abbreviate("abcdefghijklmno", 10, 10) = "...ijklmno"
	 * StringUtils.abbreviate("abcdefghijklmno", 12, 10) = "...ijklmno"
	 * StringUtils.abbreviate("abcdefghij", 0, 3)        = IllegalArgumentException
	 * StringUtils.abbreviate("abcdefghij", 5, 6)        = IllegalArgumentException
	 * 
* * @param str the String to check, may be null * @param offset left edge of source String * @param maxWidth maximum length of result String, must be at least 4 * @return abbreviated String, {@code null} if null String input * @throws IllegalArgumentException if the width is too small * @since 2.0 */ public static String abbreviate(final String str, final int offset, final int maxWidth) { return abbreviate(str, "...", offset, maxWidth); } /** *

* Abbreviates a String using another given String as replacement marker. This * will turn "Now is the time for all good men" into "Now is the time for..." if * "..." was defined as the replacement marker. *

* *

* Specifically: *

* * *
	 * StringUtils.abbreviate(null, "...", *)      = null
	 * StringUtils.abbreviate("abcdefg", null, *)  = "abcdefg"
	 * StringUtils.abbreviate("", "...", 4)        = ""
	 * StringUtils.abbreviate("abcdefg", ".", 5)   = "abcd."
	 * StringUtils.abbreviate("abcdefg", ".", 7)   = "abcdefg"
	 * StringUtils.abbreviate("abcdefg", ".", 8)   = "abcdefg"
	 * StringUtils.abbreviate("abcdefg", "..", 4)  = "ab.."
	 * StringUtils.abbreviate("abcdefg", "..", 3)  = "a.."
	 * StringUtils.abbreviate("abcdefg", "..", 2)  = IllegalArgumentException
	 * StringUtils.abbreviate("abcdefg", "...", 3) = IllegalArgumentException
	 * 
* * @param str the String to check, may be null * @param abbrevMarker the String used as replacement marker * @param maxWidth maximum length of result String, must be at least * {@code abbrevMarker.length + 1} * @return abbreviated String, {@code null} if null String input * @throws IllegalArgumentException if the width is too small * @since 3.6 */ public static String abbreviate(final String str, final String abbrevMarker, final int maxWidth) { return abbreviate(str, abbrevMarker, 0, maxWidth); } /** *

* Abbreviates a String using a given replacement marker. This will turn "Now is * the time for all good men" into "...is the time for..." if "..." was defined * as the replacement marker. *

* *

* Works like {@code abbreviate(String, String, int)}, but allows you to specify * a "left edge" offset. Note that this left edge is not necessarily going to be * the leftmost character in the result, or the first character following the * replacement marker, but it will appear somewhere in the result. * *

* In no case will it return a String of length greater than {@code maxWidth}. *

* *
	 * StringUtils.abbreviate(null, null, *, *)                 = null
	 * StringUtils.abbreviate("abcdefghijklmno", null, *, *)    = "abcdefghijklmno"
	 * StringUtils.abbreviate("", "...", 0, 4)                  = ""
	 * StringUtils.abbreviate("abcdefghijklmno", "---", -1, 10) = "abcdefg---"
	 * StringUtils.abbreviate("abcdefghijklmno", ",", 0, 10)    = "abcdefghi,"
	 * StringUtils.abbreviate("abcdefghijklmno", ",", 1, 10)    = "abcdefghi,"
	 * StringUtils.abbreviate("abcdefghijklmno", ",", 2, 10)    = "abcdefghi,"
	 * StringUtils.abbreviate("abcdefghijklmno", "::", 4, 10)   = "::efghij::"
	 * StringUtils.abbreviate("abcdefghijklmno", "...", 6, 10)  = "...ghij..."
	 * StringUtils.abbreviate("abcdefghijklmno", "*", 9, 10)    = "*ghijklmno"
	 * StringUtils.abbreviate("abcdefghijklmno", "'", 10, 10)   = "'ghijklmno"
	 * StringUtils.abbreviate("abcdefghijklmno", "!", 12, 10)   = "!ghijklmno"
	 * StringUtils.abbreviate("abcdefghij", "abra", 0, 4)       = IllegalArgumentException
	 * StringUtils.abbreviate("abcdefghij", "...", 5, 6)        = IllegalArgumentException
	 * 
* * @param str the String to check, may be null * @param abbrevMarker the String used as replacement marker * @param offset left edge of source String * @param maxWidth maximum length of result String, must be at least 4 * @return abbreviated String, {@code null} if null String input * @throws IllegalArgumentException if the width is too small * @since 3.6 */ public static String abbreviate(final String str, final String abbrevMarker, int offset, final int maxWidth) { if (isNotEmpty(str) && EMPTY.equals(abbrevMarker) && maxWidth > 0) { return substring(str, 0, maxWidth); } else if (isAnyEmpty(str, abbrevMarker)) { return str; } final int abbrevMarkerLength = abbrevMarker.length(); final int minAbbrevWidth = abbrevMarkerLength + 1; final int minAbbrevWidthOffset = abbrevMarkerLength + abbrevMarkerLength + 1; if (maxWidth < minAbbrevWidth) { throw new IllegalArgumentException(HString.format("Minimum abbreviation width is %d", minAbbrevWidth)); } final int strLen = str.length(); if (strLen <= maxWidth) { return str; } if (offset > strLen) { offset = strLen; } if (strLen - offset < maxWidth - abbrevMarkerLength) { offset = strLen - (maxWidth - abbrevMarkerLength); } if (offset <= abbrevMarkerLength + 1) { return str.substring(0, maxWidth - abbrevMarkerLength) + abbrevMarker; } if (maxWidth < minAbbrevWidthOffset) { throw new IllegalArgumentException( HString.format("Minimum abbreviation width with offset is %d", minAbbrevWidthOffset)); } if (offset + maxWidth - abbrevMarkerLength < strLen) { return abbrevMarker + abbreviate(str.substring(offset), abbrevMarker, maxWidth - abbrevMarkerLength); } return abbrevMarker + str.substring(strLen - (maxWidth - abbrevMarkerLength)); } /** *

* Abbreviates a String to the length passed, replacing the middle characters * with the supplied replacement String. *

* *

* This abbreviation only occurs if the following criteria is met: *

* *

* Otherwise, the returned String will be the same as the supplied String for * abbreviation. *

* *
	 * StringUtils.abbreviateMiddle(null, null, 0)      = null
	 * StringUtils.abbreviateMiddle("abc", null, 0)      = "abc"
	 * StringUtils.abbreviateMiddle("abc", ".", 0)      = "abc"
	 * StringUtils.abbreviateMiddle("abc", ".", 3)      = "abc"
	 * StringUtils.abbreviateMiddle("abcdef", ".", 4)     = "ab.f"
	 * 
* * @param str the String to abbreviate, may be null * @param middle the String to replace the middle characters with, may be null * @param length the length to abbreviate {@code str} to. * @return the abbreviated String if the above criteria is met, or the original * String supplied for abbreviation. * @since 2.5 */ public static String abbreviateMiddle(final String str, final String middle, final int length) { if (isAnyEmpty(str, middle) || length >= str.length() || length < middle.length() + 2) { return str; } final int targetSting = length - middle.length(); final int startOffset = targetSting / 2 + targetSting % 2; final int endOffset = str.length() - targetSting / 2; return str.substring(0, startOffset) + middle + str.substring(endOffset); } /** *

* Capitalizes a String changing the first character to title case as per * {@link Character#toTitleCase(int)}. No other characters are changed. *

* *

* For a word based algorithm, see * {@link org.apache.commons.lang3.text.WordUtils#capitalize(String)}. A * {@code null} input String returns {@code null}. *

* *
	 * StringUtils.capitalize(null)  = null
	 * StringUtils.capitalize("")    = ""
	 * StringUtils.capitalize("cat") = "Cat"
	 * StringUtils.capitalize("cAt") = "CAt"
	 * StringUtils.capitalize("'cat'") = "'cat'"
	 * 
* * @param str the String to capitalize, may be null * @return the capitalized String, {@code null} if null String input * @see org.apache.commons.lang3.text.WordUtils#capitalize(String) * @see #uncapitalize(String) * @since 2.0 */ public static String capitalize(final String str) { final int strLen = length(str); if (strLen == 0) { return str; } final int firstCodepoint = str.codePointAt(0); final int newCodePoint = Character.toTitleCase(firstCodepoint); if (firstCodepoint == newCodePoint) { // already capitalized return str; } final int[] newCodePoints = new int[strLen]; // cannot be longer than the char array int outOffset = 0; newCodePoints[outOffset++] = newCodePoint; // copy the first codepoint for (int inOffset = Character.charCount(firstCodepoint); inOffset < strLen;) { final int codepoint = str.codePointAt(inOffset); newCodePoints[outOffset++] = codepoint; // copy the remaining ones inOffset += Character.charCount(codepoint); } return new String(newCodePoints, 0, outOffset); } /** *

* Centers a String in a larger String of size {@code size} using the space * character (' '). *

* *

* If the size is less than the String length, the original String is returned. * A {@code null} String returns {@code null}. A negative size is treated as * zero. *

* *

* Equivalent to {@code center(str, size, " ")}. *

* *
	 * StringUtils.center(null, *)   = null
	 * StringUtils.center("", 4)     = "    "
	 * StringUtils.center("ab", -1)  = "ab"
	 * StringUtils.center("ab", 4)   = " ab "
	 * StringUtils.center("abcd", 2) = "abcd"
	 * StringUtils.center("a", 4)    = " a  "
	 * 
* * @param str the String to center, may be null * @param size the int size of new String, negative treated as zero * @return centered String, {@code null} if null String input */ public static String center(final String str, final int size) { return center(str, size, ' '); } /** *

* Centers a String in a larger String of size {@code size}. Uses a supplied * character as the value to pad the String with. *

* *

* If the size is less than the String length, the String is returned. A * {@code null} String returns {@code null}. A negative size is treated as zero. *

* *
	 * StringUtils.center(null, *, *)     = null
	 * StringUtils.center("", 4, ' ')     = "    "
	 * StringUtils.center("ab", -1, ' ')  = "ab"
	 * StringUtils.center("ab", 4, ' ')   = " ab "
	 * StringUtils.center("abcd", 2, ' ') = "abcd"
	 * StringUtils.center("a", 4, ' ')    = " a  "
	 * StringUtils.center("a", 4, 'y')    = "yayy"
	 * 
* * @param str the String to center, may be null * @param size the int size of new String, negative treated as zero * @param padChar the character to pad the new String with * @return centered String, {@code null} if null String input * @since 2.0 */ public static String center(String str, final int size, final char padChar) { if (str == null || size <= 0) { return str; } final int strLen = str.length(); final int pads = size - strLen; if (pads <= 0) { return str; } str = leftPad(str, strLen + pads / 2, padChar); str = rightPad(str, size, padChar); return str; } /** *

* Centers a String in a larger String of size {@code size}. Uses a supplied * String as the value to pad the String with. *

* *

* If the size is less than the String length, the String is returned. A * {@code null} String returns {@code null}. A negative size is treated as zero. *

* *
	 * StringUtils.center(null, *, *)     = null
	 * StringUtils.center("", 4, " ")     = "    "
	 * StringUtils.center("ab", -1, " ")  = "ab"
	 * StringUtils.center("ab", 4, " ")   = " ab "
	 * StringUtils.center("abcd", 2, " ") = "abcd"
	 * StringUtils.center("a", 4, " ")    = " a  "
	 * StringUtils.center("a", 4, "yz")   = "yayz"
	 * StringUtils.center("abc", 7, null) = "  abc  "
	 * StringUtils.center("abc", 7, "")   = "  abc  "
	 * 
* * @param str the String to center, may be null * @param size the int size of new String, negative treated as zero * @param padStr the String to pad the new String with, must not be null or * empty * @return centered String, {@code null} if null String input * @throws IllegalArgumentException if padStr is {@code null} or empty */ public static String center(String str, final int size, String padStr) { if (str == null || size <= 0) { return str; } if (isEmpty(padStr)) { padStr = SPACE; } final int strLen = str.length(); final int pads = size - strLen; if (pads <= 0) { return str; } str = leftPad(str, strLen + pads / 2, padStr); str = rightPad(str, size, padStr); return str; } /** *

* Removes one newline from end of a String if it's there, otherwise leave it * alone. A newline is "{@code \n}", "{@code \r}", or * "{@code \r\n}". *

* *

* NOTE: This method changed in 2.0. It now more closely matches Perl chomp. *

* *
	 * StringUtils.chomp(null)          = null
	 * StringUtils.chomp("")            = ""
	 * StringUtils.chomp("abc \r")      = "abc "
	 * StringUtils.chomp("abc\n")       = "abc"
	 * StringUtils.chomp("abc\r\n")     = "abc"
	 * StringUtils.chomp("abc\r\n\r\n") = "abc\r\n"
	 * StringUtils.chomp("abc\n\r")     = "abc\n"
	 * StringUtils.chomp("abc\n\rabc")  = "abc\n\rabc"
	 * StringUtils.chomp("\r")          = ""
	 * StringUtils.chomp("\n")          = ""
	 * StringUtils.chomp("\r\n")        = ""
	 * 
* * @param str the String to chomp a newline from, may be null * @return String without newline, {@code null} if null String input */ public static String chomp(final String str) { if (isEmpty(str)) { return str; } if (str.length() == 1) { final char ch = str.charAt(0); if (ch == CharUtils.CR || ch == CharUtils.LF) { return EMPTY; } return str; } int lastIdx = str.length() - 1; final char last = str.charAt(lastIdx); if (last == CharUtils.LF) { if (str.charAt(lastIdx - 1) == CharUtils.CR) { lastIdx--; } } else if (last != CharUtils.CR) { lastIdx++; } return str.substring(0, lastIdx); } /** *

* Removes {@code separator} from the end of {@code str} if it's there, * otherwise leave it alone. *

* *

* NOTE: This method changed in version 2.0. It now more closely matches Perl * chomp. For the previous behavior, use * {@link #substringBeforeLast(String, String)}. This method uses * {@link String#endsWith(String)}. *

* *
	 * StringUtils.chomp(null, *)         = null
	 * StringUtils.chomp("", *)           = ""
	 * StringUtils.chomp("foobar", "bar") = "foo"
	 * StringUtils.chomp("foobar", "baz") = "foobar"
	 * StringUtils.chomp("foo", "foo")    = ""
	 * StringUtils.chomp("foo ", "foo")   = "foo "
	 * StringUtils.chomp(" foo", "foo")   = " "
	 * StringUtils.chomp("foo", "foooo")  = "foo"
	 * StringUtils.chomp("foo", "")       = "foo"
	 * StringUtils.chomp("foo", null)     = "foo"
	 * 
* * @param str the String to chomp from, may be null * @param separator separator String, may be null * @return String without trailing separator, {@code null} if null String input * @deprecated This feature will be removed in Lang 4.0, use * {@link StringUtils#removeEnd(String, String)} instead */ @Deprecated public static String chomp(final String str, final String separator) { return removeEnd(str, separator); } /** *

* Remove the last character from a String. *

* *

* If the String ends in {@code \r\n}, then remove both of them. *

* *
	 * StringUtils.chop(null)          = null
	 * StringUtils.chop("")            = ""
	 * StringUtils.chop("abc \r")      = "abc "
	 * StringUtils.chop("abc\n")       = "abc"
	 * StringUtils.chop("abc\r\n")     = "abc"
	 * StringUtils.chop("abc")         = "ab"
	 * StringUtils.chop("abc\nabc")    = "abc\nab"
	 * StringUtils.chop("a")           = ""
	 * StringUtils.chop("\r")          = ""
	 * StringUtils.chop("\n")          = ""
	 * StringUtils.chop("\r\n")        = ""
	 * 
* * @param str the String to chop last character from, may be null * @return String without last character, {@code null} if null String input */ public static String chop(final String str) { if (str == null) { return null; } final int strLen = str.length(); if (strLen < 2) { return EMPTY; } final int lastIdx = strLen - 1; final String ret = str.substring(0, lastIdx); final char last = str.charAt(lastIdx); if (last == CharUtils.LF && ret.charAt(lastIdx - 1) == CharUtils.CR) { return ret.substring(0, lastIdx - 1); } return ret; } /** *

* Compare two Strings lexicographically, as per * {@link String#compareTo(String)}, returning : *

* * *

* This is a {@code null} safe version of : *

*
* *
	 * str1.compareTo(str2)
	 * 
* *
* *

* {@code null} value is considered less than non-{@code null} value. Two * {@code null} references are considered equal. *

* *
	 * StringUtils.compare(null, null)   = 0
	 * StringUtils.compare(null , "a")   < 0
	 * StringUtils.compare("a", null)    > 0
	 * StringUtils.compare("abc", "abc") = 0
	 * StringUtils.compare("a", "b")     < 0
	 * StringUtils.compare("b", "a")     > 0
	 * StringUtils.compare("a", "B")     > 0
	 * StringUtils.compare("ab", "abc")  < 0
	 * 
* * @see #compare(String, String, boolean) * @see String#compareTo(String) * @param str1 the String to compare from * @param str2 the String to compare to * @return < 0, 0, > 0, if {@code str1} is respectively less, equal or * greater than {@code str2} * @since 3.5 */ public static int compare(final String str1, final String str2) { return compare(str1, str2, true); } /** *

* Compare two Strings lexicographically, as per * {@link String#compareTo(String)}, returning : *

* * *

* This is a {@code null} safe version of : *

*
* *
	 * str1.compareTo(str2)
	 * 
* *
* *

* {@code null} inputs are handled according to the {@code nullIsLess} * parameter. Two {@code null} references are considered equal. *

* *
	 * StringUtils.compare(null, null, *)     = 0
	 * StringUtils.compare(null , "a", true)  < 0
	 * StringUtils.compare(null , "a", false) > 0
	 * StringUtils.compare("a", null, true)   > 0
	 * StringUtils.compare("a", null, false)  < 0
	 * StringUtils.compare("abc", "abc", *)   = 0
	 * StringUtils.compare("a", "b", *)       < 0
	 * StringUtils.compare("b", "a", *)       > 0
	 * StringUtils.compare("a", "B", *)       > 0
	 * StringUtils.compare("ab", "abc", *)    < 0
	 * 
* * @see String#compareTo(String) * @param str1 the String to compare from * @param str2 the String to compare to * @param nullIsLess whether consider {@code null} value less than * non-{@code null} value * @return < 0, 0, > 0, if {@code str1} is respectively less, equal ou * greater than {@code str2} * @since 3.5 */ public static int compare(final String str1, final String str2, final boolean nullIsLess) { if (str1 == str2) { // NOSONARLINT this intentionally uses == to allow for both null return 0; } if (str1 == null) { return nullIsLess ? -1 : 1; } if (str2 == null) { return nullIsLess ? 1 : -1; } return str1.compareTo(str2); } /** *

* Compare two Strings lexicographically, ignoring case differences, as per * {@link String#compareToIgnoreCase(String)}, returning : *

* * *

* This is a {@code null} safe version of : *

*
* *
	 * str1.compareToIgnoreCase(str2)
	 * 
* *
* *

* {@code null} value is considered less than non-{@code null} value. Two * {@code null} references are considered equal. Comparison is case insensitive. *

* *
	 * StringUtils.compareIgnoreCase(null, null)   = 0
	 * StringUtils.compareIgnoreCase(null , "a")   < 0
	 * StringUtils.compareIgnoreCase("a", null)    > 0
	 * StringUtils.compareIgnoreCase("abc", "abc") = 0
	 * StringUtils.compareIgnoreCase("abc", "ABC") = 0
	 * StringUtils.compareIgnoreCase("a", "b")     < 0
	 * StringUtils.compareIgnoreCase("b", "a")     > 0
	 * StringUtils.compareIgnoreCase("a", "B")     < 0
	 * StringUtils.compareIgnoreCase("A", "b")     < 0
	 * StringUtils.compareIgnoreCase("ab", "ABC")  < 0
	 * 
* * @see #compareIgnoreCase(String, String, boolean) * @see String#compareToIgnoreCase(String) * @param str1 the String to compare from * @param str2 the String to compare to * @return < 0, 0, > 0, if {@code str1} is respectively less, equal ou * greater than {@code str2}, ignoring case differences. * @since 3.5 */ public static int compareIgnoreCase(final String str1, final String str2) { return compareIgnoreCase(str1, str2, true); } /** *

* Compare two Strings lexicographically, ignoring case differences, as per * {@link String#compareToIgnoreCase(String)}, returning : *

* * *

* This is a {@code null} safe version of : *

*
* *
	 * str1.compareToIgnoreCase(str2)
	 * 
* *
* *

* {@code null} inputs are handled according to the {@code nullIsLess} * parameter. Two {@code null} references are considered equal. Comparison is * case insensitive. *

* *
	 * StringUtils.compareIgnoreCase(null, null, *)     = 0
	 * StringUtils.compareIgnoreCase(null , "a", true)  < 0
	 * StringUtils.compareIgnoreCase(null , "a", false) > 0
	 * StringUtils.compareIgnoreCase("a", null, true)   > 0
	 * StringUtils.compareIgnoreCase("a", null, false)  < 0
	 * StringUtils.compareIgnoreCase("abc", "abc", *)   = 0
	 * StringUtils.compareIgnoreCase("abc", "ABC", *)   = 0
	 * StringUtils.compareIgnoreCase("a", "b", *)       < 0
	 * StringUtils.compareIgnoreCase("b", "a", *)       > 0
	 * StringUtils.compareIgnoreCase("a", "B", *)       < 0
	 * StringUtils.compareIgnoreCase("A", "b", *)       < 0
	 * StringUtils.compareIgnoreCase("ab", "abc", *)    < 0
	 * 
* * @see String#compareToIgnoreCase(String) * @param str1 the String to compare from * @param str2 the String to compare to * @param nullIsLess whether consider {@code null} value less than * non-{@code null} value * @return < 0, 0, > 0, if {@code str1} is respectively less, equal ou * greater than {@code str2}, ignoring case differences. * @since 3.5 */ public static int compareIgnoreCase(final String str1, final String str2, final boolean nullIsLess) { if (str1 == str2) { // NOSONARLINT this intentionally uses == to allow for both null return 0; } if (str1 == null) { return nullIsLess ? -1 : 1; } if (str2 == null) { return nullIsLess ? 1 : -1; } return str1.compareToIgnoreCase(str2); } /** *

* Checks if CharSequence contains a search CharSequence, handling {@code null}. * This method uses {@link String#indexOf(String)} if possible. *

* *

* A {@code null} CharSequence will return {@code false}. *

* *
	 * StringUtils.contains(null, *)     = false
	 * StringUtils.contains(*, null)     = false
	 * StringUtils.contains("", "")      = true
	 * StringUtils.contains("abc", "")   = true
	 * StringUtils.contains("abc", "a")  = true
	 * StringUtils.contains("abc", "z")  = false
	 * 
* * @param seq the CharSequence to check, may be null * @param searchSeq the CharSequence to find, may be null * @return true if the CharSequence contains the search CharSequence, false if * not or {@code null} string input * @since 2.0 * @since 3.0 Changed signature from contains(String, String) to * contains(CharSequence, CharSequence) */ public static boolean contains(final CharSequence seq, final CharSequence searchSeq) { if (seq == null || searchSeq == null) { return false; } return CharSequenceUtils.indexOf(seq, searchSeq, 0) >= 0; } /** *

* Checks if CharSequence contains a search character, handling {@code null}. * This method uses {@link String#indexOf(int)} if possible. *

* *

* A {@code null} or empty ("") CharSequence will return {@code false}. *

* *
	 * StringUtils.contains(null, *)    = false
	 * StringUtils.contains("", *)      = false
	 * StringUtils.contains("abc", 'a') = true
	 * StringUtils.contains("abc", 'z') = false
	 * 
* * @param seq the CharSequence to check, may be null * @param searchChar the character to find * @return true if the CharSequence contains the search character, false if not * or {@code null} string input * @since 2.0 * @since 3.0 Changed signature from contains(String, int) to * contains(CharSequence, int) */ public static boolean contains(final CharSequence seq, final int searchChar) { if (isEmpty(seq)) { return false; } return CharSequenceUtils.indexOf(seq, searchChar, 0) >= 0; } /** *

* Checks if the CharSequence contains any character in the given set of * characters. *

* *

* A {@code null} CharSequence will return {@code false}. A {@code null} or zero * length search array will return {@code false}. *

* *
	 * StringUtils.containsAny(null, *)                  = false
	 * StringUtils.containsAny("", *)                    = false
	 * StringUtils.containsAny(*, null)                  = false
	 * StringUtils.containsAny(*, [])                    = false
	 * StringUtils.containsAny("zzabyycdxx", ['z', 'a']) = true
	 * StringUtils.containsAny("zzabyycdxx", ['b', 'y']) = true
	 * StringUtils.containsAny("zzabyycdxx", ['z', 'y']) = true
	 * StringUtils.containsAny("aba", ['z'])             = false
	 * 
* * @param cs the CharSequence to check, may be null * @param searchChars the chars to search for, may be null * @return the {@code true} if any of the chars are found, {@code false} if no * match or null input * @since 2.4 * @since 3.0 Changed signature from containsAny(String, char[]) to * containsAny(CharSequence, char...) */ public static boolean containsAny(final CharSequence cs, final char... searchChars) { if (isEmpty(cs) || searchChars.length == 0) { return false; } final int csLength = cs.length(); final int searchLength = searchChars.length; final int csLast = csLength - 1; final int searchLast = searchLength - 1; for (int i = 0; i < csLength; i++) { final char ch = cs.charAt(i); for (int j = 0; j < searchLength; j++) { if (searchChars[j] == ch) { if (Character.isHighSurrogate(ch)) { if (j == searchLast) { // missing low surrogate, fine, like String.indexOf(String) return true; } if (i < csLast && searchChars[j + 1] == cs.charAt(i + 1)) { return true; } } else { // ch is in the Basic Multilingual Plane return true; } } } } return false; } /** *

* Checks if the CharSequence contains any character in the given set of * characters. *

* *

* A {@code null} CharSequence will return {@code false}. A {@code null} search * CharSequence will return {@code false}. *

* *
	 * StringUtils.containsAny(null, *)               = false
	 * StringUtils.containsAny("", *)                 = false
	 * StringUtils.containsAny(*, null)               = false
	 * StringUtils.containsAny(*, "")                 = false
	 * StringUtils.containsAny("zzabyycdxx", "za")    = true
	 * StringUtils.containsAny("zzabyycdxx", "by")    = true
	 * StringUtils.containsAny("zzabyycdxx", "zy")    = true
	 * StringUtils.containsAny("zzabyycdxx", "\tx")   = true
	 * StringUtils.containsAny("zzabyycdxx", "$.#yF") = true
	 * StringUtils.containsAny("aba", "z")            = false
	 * 
* * @param cs the CharSequence to check, may be null * @param searchChars the chars to search for, may be null * @return the {@code true} if any of the chars are found, {@code false} if no * match or null input * @since 2.4 * @since 3.0 Changed signature from containsAny(String, String) to * containsAny(CharSequence, CharSequence) */ public static boolean containsAny(final CharSequence cs, final CharSequence searchChars) { if (searchChars == null) { return false; } return containsAny(cs, CharSequenceUtils.toCharArray(searchChars)); } /** *

* Checks if the CharSequence contains any of the CharSequences in the given * array. *

* *

* A {@code null} {@code cs} CharSequence will return {@code false}. A * {@code null} or zero length search array will return {@code false}. *

* *
	 * StringUtils.containsAny(null, *)            = false
	 * StringUtils.containsAny("", *)              = false
	 * StringUtils.containsAny(*, null)            = false
	 * StringUtils.containsAny(*, [])              = false
	 * StringUtils.containsAny("abcd", "ab", null) = true
	 * StringUtils.containsAny("abcd", "ab", "cd") = true
	 * StringUtils.containsAny("abc", "d", "abc")  = true
	 * 
* * * @param cs The CharSequence to check, may be null * @param searchCharSequences The array of CharSequences to search for, may be * null. Individual CharSequences may be null as * well. * @return {@code true} if any of the search CharSequences are found, * {@code false} otherwise * @since 3.4 */ public static boolean containsAny(final CharSequence cs, final CharSequence... searchCharSequences) { return containsAny(StringUtils::contains, cs, searchCharSequences); } /** *

* Checks if the CharSequence contains any of the CharSequences in the given * array. *

* *

* A {@code null} {@code cs} CharSequence will return {@code false}. A * {@code null} or zero length search array will return {@code false}. *

* * @param cs The CharSequence to check, may be null * @param searchCharSequences The array of CharSequences to search for, may be * null. Individual CharSequences may be null as * well. * @return {@code true} if any of the search CharSequences are found, * {@code false} otherwise * @since 3.12.0 */ private static boolean containsAny(final ToBooleanBiFunction test, final CharSequence cs, final CharSequence... searchCharSequences) { if (isEmpty(cs) || searchCharSequences.length == 0) { return false; } for (final CharSequence searchCharSequence : searchCharSequences) { if (test.applyAsBoolean(cs, searchCharSequence)) { return true; } } return false; } /** *

* Checks if the CharSequence contains any of the CharSequences in the given * array, ignoring case. *

* *

* A {@code null} {@code cs} CharSequence will return {@code false}. A * {@code null} or zero length search array will return {@code false}. *

* *
	 * StringUtils.containsAny(null, *)            = false
	 * StringUtils.containsAny("", *)              = false
	 * StringUtils.containsAny(*, null)            = false
	 * StringUtils.containsAny(*, [])              = false
	 * StringUtils.containsAny("abcd", "ab", null) = true
	 * StringUtils.containsAny("abcd", "ab", "cd") = true
	 * StringUtils.containsAny("abc", "d", "abc")  = true
	 * StringUtils.containsAny("abc", "D", "ABC")  = true
	 * StringUtils.containsAny("ABC", "d", "abc")  = true
	 * 
* * * @param cs The CharSequence to check, may be null * @param searchCharSequences The array of CharSequences to search for, may be * null. Individual CharSequences may be null as * well. * @return {@code true} if any of the search CharSequences are found, * {@code false} otherwise * @since 3.12.0 */ public static boolean containsAnyIgnoreCase(final CharSequence cs, final CharSequence... searchCharSequences) { return containsAny(StringUtils::containsIgnoreCase, cs, searchCharSequences); } /** *

* Checks if CharSequence contains a search CharSequence irrespective of case, * handling {@code null}. Case-insensitivity is defined as by * {@link String#equalsIgnoreCase(String)}. * *

* A {@code null} CharSequence will return {@code false}. *

* *
	 * StringUtils.containsIgnoreCase(null, *) = false
	 * StringUtils.containsIgnoreCase(*, null) = false
	 * StringUtils.containsIgnoreCase("", "") = true
	 * StringUtils.containsIgnoreCase("abc", "") = true
	 * StringUtils.containsIgnoreCase("abc", "a") = true
	 * StringUtils.containsIgnoreCase("abc", "z") = false
	 * StringUtils.containsIgnoreCase("abc", "A") = true
	 * StringUtils.containsIgnoreCase("abc", "Z") = false
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @return true if the CharSequence contains the search CharSequence * irrespective of case or false if not or {@code null} string input * @since 3.0 Changed signature from containsIgnoreCase(String, String) to * containsIgnoreCase(CharSequence, CharSequence) */ public static boolean containsIgnoreCase(final CharSequence str, final CharSequence searchStr) { if (str == null || searchStr == null) { return false; } final int len = searchStr.length(); final int max = str.length() - len; for (int i = 0; i <= max; i++) { if (CharSequenceUtils.regionMatches(str, true, i, searchStr, 0, len)) { return true; } } return false; } /** *

* Checks that the CharSequence does not contain certain characters. *

* *

* A {@code null} CharSequence will return {@code true}. A {@code null} invalid * character array will return {@code true}. An empty CharSequence (length()=0) * always returns true. *

* *
	 * StringUtils.containsNone(null, *)       = true
	 * StringUtils.containsNone(*, null)       = true
	 * StringUtils.containsNone("", *)         = true
	 * StringUtils.containsNone("ab", '')      = true
	 * StringUtils.containsNone("abab", 'xyz') = true
	 * StringUtils.containsNone("ab1", 'xyz')  = true
	 * StringUtils.containsNone("abz", 'xyz')  = false
	 * 
* * @param cs the CharSequence to check, may be null * @param searchChars an array of invalid chars, may be null * @return true if it contains none of the invalid chars, or is null * @since 2.0 * @since 3.0 Changed signature from containsNone(String, char[]) to * containsNone(CharSequence, char...) */ public static boolean containsNone(final CharSequence cs, final char... searchChars) { if (cs == null || searchChars == null) { return true; } final int csLen = cs.length(); final int csLast = csLen - 1; final int searchLen = searchChars.length; final int searchLast = searchLen - 1; for (int i = 0; i < csLen; i++) { final char ch = cs.charAt(i); for (int j = 0; j < searchLen; j++) { if (searchChars[j] == ch) { if (Character.isHighSurrogate(ch)) { if (j == searchLast) { // missing low surrogate, fine, like String.indexOf(String) return false; } if (i < csLast && searchChars[j + 1] == cs.charAt(i + 1)) { return false; } } else { // ch is in the Basic Multilingual Plane return false; } } } } return true; } /** *

* Checks that the CharSequence does not contain certain characters. *

* *

* A {@code null} CharSequence will return {@code true}. A {@code null} invalid * character array will return {@code true}. An empty String ("") always returns * true. *

* *
	 * StringUtils.containsNone(null, *)       = true
	 * StringUtils.containsNone(*, null)       = true
	 * StringUtils.containsNone("", *)         = true
	 * StringUtils.containsNone("ab", "")      = true
	 * StringUtils.containsNone("abab", "xyz") = true
	 * StringUtils.containsNone("ab1", "xyz")  = true
	 * StringUtils.containsNone("abz", "xyz")  = false
	 * 
* * @param cs the CharSequence to check, may be null * @param invalidChars a String of invalid chars, may be null * @return true if it contains none of the invalid chars, or is null * @since 2.0 * @since 3.0 Changed signature from containsNone(String, String) to * containsNone(CharSequence, String) */ public static boolean containsNone(final CharSequence cs, final String invalidChars) { if (invalidChars == null) { return true; } return containsNone(cs, invalidChars.toCharArray()); } /** *

* Checks if the CharSequence contains only certain characters. *

* *

* A {@code null} CharSequence will return {@code false}. A {@code null} valid * character array will return {@code false}. An empty CharSequence (length()=0) * always returns {@code true}. *

* *
	 * StringUtils.containsOnly(null, *)       = false
	 * StringUtils.containsOnly(*, null)       = false
	 * StringUtils.containsOnly("", *)         = true
	 * StringUtils.containsOnly("ab", '')      = false
	 * StringUtils.containsOnly("abab", 'abc') = true
	 * StringUtils.containsOnly("ab1", 'abc')  = false
	 * StringUtils.containsOnly("abz", 'abc')  = false
	 * 
* * @param cs the String to check, may be null * @param valid an array of valid chars, may be null * @return true if it only contains valid chars and is non-null * @since 3.0 Changed signature from containsOnly(String, char[]) to * containsOnly(CharSequence, char...) */ public static boolean containsOnly(final CharSequence cs, final char... valid) { // All these pre-checks are to maintain API with an older version if (valid == null || cs == null) { return false; } if (cs.length() == 0) { return true; } if (valid.length == 0) { return false; } return indexOfAnyBut(cs, valid) == INDEX_NOT_FOUND; } /** *

* Checks if the CharSequence contains only certain characters. *

* *

* A {@code null} CharSequence will return {@code false}. A {@code null} valid * character String will return {@code false}. An empty String (length()=0) * always returns {@code true}. *

* *
	 * StringUtils.containsOnly(null, *)       = false
	 * StringUtils.containsOnly(*, null)       = false
	 * StringUtils.containsOnly("", *)         = true
	 * StringUtils.containsOnly("ab", "")      = false
	 * StringUtils.containsOnly("abab", "abc") = true
	 * StringUtils.containsOnly("ab1", "abc")  = false
	 * StringUtils.containsOnly("abz", "abc")  = false
	 * 
* * @param cs the CharSequence to check, may be null * @param validChars a String of valid chars, may be null * @return true if it only contains valid chars and is non-null * @since 2.0 * @since 3.0 Changed signature from containsOnly(String, String) to * containsOnly(CharSequence, String) */ public static boolean containsOnly(final CharSequence cs, final String validChars) { if (cs == null || validChars == null) { return false; } return containsOnly(cs, validChars.toCharArray()); } /** *

* Check whether the given CharSequence contains any whitespace characters. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* * @param seq the CharSequence to check (may be {@code null}) * @return {@code true} if the CharSequence is not empty and contains at least 1 * (breaking) whitespace character * @since 3.0 */ // From org.springframework.util.StringUtils, under Apache License 2.0 public static boolean containsWhitespace(final CharSequence seq) { if (isEmpty(seq)) { return false; } final int strLen = seq.length(); for (int i = 0; i < strLen; i++) { if (Character.isWhitespace(seq.charAt(i))) { return true; } } return false; } private static void convertRemainingAccentCharacters(final StringBuilder decomposed) { for (int i = 0; i < decomposed.length(); i++) { if (decomposed.charAt(i) == '\u0141') { decomposed.setCharAt(i, 'L'); } else if (decomposed.charAt(i) == '\u0142') { decomposed.setCharAt(i, 'l'); } } } /** *

* Counts how many times the char appears in the given string. *

* *

* A {@code null} or empty ("") String input returns {@code 0}. *

* *
	 * StringUtils.countMatches(null, *)       = 0
	 * StringUtils.countMatches("", *)         = 0
	 * StringUtils.countMatches("abba", 0)  = 0
	 * StringUtils.countMatches("abba", 'a')   = 2
	 * StringUtils.countMatches("abba", 'b')  = 2
	 * StringUtils.countMatches("abba", 'x') = 0
	 * 
* * @param str the CharSequence to check, may be null * @param ch the char to count * @return the number of occurrences, 0 if the CharSequence is {@code null} * @since 3.4 */ public static int countMatches(final CharSequence str, final char ch) { if (isEmpty(str)) { return 0; } int count = 0; // We could also call str.toCharArray() for faster look ups but that would // generate more garbage. for (int i = 0; i < str.length(); i++) { if (ch == str.charAt(i)) { count++; } } return count; } /** *

* Counts how many times the substring appears in the larger string. Note that * the code only counts non-overlapping matches. *

* *

* A {@code null} or empty ("") String input returns {@code 0}. *

* *
	 * StringUtils.countMatches(null, *)       = 0
	 * StringUtils.countMatches("", *)         = 0
	 * StringUtils.countMatches("abba", null)  = 0
	 * StringUtils.countMatches("abba", "")    = 0
	 * StringUtils.countMatches("abba", "a")   = 2
	 * StringUtils.countMatches("abba", "ab")  = 1
	 * StringUtils.countMatches("abba", "xxx") = 0
	 * StringUtils.countMatches("ababa", "aba") = 1
	 * 
* * @param str the CharSequence to check, may be null * @param sub the substring to count, may be null * @return the number of occurrences, 0 if either CharSequence is {@code null} * @since 3.0 Changed signature from countMatches(String, String) to * countMatches(CharSequence, CharSequence) */ public static int countMatches(final CharSequence str, final CharSequence sub) { if (isEmpty(str) || isEmpty(sub)) { return 0; } int count = 0; int idx = 0; while ((idx = CharSequenceUtils.indexOf(str, sub, idx)) != INDEX_NOT_FOUND) { count++; idx += sub.length(); } return count; } /** *

* Returns either the passed in CharSequence, or if the CharSequence is * whitespace, empty ("") or {@code null}, the value of {@code defaultStr}. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.defaultIfBlank(null, "NULL")  = "NULL"
	 * StringUtils.defaultIfBlank("", "NULL")    = "NULL"
	 * StringUtils.defaultIfBlank(" ", "NULL")   = "NULL"
	 * StringUtils.defaultIfBlank("bat", "NULL") = "bat"
	 * StringUtils.defaultIfBlank("", null)      = null
	 * 
* * @param the specific kind of CharSequence * @param str the CharSequence to check, may be null * @param defaultStr the default CharSequence to return if the input is * whitespace, empty ("") or {@code null}, may be null * @return the passed in CharSequence, or the default * @see StringUtils#defaultString(String, String) */ public static T defaultIfBlank(final T str, final T defaultStr) { return isBlank(str) ? defaultStr : str; } /** *

* Returns either the passed in CharSequence, or if the CharSequence is empty or * {@code null}, the value of {@code defaultStr}. *

* *
	 * StringUtils.defaultIfEmpty(null, "NULL")  = "NULL"
	 * StringUtils.defaultIfEmpty("", "NULL")    = "NULL"
	 * StringUtils.defaultIfEmpty(" ", "NULL")   = " "
	 * StringUtils.defaultIfEmpty("bat", "NULL") = "bat"
	 * StringUtils.defaultIfEmpty("", null)      = null
	 * 
* * @param the specific kind of CharSequence * @param str the CharSequence to check, may be null * @param defaultStr the default CharSequence to return if the input is empty * ("") or {@code null}, may be null * @return the passed in CharSequence, or the default * @see StringUtils#defaultString(String, String) */ public static T defaultIfEmpty(final T str, final T defaultStr) { return isEmpty(str) ? defaultStr : str; } /** *

* Returns either the passed in String, or if the String is {@code null}, an * empty String (""). *

* *
	 * StringUtils.defaultString(null)  = ""
	 * StringUtils.defaultString("")    = ""
	 * StringUtils.defaultString("bat") = "bat"
	 * 
* * @see ObjectUtils#toString(Object) * @see String#valueOf(Object) * @param str the String to check, may be null * @return the passed in String, or the empty String if it was {@code null} */ public static String defaultString(final String str) { return defaultString(str, EMPTY); } /** *

* Returns either the passed in String, or if the String is {@code null}, the * value of {@code defaultStr}. *

* *
	 * StringUtils.defaultString(null, "NULL")  = "NULL"
	 * StringUtils.defaultString("", "NULL")    = ""
	 * StringUtils.defaultString("bat", "NULL") = "bat"
	 * 
* * @see ObjectUtils#toString(Object,String) * @see String#valueOf(Object) * @param str the String to check, may be null * @param defaultStr the default String to return if the input is {@code null}, * may be null * @return the passed in String, or the default if it was {@code null} */ public static String defaultString(final String str, final String defaultStr) { return str == null ? defaultStr : str; } /** *

* Deletes all whitespaces from a String as defined by * {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.deleteWhitespace(null)         = null
	 * StringUtils.deleteWhitespace("")           = ""
	 * StringUtils.deleteWhitespace("abc")        = "abc"
	 * StringUtils.deleteWhitespace("   ab  c  ") = "abc"
	 * 
* * @param str the String to delete whitespace from, may be null * @return the String without whitespaces, {@code null} if null String input */ public static String deleteWhitespace(final String str) { if (isEmpty(str)) { return str; } final int sz = str.length(); final char[] chs = new char[sz]; int count = 0; for (int i = 0; i < sz; i++) { if (!Character.isWhitespace(str.charAt(i))) { chs[count++] = str.charAt(i); } } if (count == sz) { return str; } if (count == 0) { return EMPTY; } return new String(chs, 0, count); } /** *

* Compares two Strings, and returns the portion where they differ. More * precisely, return the remainder of the second String, starting from where * it's different from the first. This means that the difference between "abc" * and "ab" is the empty String and not "c". *

* *

* For example, {@code difference("i am a machine", "i am a robot") -> "robot"}. *

* *
	 * StringUtils.difference(null, null) = null
	 * StringUtils.difference("", "") = ""
	 * StringUtils.difference("", "abc") = "abc"
	 * StringUtils.difference("abc", "") = ""
	 * StringUtils.difference("abc", "abc") = ""
	 * StringUtils.difference("abc", "ab") = ""
	 * StringUtils.difference("ab", "abxyz") = "xyz"
	 * StringUtils.difference("abcde", "abxyz") = "xyz"
	 * StringUtils.difference("abcde", "xyz") = "xyz"
	 * 
* * @param str1 the first String, may be null * @param str2 the second String, may be null * @return the portion of str2 where it differs from str1; returns the empty * String if they are equal * @see #indexOfDifference(CharSequence,CharSequence) * @since 2.0 */ public static String difference(final String str1, final String str2) { if (str1 == null) { return str2; } if (str2 == null) { return str1; } final int at = indexOfDifference(str1, str2); if (at == INDEX_NOT_FOUND) { return EMPTY; } return str2.substring(at); } /** *

* Check if a CharSequence ends with a specified suffix. *

* *

* {@code null}s are handled without exceptions. Two {@code null} references are * considered to be equal. The comparison is case sensitive. *

* *
	 * StringUtils.endsWith(null, null)      = true
	 * StringUtils.endsWith(null, "def")     = false
	 * StringUtils.endsWith("abcdef", null)  = false
	 * StringUtils.endsWith("abcdef", "def") = true
	 * StringUtils.endsWith("ABCDEF", "def") = false
	 * StringUtils.endsWith("ABCDEF", "cde") = false
	 * StringUtils.endsWith("ABCDEF", "")    = true
	 * 
* * @see java.lang.String#endsWith(String) * @param str the CharSequence to check, may be null * @param suffix the suffix to find, may be null * @return {@code true} if the CharSequence ends with the suffix, case * sensitive, or both {@code null} * @since 2.4 * @since 3.0 Changed signature from endsWith(String, String) to * endsWith(CharSequence, CharSequence) */ public static boolean endsWith(final CharSequence str, final CharSequence suffix) { return endsWith(str, suffix, false); } /** *

* Check if a CharSequence ends with a specified suffix (optionally case * insensitive). *

* * @see java.lang.String#endsWith(String) * @param str the CharSequence to check, may be null * @param suffix the suffix to find, may be null * @param ignoreCase indicates whether the compare should ignore case (case * insensitive) or not. * @return {@code true} if the CharSequence starts with the prefix or both * {@code null} */ private static boolean endsWith(final CharSequence str, final CharSequence suffix, final boolean ignoreCase) { if (str == null || suffix == null) { return str == suffix; } if (suffix.length() > str.length()) { return false; } final int strOffset = str.length() - suffix.length(); return CharSequenceUtils.regionMatches(str, ignoreCase, strOffset, suffix, 0, suffix.length()); } /** *

* Check if a CharSequence ends with any of the provided case-sensitive * suffixes. *

* *
	 * StringUtils.endsWithAny(null, null)      = false
	 * StringUtils.endsWithAny(null, new String[] {"abc"})  = false
	 * StringUtils.endsWithAny("abcxyz", null)     = false
	 * StringUtils.endsWithAny("abcxyz", new String[] {""}) = true
	 * StringUtils.endsWithAny("abcxyz", new String[] {"xyz"}) = true
	 * StringUtils.endsWithAny("abcxyz", new String[] {null, "xyz", "abc"}) = true
	 * StringUtils.endsWithAny("abcXYZ", "def", "XYZ") = true
	 * StringUtils.endsWithAny("abcXYZ", "def", "xyz") = false
	 * 
* * @param sequence the CharSequence to check, may be null * @param searchStrings the case-sensitive CharSequences to find, may be empty * or contain {@code null} * @see StringUtils#endsWith(CharSequence, CharSequence) * @return {@code true} if the input {@code sequence} is {@code null} AND no * {@code searchStrings} are provided, or the input {@code sequence} * ends in any of the provided case-sensitive {@code searchStrings}. * @since 3.0 */ public static boolean endsWithAny(final CharSequence sequence, final CharSequence... searchStrings) { if (isEmpty(sequence) || searchStrings.length == 0) { return false; } for (final CharSequence searchString : searchStrings) { if (endsWith(sequence, searchString)) { return true; } } return false; } /** *

* Case insensitive check if a CharSequence ends with a specified suffix. *

* *

* {@code null}s are handled without exceptions. Two {@code null} references are * considered to be equal. The comparison is case insensitive. *

* *
	 * StringUtils.endsWithIgnoreCase(null, null)      = true
	 * StringUtils.endsWithIgnoreCase(null, "def")     = false
	 * StringUtils.endsWithIgnoreCase("abcdef", null)  = false
	 * StringUtils.endsWithIgnoreCase("abcdef", "def") = true
	 * StringUtils.endsWithIgnoreCase("ABCDEF", "def") = true
	 * StringUtils.endsWithIgnoreCase("ABCDEF", "cde") = false
	 * 
* * @see java.lang.String#endsWith(String) * @param str the CharSequence to check, may be null * @param suffix the suffix to find, may be null * @return {@code true} if the CharSequence ends with the suffix, case * insensitive, or both {@code null} * @since 2.4 * @since 3.0 Changed signature from endsWithIgnoreCase(String, String) to * endsWithIgnoreCase(CharSequence, CharSequence) */ public static boolean endsWithIgnoreCase(final CharSequence str, final CharSequence suffix) { return endsWith(str, suffix, true); } /** *

* Compares two CharSequences, returning {@code true} if they represent equal * sequences of characters. *

* *

* {@code null}s are handled without exceptions. Two {@code null} references are * considered to be equal. The comparison is case sensitive. *

* *
	 * StringUtils.equals(null, null)   = true
	 * StringUtils.equals(null, "abc")  = false
	 * StringUtils.equals("abc", null)  = false
	 * StringUtils.equals("abc", "abc") = true
	 * StringUtils.equals("abc", "ABC") = false
	 * 
* * @param cs1 the first CharSequence, may be {@code null} * @param cs2 the second CharSequence, may be {@code null} * @return {@code true} if the CharSequences are equal (case-sensitive), or both * {@code null} * @since 3.0 Changed signature from equals(String, String) to * equals(CharSequence, CharSequence) * @see Object#equals(Object) * @see #equalsIgnoreCase(CharSequence, CharSequence) */ public static boolean equals(final CharSequence cs1, final CharSequence cs2) { if (cs1 == cs2) { return true; } if (cs1 == null || cs2 == null) { return false; } if (cs1.length() != cs2.length()) { return false; } if (cs1 instanceof String && cs2 instanceof String) { return cs1.equals(cs2); } // Step-wise comparison final int length = cs1.length(); for (int i = 0; i < length; i++) { if (cs1.charAt(i) != cs2.charAt(i)) { return false; } } return true; } /** *

* Compares given {@code string} to a CharSequences vararg of * {@code searchStrings}, returning {@code true} if the {@code string} is equal * to any of the {@code searchStrings}. *

* *
	 * StringUtils.equalsAny(null, (CharSequence[]) null) = false
	 * StringUtils.equalsAny(null, null, null)    = true
	 * StringUtils.equalsAny(null, "abc", "def")  = false
	 * StringUtils.equalsAny("abc", null, "def")  = false
	 * StringUtils.equalsAny("abc", "abc", "def") = true
	 * StringUtils.equalsAny("abc", "ABC", "DEF") = false
	 * 
* * @param string to compare, may be {@code null}. * @param searchStrings a vararg of strings, may be {@code null}. * @return {@code true} if the string is equal (case-sensitive) to any other * element of {@code searchStrings}; {@code false} if * {@code searchStrings} is null or contains no matches. * @since 3.5 */ public static boolean equalsAny(final CharSequence string, final CharSequence... searchStrings) { if (searchStrings.length > 0) { for (final CharSequence next : searchStrings) { if (equals(string, next)) { return true; } } } return false; } /** *

* Compares given {@code string} to a CharSequences vararg of * {@code searchStrings}, returning {@code true} if the {@code string} is equal * to any of the {@code searchStrings}, ignoring case. *

* *
	 * StringUtils.equalsAnyIgnoreCase(null, (CharSequence[]) null) = false
	 * StringUtils.equalsAnyIgnoreCase(null, null, null)    = true
	 * StringUtils.equalsAnyIgnoreCase(null, "abc", "def")  = false
	 * StringUtils.equalsAnyIgnoreCase("abc", null, "def")  = false
	 * StringUtils.equalsAnyIgnoreCase("abc", "abc", "def") = true
	 * StringUtils.equalsAnyIgnoreCase("abc", "ABC", "DEF") = true
	 * 
* * @param string to compare, may be {@code null}. * @param searchStrings a vararg of strings, may be {@code null}. * @return {@code true} if the string is equal (case-insensitive) to any other * element of {@code searchStrings}; {@code false} if * {@code searchStrings} is null or contains no matches. * @since 3.5 */ public static boolean equalsAnyIgnoreCase(final CharSequence string, final CharSequence... searchStrings) { if (searchStrings.length > 0) { for (final CharSequence next : searchStrings) { if (equalsIgnoreCase(string, next)) { return true; } } } return false; } /** *

* Compares two CharSequences, returning {@code true} if they represent equal * sequences of characters, ignoring case. *

* *

* {@code null}s are handled without exceptions. Two {@code null} references are * considered equal. The comparison is case insensitive. *

* *
	 * StringUtils.equalsIgnoreCase(null, null)   = true
	 * StringUtils.equalsIgnoreCase(null, "abc")  = false
	 * StringUtils.equalsIgnoreCase("abc", null)  = false
	 * StringUtils.equalsIgnoreCase("abc", "abc") = true
	 * StringUtils.equalsIgnoreCase("abc", "ABC") = true
	 * 
* * @param cs1 the first CharSequence, may be {@code null} * @param cs2 the second CharSequence, may be {@code null} * @return {@code true} if the CharSequences are equal (case-insensitive), or * both {@code null} * @since 3.0 Changed signature from equalsIgnoreCase(String, String) to * equalsIgnoreCase(CharSequence, CharSequence) * @see #equals(CharSequence, CharSequence) */ public static boolean equalsIgnoreCase(final CharSequence cs1, final CharSequence cs2) { if (cs1 == cs2) { return true; } if (cs1 == null || cs2 == null) { return false; } if (cs1.length() != cs2.length()) { return false; } return CharSequenceUtils.regionMatches(cs1, true, 0, cs2, 0, cs1.length()); } /** *

* Returns the first value in the array which is not empty (""), {@code null} or * whitespace only. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *

* If all values are blank or the array is {@code null} or empty then * {@code null} is returned. *

* *
	 * StringUtils.firstNonBlank(null, null, null)     = null
	 * StringUtils.firstNonBlank(null, "", " ")        = null
	 * StringUtils.firstNonBlank("abc")                = "abc"
	 * StringUtils.firstNonBlank(null, "xyz")          = "xyz"
	 * StringUtils.firstNonBlank(null, "", " ", "xyz") = "xyz"
	 * StringUtils.firstNonBlank(null, "xyz", "abc")   = "xyz"
	 * StringUtils.firstNonBlank()                     = null
	 * 
* * @param the specific kind of CharSequence * @param values the values to test, may be {@code null} or empty * @return the first value from {@code values} which is not blank, or * {@code null} if there are no non-blank values * @since 3.8 */ @SafeVarargs public static T firstNonBlank(final T... values) { if (values != null) { for (final T val : values) { if (isNotBlank(val)) { return val; } } } return null; } /** *

* Returns the first value in the array which is not empty. *

* *

* If all values are empty or the array is {@code null} or empty then * {@code null} is returned. *

* *
	 * StringUtils.firstNonEmpty(null, null, null)   = null
	 * StringUtils.firstNonEmpty(null, null, "")     = null
	 * StringUtils.firstNonEmpty(null, "", " ")      = " "
	 * StringUtils.firstNonEmpty("abc")              = "abc"
	 * StringUtils.firstNonEmpty(null, "xyz")        = "xyz"
	 * StringUtils.firstNonEmpty("", "xyz")          = "xyz"
	 * StringUtils.firstNonEmpty(null, "xyz", "abc") = "xyz"
	 * StringUtils.firstNonEmpty()                   = null
	 * 
* * @param the specific kind of CharSequence * @param values the values to test, may be {@code null} or empty * @return the first value from {@code values} which is not empty, or * {@code null} if there are no non-empty values * @since 3.8 */ @SafeVarargs public static T firstNonEmpty(final T... values) { if (values != null) { for (final T val : values) { if (isNotEmpty(val)) { return val; } } } return null; } /** * Calls {@link String#getBytes(Charset)} in a null-safe manner. * * @param string input string * @param charset The {@link Charset} to encode the {@code String}. If null, * then use the default Charset. * @return The empty byte[] if {@code string} is null, the result of * {@link String#getBytes(Charset)} otherwise. * @see String#getBytes(Charset) * @since 3.10 */ public static byte[] getBytes(final String string, final Charset charset) { return string == null ? new byte[0] : string.getBytes(Charsets.toCharset(charset)); } /** * Calls {@link String#getBytes(String)} in a null-safe manner. * * @param string input string * @param charset The {@link Charset} name to encode the {@code String}. If * null, then use the default Charset. * @return The empty byte[] if {@code string} is null, the result of * {@link String#getBytes(String)} otherwise. * @throws UnsupportedEncodingException Thrown when the named charset is not * supported. * @see String#getBytes(String) * @since 3.10 */ public static byte[] getBytes(final String string, final String charset) throws UnsupportedEncodingException { return string == null ? new byte[0] : string.getBytes(Charsets.toCharsetName(charset)); } /** *

* Compares all Strings in an array and returns the initial sequence of * characters that is common to all of them. *

* *

* For example, {@code getCommonPrefix(new String[] {"i am a machine", "i am a * robot"}) -> "i am a "} *

* *
	 * StringUtils.getCommonPrefix(null) = ""
	 * StringUtils.getCommonPrefix(new String[] {}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"abc"}) = "abc"
	 * StringUtils.getCommonPrefix(new String[] {null, null}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"", ""}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"", null}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"abc", null, null}) = ""
	 * StringUtils.getCommonPrefix(new String[] {null, null, "abc"}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"", "abc"}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"abc", ""}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"abc", "abc"}) = "abc"
	 * StringUtils.getCommonPrefix(new String[] {"abc", "a"}) = "a"
	 * StringUtils.getCommonPrefix(new String[] {"ab", "abxyz"}) = "ab"
	 * StringUtils.getCommonPrefix(new String[] {"abcde", "abxyz"}) = "ab"
	 * StringUtils.getCommonPrefix(new String[] {"abcde", "xyz"}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"xyz", "abcde"}) = ""
	 * StringUtils.getCommonPrefix(new String[] {"i am a machine", "i am a robot"}) = "i am a "
	 * 
* * @param strs array of String objects, entries may be null * @return the initial sequence of characters that are common to all Strings in * the array; empty String if the array is null, the elements are all * null or if there is no common prefix. * @since 2.4 */ public static String getCommonPrefix(final String... strs) { if (strs.length == 0) { return EMPTY; } final int smallestIndexOfDiff = indexOfDifference(strs); if (smallestIndexOfDiff == INDEX_NOT_FOUND) { // all strings were identical if (strs[0] == null) { return EMPTY; } return strs[0]; } else if (smallestIndexOfDiff == 0) { // there were no common initial characters return EMPTY; } else { // we found a common initial character sequence return strs[0].substring(0, smallestIndexOfDiff); } } /** *

* Checks if a String {@code str} contains Unicode digits, if yes then * concatenate all the digits in {@code str} and return it as a String. *

* *

* An empty ("") String will be returned if no digits found in {@code str}. *

* *
	 * StringUtils.getDigits(null)  = null
	 * StringUtils.getDigits("")    = ""
	 * StringUtils.getDigits("abc") = ""
	 * StringUtils.getDigits("1000$") = "1000"
	 * StringUtils.getDigits("1123~45") = "112345"
	 * StringUtils.getDigits("(541) 754-3010") = "5417543010"
	 * StringUtils.getDigits("\u0967\u0968\u0969") = "\u0967\u0968\u0969"
	 * 
* * @param str the String to extract digits from, may be null * @return String with only digits, or an empty ("") String if no digits found, * or {@code null} String if {@code str} is null * @since 3.6 */ public static String getDigits(final String str) { if (isEmpty(str)) { return str; } final int sz = str.length(); final StringBuilder strDigits = new StringBuilder(sz); for (int i = 0; i < sz; i++) { final char tempChar = str.charAt(i); if (Character.isDigit(tempChar)) { strDigits.append(tempChar); } } return strDigits.toString(); } /** *

* Find the Fuzzy Distance which indicates the similarity score between two * Strings. *

* *

* This string matching algorithm is similar to the algorithms of editors such * as Sublime Text, TextMate, Atom and others. One point is given for every * matched character. Subsequent matches yield two bonus points. A higher score * indicates a higher similarity. *

* *
	 * StringUtils.getFuzzyDistance(null, null, null)                                    = IllegalArgumentException
	 * StringUtils.getFuzzyDistance("", "", Locale.ENGLISH)                              = 0
	 * StringUtils.getFuzzyDistance("Workshop", "b", Locale.ENGLISH)                     = 0
	 * StringUtils.getFuzzyDistance("Room", "o", Locale.ENGLISH)                         = 1
	 * StringUtils.getFuzzyDistance("Workshop", "w", Locale.ENGLISH)                     = 1
	 * StringUtils.getFuzzyDistance("Workshop", "ws", Locale.ENGLISH)                    = 2
	 * StringUtils.getFuzzyDistance("Workshop", "wo", Locale.ENGLISH)                    = 4
	 * StringUtils.getFuzzyDistance("Apache Software Foundation", "asf", Locale.ENGLISH) = 3
	 * 
* * @param term a full term that should be matched against, must not be null * @param query the query that will be matched against a term, must not be null * @param locale This string matching logic is case insensitive. A locale is * necessary to normalize both Strings to lower case. * @return result score * @throws IllegalArgumentException if either String input {@code null} or * Locale input {@code null} * @since 3.4 * @deprecated as of 3.6, use commons-text * FuzzyScore instead */ @Deprecated public static int getFuzzyDistance(final CharSequence term, final CharSequence query, final Locale locale) { if (term == null || query == null) { throw new IllegalArgumentException("Strings must not be null"); } else if (locale == null) { throw new IllegalArgumentException("Locale must not be null"); } // fuzzy logic is case insensitive. We normalize the Strings to lower // case right from the start. Turning characters to lower case // via Character.toLowerCase(char) is unfortunately insufficient // as it does not accept a locale. final String termLowerCase = term.toString().toLowerCase(locale); final String queryLowerCase = query.toString().toLowerCase(locale); // the resulting score int score = 0; // the position in the term which will be scanned next for potential // query character matches int termIndex = 0; // index of the previously matched character in the term int previousMatchingCharacterIndex = Integer.MIN_VALUE; for (int queryIndex = 0; queryIndex < queryLowerCase.length(); queryIndex++) { final char queryChar = queryLowerCase.charAt(queryIndex); boolean termCharacterMatchFound = false; for (; termIndex < termLowerCase.length() && !termCharacterMatchFound; termIndex++) { final char termChar = termLowerCase.charAt(termIndex); if (queryChar == termChar) { // simple character matches result in one point score++; // subsequent character matches further improve // the score. if (previousMatchingCharacterIndex + 1 == termIndex) { score += 2; } previousMatchingCharacterIndex = termIndex; // we can leave the nested loop. Every character in the // query can match at most one character in the term. termCharacterMatchFound = true; } } } return score; } /** *

* Returns either the passed in CharSequence, or if the CharSequence is * whitespace, empty ("") or {@code null}, the value supplied by * {@code defaultStrSupplier}. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *

* Caller responsible for thread-safety and exception handling of default value * supplier *

* *
	 * {@code
	 * StringUtils.getIfBlank(null, () -> "NULL")   = "NULL"
	 * StringUtils.getIfBlank("", () -> "NULL")     = "NULL"
	 * StringUtils.getIfBlank(" ", () -> "NULL")    = "NULL"
	 * StringUtils.getIfBlank("bat", () -> "NULL")  = "bat"
	 * StringUtils.getIfBlank("", () -> null)       = null
	 * StringUtils.getIfBlank("", null)             = null
	 * }
	 * 
* * @param the specific kind of CharSequence * @param str the CharSequence to check, may be null * @param defaultSupplier the supplier of default CharSequence to return if the * input is whitespace, empty ("") or {@code null}, may * be null * @return the passed in CharSequence, or the default * @see StringUtils#defaultString(String, String) * @since 3.10 */ public static T getIfBlank(final T str, final Supplier defaultSupplier) { return isBlank(str) ? defaultSupplier == null ? null : defaultSupplier.get() : str; } /** *

* Returns either the passed in CharSequence, or if the CharSequence is empty or * {@code null}, the value supplied by {@code defaultStrSupplier}. *

* *

* Caller responsible for thread-safety and exception handling of default value * supplier *

* *
	 * {@code
	 * StringUtils.getIfEmpty(null, () -> "NULL")    = "NULL"
	 * StringUtils.getIfEmpty("", () -> "NULL")      = "NULL"
	 * StringUtils.getIfEmpty(" ", () -> "NULL")     = " "
	 * StringUtils.getIfEmpty("bat", () -> "NULL")   = "bat"
	 * StringUtils.getIfEmpty("", () -> null)        = null
	 * StringUtils.getIfEmpty("", null)              = null
	 * }
	 * 
* * @param the specific kind of CharSequence * @param str the CharSequence to check, may be null * @param defaultSupplier the supplier of default CharSequence to return if the * input is empty ("") or {@code null}, may be null * @return the passed in CharSequence, or the default * @see StringUtils#defaultString(String, String) * @since 3.10 */ public static T getIfEmpty(final T str, final Supplier defaultSupplier) { return isEmpty(str) ? defaultSupplier == null ? null : defaultSupplier.get() : str; } /** *

* Find the Jaro Winkler Distance which indicates the similarity score between * two Strings. *

* *

* The Jaro measure is the weighted sum of percentage of matched characters from * each file and transposed characters. Winkler increased this measure for * matching initial characters. *

* *

* This implementation is based on the Jaro Winkler similarity algorithm from * http://en.wikipedia.org/wiki/Jaro%E2%80%93Winkler_distance. *

* *
	 * StringUtils.getJaroWinklerDistance(null, null)          = IllegalArgumentException
	 * StringUtils.getJaroWinklerDistance("", "")              = 0.0
	 * StringUtils.getJaroWinklerDistance("", "a")             = 0.0
	 * StringUtils.getJaroWinklerDistance("aaapppp", "")       = 0.0
	 * StringUtils.getJaroWinklerDistance("frog", "fog")       = 0.93
	 * StringUtils.getJaroWinklerDistance("fly", "ant")        = 0.0
	 * StringUtils.getJaroWinklerDistance("elephant", "hippo") = 0.44
	 * StringUtils.getJaroWinklerDistance("hippo", "elephant") = 0.44
	 * StringUtils.getJaroWinklerDistance("hippo", "zzzzzzzz") = 0.0
	 * StringUtils.getJaroWinklerDistance("hello", "hallo")    = 0.88
	 * StringUtils.getJaroWinklerDistance("ABC Corporation", "ABC Corp") = 0.93
	 * StringUtils.getJaroWinklerDistance("D N H Enterprises Inc", "D & H Enterprises, Inc.") = 0.95
	 * StringUtils.getJaroWinklerDistance("My Gym Children's Fitness Center", "My Gym. Childrens Fitness") = 0.92
	 * StringUtils.getJaroWinklerDistance("PENNSYLVANIA", "PENNCISYLVNIA") = 0.88
	 * 
* * @param first the first String, must not be null * @param second the second String, must not be null * @return result distance * @throws IllegalArgumentException if either String input {@code null} * @since 3.3 * @deprecated as of 3.6, use commons-text * JaroWinklerDistance instead */ @Deprecated public static double getJaroWinklerDistance(final CharSequence first, final CharSequence second) { final double DEFAULT_SCALING_FACTOR = 0.1; if (first == null || second == null) { throw new IllegalArgumentException("Strings must not be null"); } final int[] mtp = matches(first, second); final double m = mtp[0]; if (m == 0) { return 0D; } final double j = (m / first.length() + m / second.length() + (m - mtp[1]) / m) / 3; final double jw = j < 0.7D ? j : j + Math.min(DEFAULT_SCALING_FACTOR, 1D / mtp[3]) * mtp[2] * (1D - j); return Math.round(jw * 100.0D) / 100.0D; } /** *

* Find the Levenshtein distance between two Strings. *

* *

* This is the number of changes needed to change one String into another, where * each change is a single character modification (deletion, insertion or * substitution). *

* *

* The implementation uses a single-dimensional array of length s.length() + 1. * See * http://blog.softwx.net/2014/12/optimizing-levenshtein-algorithm-in-c.html * for details. *

* *
	 * StringUtils.getLevenshteinDistance(null, *)             = IllegalArgumentException
	 * StringUtils.getLevenshteinDistance(*, null)             = IllegalArgumentException
	 * StringUtils.getLevenshteinDistance("", "")              = 0
	 * StringUtils.getLevenshteinDistance("", "a")             = 1
	 * StringUtils.getLevenshteinDistance("aaapppp", "")       = 7
	 * StringUtils.getLevenshteinDistance("frog", "fog")       = 1
	 * StringUtils.getLevenshteinDistance("fly", "ant")        = 3
	 * StringUtils.getLevenshteinDistance("elephant", "hippo") = 7
	 * StringUtils.getLevenshteinDistance("hippo", "elephant") = 7
	 * StringUtils.getLevenshteinDistance("hippo", "zzzzzzzz") = 8
	 * StringUtils.getLevenshteinDistance("hello", "hallo")    = 1
	 * 
* * @param s the first String, must not be null * @param t the second String, must not be null * @return result distance * @throws IllegalArgumentException if either String input {@code null} * @since 3.0 Changed signature from getLevenshteinDistance(String, String) to * getLevenshteinDistance(CharSequence, CharSequence) * @deprecated as of 3.6, use commons-text * LevenshteinDistance instead */ @Deprecated public static int getLevenshteinDistance(CharSequence s, CharSequence t) { if (s == null || t == null) { throw new IllegalArgumentException("Strings must not be null"); } int n = s.length(); int m = t.length(); if (n == 0) { return m; } else if (m == 0) { return n; } if (n > m) { // swap the input strings to consume less memory final CharSequence tmp = s; s = t; t = tmp; n = m; m = t.length(); } final int[] p = new int[n + 1]; // indexes into strings s and t int i; // iterates through s int j; // iterates through t int upper_left; int upper; char t_j; // jth character of t int cost; for (i = 0; i <= n; i++) { p[i] = i; } for (j = 1; j <= m; j++) { upper_left = p[0]; t_j = t.charAt(j - 1); p[0] = j; for (i = 1; i <= n; i++) { upper = p[i]; cost = s.charAt(i - 1) == t_j ? 0 : 1; // minimum of cell to the left+1, to the top+1, diagonally left and up +cost p[i] = Math.min(Math.min(p[i - 1] + 1, p[i] + 1), upper_left + cost); upper_left = upper; } } return p[n]; } /** *

* Find the Levenshtein distance between two Strings if it's less than or equal * to a given threshold. *

* *

* This is the number of changes needed to change one String into another, where * each change is a single character modification (deletion, insertion or * substitution). *

* *

* This implementation follows from Algorithms on Strings, Trees and Sequences * by Dan Gusfield and Chas Emerick's implementation of the Levenshtein distance * algorithm from http://www.merriampark.com/ld.htm *

* *
	 * StringUtils.getLevenshteinDistance(null, *, *)             = IllegalArgumentException
	 * StringUtils.getLevenshteinDistance(*, null, *)             = IllegalArgumentException
	 * StringUtils.getLevenshteinDistance(*, *, -1)               = IllegalArgumentException
	 * StringUtils.getLevenshteinDistance("", "", 0)              = 0
	 * StringUtils.getLevenshteinDistance("aaapppp", "", 8)       = 7
	 * StringUtils.getLevenshteinDistance("aaapppp", "", 7)       = 7
	 * StringUtils.getLevenshteinDistance("aaapppp", "", 6))      = -1
	 * StringUtils.getLevenshteinDistance("elephant", "hippo", 7) = 7
	 * StringUtils.getLevenshteinDistance("elephant", "hippo", 6) = -1
	 * StringUtils.getLevenshteinDistance("hippo", "elephant", 7) = 7
	 * StringUtils.getLevenshteinDistance("hippo", "elephant", 6) = -1
	 * 
* * @param s the first String, must not be null * @param t the second String, must not be null * @param threshold the target threshold, must not be negative * @return result distance, or {@code -1} if the distance would be greater than * the threshold * @throws IllegalArgumentException if either String input {@code null} or * negative threshold * @deprecated as of 3.6, use commons-text * LevenshteinDistance instead */ @Deprecated public static int getLevenshteinDistance(CharSequence s, CharSequence t, final int threshold) { if (s == null || t == null) { throw new IllegalArgumentException("Strings must not be null"); } if (threshold < 0) { throw new IllegalArgumentException("Threshold must not be negative"); } /* * This implementation only computes the distance if it's less than or equal to * the threshold value, returning -1 if it's greater. The advantage is * performance: unbounded distance is O(nm), but a bound of k allows us to * reduce it to O(km) time by only computing a diagonal stripe of width 2k + 1 * of the cost table. It is also possible to use this to compute the unbounded * Levenshtein distance by starting the threshold at 1 and doubling each time * until the distance is found; this is O(dm), where d is the distance. * * One subtlety comes from needing to ignore entries on the border of our stripe * eg. p[] = |#|#|#|* d[] = *|#|#|#| We must ignore the entry to the left of the * leftmost member We must ignore the entry above the rightmost member * * Another subtlety comes from our stripe running off the matrix if the strings * aren't of the same size. Since string s is always swapped to be the shorter * of the two, the stripe will always run off to the upper right instead of the * lower left of the matrix. * * As a concrete example, suppose s is of length 5, t is of length 7, and our * threshold is 1. In this case we're going to walk a stripe of length 3. The * matrix would look like so: * * 1 2 3 4 5 1 |#|#| | | | 2 |#|#|#| | | 3 | |#|#|#| | 4 | | |#|#|#| 5 | | | * |#|#| 6 | | | | |#| 7 | | | | | | * * Note how the stripe leads off the table as there is no possible way to turn a * string of length 5 into one of length 7 in edit distance of 1. * * Additionally, this implementation decreases memory usage by using two * single-dimensional arrays and swapping them back and forth instead of * allocating an entire n by m matrix. This requires a few minor changes, such * as immediately returning when it's detected that the stripe has run off the * matrix and initially filling the arrays with large values so that entries we * don't compute are ignored. * * See Algorithms on Strings, Trees and Sequences by Dan Gusfield for some * discussion. */ int n = s.length(); // length of s int m = t.length(); // length of t // if one string is empty, the edit distance is necessarily the length of the // other if (n == 0) { return m <= threshold ? m : -1; } else if (m == 0) { return n <= threshold ? n : -1; } else if (Math.abs(n - m) > threshold) { // no need to calculate the distance if the length difference is greater than // the threshold return -1; } if (n > m) { // swap the two strings to consume less memory final CharSequence tmp = s; s = t; t = tmp; n = m; m = t.length(); } int[] p = new int[n + 1]; // 'previous' cost array, horizontally int[] d = new int[n + 1]; // cost array, horizontally int[] _d; // placeholder to assist in swapping p and d // fill in starting table values final int boundary = Math.min(n, threshold) + 1; for (int i = 0; i < boundary; i++) { p[i] = i; } // these fills ensure that the value above the rightmost entry of our // stripe will be ignored in following loop iterations Arrays.fill(p, boundary, p.length, Integer.MAX_VALUE); Arrays.fill(d, Integer.MAX_VALUE); // iterates through t for (int j = 1; j <= m; j++) { final char t_j = t.charAt(j - 1); // jth character of t d[0] = j; // compute stripe indices, constrain to array size final int min = Math.max(1, j - threshold); final int max = j > Integer.MAX_VALUE - threshold ? n : Math.min(n, j + threshold); // the stripe may lead off of the table if s and t are of different sizes if (min > max) { return -1; } // ignore entry left of leftmost if (min > 1) { d[min - 1] = Integer.MAX_VALUE; } // iterates through [min, max] in s for (int i = min; i <= max; i++) { if (s.charAt(i - 1) == t_j) { // diagonally left and up d[i] = p[i - 1]; } else { // 1 + minimum of cell to the left, to the top, diagonally left and up d[i] = 1 + Math.min(Math.min(d[i - 1], p[i]), p[i - 1]); } } // copy current distance counts to 'previous row' distance counts _d = p; p = d; d = _d; } // if p[n] is greater than the threshold, there's no guarantee on it being the // correct // distance if (p[n] <= threshold) { return p[n]; } return -1; } /** *

* Finds the first index within a CharSequence, handling {@code null}. This * method uses {@link String#indexOf(String, int)} if possible. *

* *

* A {@code null} CharSequence will return {@code -1}. *

* *
	 * StringUtils.indexOf(null, *)          = -1
	 * StringUtils.indexOf(*, null)          = -1
	 * StringUtils.indexOf("", "")           = 0
	 * StringUtils.indexOf("", *)            = -1 (except when * = "")
	 * StringUtils.indexOf("aabaabaa", "a")  = 0
	 * StringUtils.indexOf("aabaabaa", "b")  = 2
	 * StringUtils.indexOf("aabaabaa", "ab") = 1
	 * StringUtils.indexOf("aabaabaa", "")   = 0
	 * 
* * @param seq the CharSequence to check, may be null * @param searchSeq the CharSequence to find, may be null * @return the first index of the search CharSequence, -1 if no match or * {@code null} string input * @since 2.0 * @since 3.0 Changed signature from indexOf(String, String) to * indexOf(CharSequence, CharSequence) */ public static int indexOf(final CharSequence seq, final CharSequence searchSeq) { if (seq == null || searchSeq == null) { return INDEX_NOT_FOUND; } return CharSequenceUtils.indexOf(seq, searchSeq, 0); } /** *

* Finds the first index within a CharSequence, handling {@code null}. This * method uses {@link String#indexOf(String, int)} if possible. *

* *

* A {@code null} CharSequence will return {@code -1}. A negative start position * is treated as zero. An empty ("") search CharSequence always matches. A start * position greater than the string length only matches an empty search * CharSequence. *

* *
	 * StringUtils.indexOf(null, *, *)          = -1
	 * StringUtils.indexOf(*, null, *)          = -1
	 * StringUtils.indexOf("", "", 0)           = 0
	 * StringUtils.indexOf("", *, 0)            = -1 (except when * = "")
	 * StringUtils.indexOf("aabaabaa", "a", 0)  = 0
	 * StringUtils.indexOf("aabaabaa", "b", 0)  = 2
	 * StringUtils.indexOf("aabaabaa", "ab", 0) = 1
	 * StringUtils.indexOf("aabaabaa", "b", 3)  = 5
	 * StringUtils.indexOf("aabaabaa", "b", 9)  = -1
	 * StringUtils.indexOf("aabaabaa", "b", -1) = 2
	 * StringUtils.indexOf("aabaabaa", "", 2)   = 2
	 * StringUtils.indexOf("abc", "", 9)        = 3
	 * 
* * @param seq the CharSequence to check, may be null * @param searchSeq the CharSequence to find, may be null * @param startPos the start position, negative treated as zero * @return the first index of the search CharSequence (always ≥ startPos), -1 * if no match or {@code null} string input * @since 2.0 * @since 3.0 Changed signature from indexOf(String, String, int) to * indexOf(CharSequence, CharSequence, int) */ public static int indexOf(final CharSequence seq, final CharSequence searchSeq, final int startPos) { if (seq == null || searchSeq == null) { return INDEX_NOT_FOUND; } return CharSequenceUtils.indexOf(seq, searchSeq, startPos); } /** * Returns the index within {@code seq} of the first occurrence of the specified * character. If a character with value {@code searchChar} occurs in the * character sequence represented by {@code seq} {@code CharSequence} object, * then the index (in Unicode code units) of the first such occurrence is * returned. For values of {@code searchChar} in the range from 0 to 0xFFFF * (inclusive), this is the smallest value k such that:
* *
	 * this.charAt(k) == searchChar
	 * 
* *
is true. For other values of {@code searchChar}, it is the * smallest value k such that:
* *
	 * this.codePointAt(k) == searchChar
	 * 
* *
is true. In either case, if no such character occurs in * {@code seq}, then {@code INDEX_NOT_FOUND (-1)} is returned. * *

* Furthermore, a {@code null} or empty ("") CharSequence will return * {@code INDEX_NOT_FOUND (-1)}. *

* *
	 * StringUtils.indexOf(null, *)         = -1
	 * StringUtils.indexOf("", *)           = -1
	 * StringUtils.indexOf("aabaabaa", 'a') = 0
	 * StringUtils.indexOf("aabaabaa", 'b') = 2
	 * 
* * @param seq the CharSequence to check, may be null * @param searchChar the character to find * @return the first index of the search character, -1 if no match or * {@code null} string input * @since 2.0 * @since 3.0 Changed signature from indexOf(String, int) to * indexOf(CharSequence, int) * @since 3.6 Updated {@link CharSequenceUtils} call to behave more like * {@code String} */ public static int indexOf(final CharSequence seq, final int searchChar) { if (isEmpty(seq)) { return INDEX_NOT_FOUND; } return CharSequenceUtils.indexOf(seq, searchChar, 0); } /** * * Returns the index within {@code seq} of the first occurrence of the specified * character, starting the search at the specified index. *

* If a character with value {@code searchChar} occurs in the character sequence * represented by the {@code seq} {@code CharSequence} object at an index no * smaller than {@code startPos}, then the index of the first such occurrence is * returned. For values of {@code searchChar} in the range from 0 to 0xFFFF * (inclusive), this is the smallest value k such that:

* *
	 * (this.charAt(k) == searchChar) && (k >= startPos)
	 * 
* *
is true. For other values of {@code searchChar}, it is the * smallest value k such that:
* *
	 * (this.codePointAt(k) == searchChar) && (k >= startPos)
	 * 
* *
is true. In either case, if no such character occurs in * {@code seq} at or after position {@code startPos}, then {@code -1} is * returned. * *

* There is no restriction on the value of {@code startPos}. If it is negative, * it has the same effect as if it were zero: this entire string may be * searched. If it is greater than the length of this string, it has the same * effect as if it were equal to the length of this string: * {@code (INDEX_NOT_FOUND) -1} is returned. Furthermore, a {@code null} or * empty ("") CharSequence will return {@code (INDEX_NOT_FOUND) -1}. * *

* All indices are specified in {@code char} values (Unicode code units). * *

	 * StringUtils.indexOf(null, *, *)          = -1
	 * StringUtils.indexOf("", *, *)            = -1
	 * StringUtils.indexOf("aabaabaa", 'b', 0)  = 2
	 * StringUtils.indexOf("aabaabaa", 'b', 3)  = 5
	 * StringUtils.indexOf("aabaabaa", 'b', 9)  = -1
	 * StringUtils.indexOf("aabaabaa", 'b', -1) = 2
	 * 
* * @param seq the CharSequence to check, may be null * @param searchChar the character to find * @param startPos the start position, negative treated as zero * @return the first index of the search character (always ≥ startPos), -1 if * no match or {@code null} string input * @since 2.0 * @since 3.0 Changed signature from indexOf(String, int, int) to * indexOf(CharSequence, int, int) * @since 3.6 Updated {@link CharSequenceUtils} call to behave more like * {@code String} */ public static int indexOf(final CharSequence seq, final int searchChar, final int startPos) { if (isEmpty(seq)) { return INDEX_NOT_FOUND; } return CharSequenceUtils.indexOf(seq, searchChar, startPos); } /** *

* Search a CharSequence to find the first index of any character in the given * set of characters. *

* *

* A {@code null} String will return {@code -1}. A {@code null} or zero length * search array will return {@code -1}. *

* *
	 * StringUtils.indexOfAny(null, *)                  = -1
	 * StringUtils.indexOfAny("", *)                    = -1
	 * StringUtils.indexOfAny(*, null)                  = -1
	 * StringUtils.indexOfAny(*, [])                    = -1
	 * StringUtils.indexOfAny("zzabyycdxx", ['z', 'a']) = 0
	 * StringUtils.indexOfAny("zzabyycdxx", ['b', 'y']) = 3
	 * StringUtils.indexOfAny("aba", ['z'])             = -1
	 * 
* * @param cs the CharSequence to check, may be null * @param searchChars the chars to search for, may be null * @return the index of any of the chars, -1 if no match or null input * @since 2.0 * @since 3.0 Changed signature from indexOfAny(String, char[]) to * indexOfAny(CharSequence, char...) */ public static int indexOfAny(final CharSequence cs, final char... searchChars) { if (isEmpty(cs) || searchChars.length == 0) { return INDEX_NOT_FOUND; } final int csLen = cs.length(); final int csLast = csLen - 1; final int searchLen = searchChars.length; final int searchLast = searchLen - 1; for (int i = 0; i < csLen; i++) { final char ch = cs.charAt(i); for (int j = 0; j < searchLen; j++) { if (searchChars[j] == ch) { if (i < csLast && j < searchLast && Character.isHighSurrogate(ch)) { // ch is a supplementary character if (searchChars[j + 1] == cs.charAt(i + 1)) { return i; } } else { return i; } } } } return INDEX_NOT_FOUND; } /** *

* Find the first index of any of a set of potential substrings. *

* *

* A {@code null} CharSequence will return {@code -1}. A {@code null} or zero * length search array will return {@code -1}. A {@code null} search array entry * will be ignored, but a search array containing "" will return {@code 0} if * {@code str} is not null. This method uses {@link String#indexOf(String)} if * possible. *

* *
	 * StringUtils.indexOfAny(null, *)                      = -1
	 * StringUtils.indexOfAny(*, null)                      = -1
	 * StringUtils.indexOfAny(*, [])                        = -1
	 * StringUtils.indexOfAny("zzabyycdxx", ["ab", "cd"])   = 2
	 * StringUtils.indexOfAny("zzabyycdxx", ["cd", "ab"])   = 2
	 * StringUtils.indexOfAny("zzabyycdxx", ["mn", "op"])   = -1
	 * StringUtils.indexOfAny("zzabyycdxx", ["zab", "aby"]) = 1
	 * StringUtils.indexOfAny("zzabyycdxx", [""])           = 0
	 * StringUtils.indexOfAny("", [""])                     = 0
	 * StringUtils.indexOfAny("", ["a"])                    = -1
	 * 
* * @param str the CharSequence to check, may be null * @param searchStrs the CharSequences to search for, may be null * @return the first index of any of the searchStrs in str, -1 if no match * @since 3.0 Changed signature from indexOfAny(String, String[]) to * indexOfAny(CharSequence, CharSequence...) */ public static int indexOfAny(final CharSequence str, final CharSequence... searchStrs) { if (str == null || searchStrs == null) { return INDEX_NOT_FOUND; } // String's can't have a MAX_VALUEth index. int ret = Integer.MAX_VALUE; int tmp = 0; for (final CharSequence search : searchStrs) { if (search == null) { continue; } tmp = CharSequenceUtils.indexOf(str, search, 0); if (tmp == INDEX_NOT_FOUND) { continue; } if (tmp < ret) { ret = tmp; } } return ret == Integer.MAX_VALUE ? INDEX_NOT_FOUND : ret; } /** *

* Search a CharSequence to find the first index of any character in the given * set of characters. *

* *

* A {@code null} String will return {@code -1}. A {@code null} search string * will return {@code -1}. *

* *
	 * StringUtils.indexOfAny(null, *)            = -1
	 * StringUtils.indexOfAny("", *)              = -1
	 * StringUtils.indexOfAny(*, null)            = -1
	 * StringUtils.indexOfAny(*, "")              = -1
	 * StringUtils.indexOfAny("zzabyycdxx", "za") = 0
	 * StringUtils.indexOfAny("zzabyycdxx", "by") = 3
	 * StringUtils.indexOfAny("aba", "z")         = -1
	 * 
* * @param cs the CharSequence to check, may be null * @param searchChars the chars to search for, may be null * @return the index of any of the chars, -1 if no match or null input * @since 2.0 * @since 3.0 Changed signature from indexOfAny(String, String) to * indexOfAny(CharSequence, String) */ public static int indexOfAny(final CharSequence cs, final String searchChars) { if (isEmpty(cs) || isEmpty(searchChars)) { return INDEX_NOT_FOUND; } return indexOfAny(cs, searchChars.toCharArray()); } /** *

* Searches a CharSequence to find the first index of any character not in the * given set of characters. *

* *

* A {@code null} CharSequence will return {@code -1}. A {@code null} or zero * length search array will return {@code -1}. *

* *
	 * StringUtils.indexOfAnyBut(null, *)                              = -1
	 * StringUtils.indexOfAnyBut("", *)                                = -1
	 * StringUtils.indexOfAnyBut(*, null)                              = -1
	 * StringUtils.indexOfAnyBut(*, [])                                = -1
	 * StringUtils.indexOfAnyBut("zzabyycdxx", new char[] {'z', 'a'} ) = 3
	 * StringUtils.indexOfAnyBut("aba", new char[] {'z'} )             = 0
	 * StringUtils.indexOfAnyBut("aba", new char[] {'a', 'b'} )        = -1
	 * 
	 * 
* * @param cs the CharSequence to check, may be null * @param searchChars the chars to search for, may be null * @return the index of any of the chars, -1 if no match or null input * @since 2.0 * @since 3.0 Changed signature from indexOfAnyBut(String, char[]) to * indexOfAnyBut(CharSequence, char...) */ public static int indexOfAnyBut(final CharSequence cs, final char... searchChars) { if (isEmpty(cs) || searchChars.length == 0) { return INDEX_NOT_FOUND; } final int csLen = cs.length(); final int csLast = csLen - 1; final int searchLen = searchChars.length; final int searchLast = searchLen - 1; outer: for (int i = 0; i < csLen; i++) { final char ch = cs.charAt(i); for (int j = 0; j < searchLen; j++) { if (searchChars[j] == ch) { if (i < csLast && j < searchLast && Character.isHighSurrogate(ch)) { if (searchChars[j + 1] == cs.charAt(i + 1)) { continue outer; } } else { continue outer; } } } return i; } return INDEX_NOT_FOUND; } /** *

* Search a CharSequence to find the first index of any character not in the * given set of characters. *

* *

* A {@code null} CharSequence will return {@code -1}. A {@code null} or empty * search string will return {@code -1}. *

* *
	 * StringUtils.indexOfAnyBut(null, *)            = -1
	 * StringUtils.indexOfAnyBut("", *)              = -1
	 * StringUtils.indexOfAnyBut(*, null)            = -1
	 * StringUtils.indexOfAnyBut(*, "")              = -1
	 * StringUtils.indexOfAnyBut("zzabyycdxx", "za") = 3
	 * StringUtils.indexOfAnyBut("zzabyycdxx", "")   = -1
	 * StringUtils.indexOfAnyBut("aba", "ab")        = -1
	 * 
* * @param seq the CharSequence to check, may be null * @param searchChars the chars to search for, may be null * @return the index of any of the chars, -1 if no match or null input * @since 2.0 * @since 3.0 Changed signature from indexOfAnyBut(String, String) to * indexOfAnyBut(CharSequence, CharSequence) */ public static int indexOfAnyBut(final CharSequence seq, final CharSequence searchChars) { if (isEmpty(seq) || isEmpty(searchChars)) { return INDEX_NOT_FOUND; } final int strLen = seq.length(); for (int i = 0; i < strLen; i++) { final char ch = seq.charAt(i); final boolean chFound = CharSequenceUtils.indexOf(searchChars, ch, 0) >= 0; if (i + 1 < strLen && Character.isHighSurrogate(ch)) { final char ch2 = seq.charAt(i + 1); if (chFound && CharSequenceUtils.indexOf(searchChars, ch2, 0) < 0) { return i; } } else if (!chFound) { return i; } } return INDEX_NOT_FOUND; } /** *

* Compares all CharSequences in an array and returns the index at which the * CharSequences begin to differ. *

* *

* For example, {@code indexOfDifference(new String[] {"i am a machine", "i am a * robot"}) -> 7} *

* *
	 * StringUtils.indexOfDifference(null) = -1
	 * StringUtils.indexOfDifference(new String[] {}) = -1
	 * StringUtils.indexOfDifference(new String[] {"abc"}) = -1
	 * StringUtils.indexOfDifference(new String[] {null, null}) = -1
	 * StringUtils.indexOfDifference(new String[] {"", ""}) = -1
	 * StringUtils.indexOfDifference(new String[] {"", null}) = 0
	 * StringUtils.indexOfDifference(new String[] {"abc", null, null}) = 0
	 * StringUtils.indexOfDifference(new String[] {null, null, "abc"}) = 0
	 * StringUtils.indexOfDifference(new String[] {"", "abc"}) = 0
	 * StringUtils.indexOfDifference(new String[] {"abc", ""}) = 0
	 * StringUtils.indexOfDifference(new String[] {"abc", "abc"}) = -1
	 * StringUtils.indexOfDifference(new String[] {"abc", "a"}) = 1
	 * StringUtils.indexOfDifference(new String[] {"ab", "abxyz"}) = 2
	 * StringUtils.indexOfDifference(new String[] {"abcde", "abxyz"}) = 2
	 * StringUtils.indexOfDifference(new String[] {"abcde", "xyz"}) = 0
	 * StringUtils.indexOfDifference(new String[] {"xyz", "abcde"}) = 0
	 * StringUtils.indexOfDifference(new String[] {"i am a machine", "i am a robot"}) = 7
	 * 
* * @param css array of CharSequences, entries may be null * @return the index where the strings begin to differ; -1 if they are all equal * @since 2.4 * @since 3.0 Changed signature from indexOfDifference(String...) to * indexOfDifference(CharSequence...) */ public static int indexOfDifference(final CharSequence... css) { if (css.length <= 1) { return INDEX_NOT_FOUND; } boolean anyStringNull = false; boolean allStringsNull = true; final int arrayLen = css.length; int shortestStrLen = Integer.MAX_VALUE; int longestStrLen = 0; // find the min and max string lengths; this avoids checking to make // sure we are not exceeding the length of the string each time through // the bottom loop. for (final CharSequence cs : css) { if (cs == null) { anyStringNull = true; shortestStrLen = 0; } else { allStringsNull = false; shortestStrLen = Math.min(cs.length(), shortestStrLen); longestStrLen = Math.max(cs.length(), longestStrLen); } } // handle lists containing all nulls or all empty strings if (allStringsNull || longestStrLen == 0 && !anyStringNull) { return INDEX_NOT_FOUND; } // handle lists containing some nulls or some empty strings if (shortestStrLen == 0) { return 0; } // find the position with the first difference across all strings int firstDiff = -1; for (int stringPos = 0; stringPos < shortestStrLen; stringPos++) { final char comparisonChar = css[0].charAt(stringPos); for (int arrayPos = 1; arrayPos < arrayLen; arrayPos++) { if (css[arrayPos].charAt(stringPos) != comparisonChar) { firstDiff = stringPos; break; } } if (firstDiff != -1) { break; } } if (firstDiff == -1 && shortestStrLen != longestStrLen) { // we compared all of the characters up to the length of the // shortest string and didn't find a match, but the string lengths // vary, so return the length of the shortest string. return shortestStrLen; } return firstDiff; } /** *

* Compares two CharSequences, and returns the index at which the CharSequences * begin to differ. *

* *

* For example, {@code indexOfDifference("i am a machine", "i am a robot") -> 7} *

* *
	 * StringUtils.indexOfDifference(null, null) = -1
	 * StringUtils.indexOfDifference("", "") = -1
	 * StringUtils.indexOfDifference("", "abc") = 0
	 * StringUtils.indexOfDifference("abc", "") = 0
	 * StringUtils.indexOfDifference("abc", "abc") = -1
	 * StringUtils.indexOfDifference("ab", "abxyz") = 2
	 * StringUtils.indexOfDifference("abcde", "abxyz") = 2
	 * StringUtils.indexOfDifference("abcde", "xyz") = 0
	 * 
* * @param cs1 the first CharSequence, may be null * @param cs2 the second CharSequence, may be null * @return the index where cs1 and cs2 begin to differ; -1 if they are equal * @since 2.0 * @since 3.0 Changed signature from indexOfDifference(String, String) to * indexOfDifference(CharSequence, CharSequence) */ public static int indexOfDifference(final CharSequence cs1, final CharSequence cs2) { if (cs1 == cs2) { return INDEX_NOT_FOUND; } if (cs1 == null || cs2 == null) { return 0; } int i; for (i = 0; i < cs1.length() && i < cs2.length(); ++i) { if (cs1.charAt(i) != cs2.charAt(i)) { break; } } if (i < cs2.length() || i < cs1.length()) { return i; } return INDEX_NOT_FOUND; } /** *

* Case in-sensitive find of the first index within a CharSequence. *

* *

* A {@code null} CharSequence will return {@code -1}. A negative start position * is treated as zero. An empty ("") search CharSequence always matches. A start * position greater than the string length only matches an empty search * CharSequence. *

* *
	 * StringUtils.indexOfIgnoreCase(null, *)          = -1
	 * StringUtils.indexOfIgnoreCase(*, null)          = -1
	 * StringUtils.indexOfIgnoreCase("", "")           = 0
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "a")  = 0
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "b")  = 2
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "ab") = 1
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @return the first index of the search CharSequence, -1 if no match or * {@code null} string input * @since 2.5 * @since 3.0 Changed signature from indexOfIgnoreCase(String, String) to * indexOfIgnoreCase(CharSequence, CharSequence) */ public static int indexOfIgnoreCase(final CharSequence str, final CharSequence searchStr) { return indexOfIgnoreCase(str, searchStr, 0); } /** *

* Case in-sensitive find of the first index within a CharSequence from the * specified position. *

* *

* A {@code null} CharSequence will return {@code -1}. A negative start position * is treated as zero. An empty ("") search CharSequence always matches. A start * position greater than the string length only matches an empty search * CharSequence. *

* *
	 * StringUtils.indexOfIgnoreCase(null, *, *)          = -1
	 * StringUtils.indexOfIgnoreCase(*, null, *)          = -1
	 * StringUtils.indexOfIgnoreCase("", "", 0)           = 0
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "A", 0)  = 0
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "B", 0)  = 2
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "AB", 0) = 1
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "B", 3)  = 5
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "B", 9)  = -1
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "B", -1) = 2
	 * StringUtils.indexOfIgnoreCase("aabaabaa", "", 2)   = 2
	 * StringUtils.indexOfIgnoreCase("abc", "", 9)        = -1
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @param startPos the start position, negative treated as zero * @return the first index of the search CharSequence (always ≥ startPos), -1 * if no match or {@code null} string input * @since 2.5 * @since 3.0 Changed signature from indexOfIgnoreCase(String, String, int) to * indexOfIgnoreCase(CharSequence, CharSequence, int) */ public static int indexOfIgnoreCase(final CharSequence str, final CharSequence searchStr, int startPos) { if (str == null || searchStr == null) { return INDEX_NOT_FOUND; } if (startPos < 0) { startPos = 0; } final int endLimit = str.length() - searchStr.length() + 1; if (startPos > endLimit) { return INDEX_NOT_FOUND; } if (searchStr.length() == 0) { return startPos; } for (int i = startPos; i < endLimit; i++) { if (CharSequenceUtils.regionMatches(str, true, i, searchStr, 0, searchStr.length())) { return i; } } return INDEX_NOT_FOUND; } /** *

* Checks if all of the CharSequences are empty (""), null or whitespace only. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.isAllBlank(null)             = true
	 * StringUtils.isAllBlank(null, "foo")      = false
	 * StringUtils.isAllBlank(null, null)       = true
	 * StringUtils.isAllBlank("", "bar")        = false
	 * StringUtils.isAllBlank("bob", "")        = false
	 * StringUtils.isAllBlank("  bob  ", null)  = false
	 * StringUtils.isAllBlank(" ", "bar")       = false
	 * StringUtils.isAllBlank("foo", "bar")     = false
	 * StringUtils.isAllBlank(new String[] {})  = true
	 * 
* * @param css the CharSequences to check, may be null or empty * @return {@code true} if all of the CharSequences are empty or null or * whitespace only * @since 3.6 */ public static boolean isAllBlank(final CharSequence... css) { if (css.length == 0) { return true; } for (final CharSequence cs : css) { if (isNotBlank(cs)) { return false; } } return true; } /** *

* Checks if all of the CharSequences are empty ("") or null. *

* *
	 * StringUtils.isAllEmpty(null)             = true
	 * StringUtils.isAllEmpty(null, "")         = true
	 * StringUtils.isAllEmpty(new String[] {})  = true
	 * StringUtils.isAllEmpty(null, "foo")      = false
	 * StringUtils.isAllEmpty("", "bar")        = false
	 * StringUtils.isAllEmpty("bob", "")        = false
	 * StringUtils.isAllEmpty("  bob  ", null)  = false
	 * StringUtils.isAllEmpty(" ", "bar")       = false
	 * StringUtils.isAllEmpty("foo", "bar")     = false
	 * 
* * @param css the CharSequences to check, may be null or empty * @return {@code true} if all of the CharSequences are empty or null * @since 3.6 */ public static boolean isAllEmpty(final CharSequence... css) { if (css.length == 0) { return true; } for (final CharSequence cs : css) { if (isNotEmpty(cs)) { return false; } } return true; } /** *

* Checks if the CharSequence contains only lowercase characters. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code false}. *

* *
	 * StringUtils.isAllLowerCase(null)   = false
	 * StringUtils.isAllLowerCase("")     = false
	 * StringUtils.isAllLowerCase("  ")   = false
	 * StringUtils.isAllLowerCase("abc")  = true
	 * StringUtils.isAllLowerCase("abC")  = false
	 * StringUtils.isAllLowerCase("ab c") = false
	 * StringUtils.isAllLowerCase("ab1c") = false
	 * StringUtils.isAllLowerCase("ab/c") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains lowercase characters, and is non-null * @since 2.5 * @since 3.0 Changed signature from isAllLowerCase(String) to * isAllLowerCase(CharSequence) */ public static boolean isAllLowerCase(final CharSequence cs) { if (isEmpty(cs)) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!Character.isLowerCase(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if the CharSequence contains only uppercase characters. *

* *

* {@code null} will return {@code false}. An empty String (length()=0) will * return {@code false}. *

* *
	 * StringUtils.isAllUpperCase(null)   = false
	 * StringUtils.isAllUpperCase("")     = false
	 * StringUtils.isAllUpperCase("  ")   = false
	 * StringUtils.isAllUpperCase("ABC")  = true
	 * StringUtils.isAllUpperCase("aBC")  = false
	 * StringUtils.isAllUpperCase("A C")  = false
	 * StringUtils.isAllUpperCase("A1C")  = false
	 * StringUtils.isAllUpperCase("A/C")  = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains uppercase characters, and is non-null * @since 2.5 * @since 3.0 Changed signature from isAllUpperCase(String) to * isAllUpperCase(CharSequence) */ public static boolean isAllUpperCase(final CharSequence cs) { if (isEmpty(cs)) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!Character.isUpperCase(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if the CharSequence contains only Unicode letters. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code false}. *

* *
	 * StringUtils.isAlpha(null)   = false
	 * StringUtils.isAlpha("")     = false
	 * StringUtils.isAlpha("  ")   = false
	 * StringUtils.isAlpha("abc")  = true
	 * StringUtils.isAlpha("ab2c") = false
	 * StringUtils.isAlpha("ab-c") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains letters, and is non-null * @since 3.0 Changed signature from isAlpha(String) to isAlpha(CharSequence) * @since 3.0 Changed "" to return false and not true */ public static boolean isAlpha(final CharSequence cs) { if (isEmpty(cs)) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!Character.isLetter(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if the CharSequence contains only Unicode letters or digits. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code false}. *

* *
	 * StringUtils.isAlphanumeric(null)   = false
	 * StringUtils.isAlphanumeric("")     = false
	 * StringUtils.isAlphanumeric("  ")   = false
	 * StringUtils.isAlphanumeric("abc")  = true
	 * StringUtils.isAlphanumeric("ab c") = false
	 * StringUtils.isAlphanumeric("ab2c") = true
	 * StringUtils.isAlphanumeric("ab-c") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains letters or digits, and is non-null * @since 3.0 Changed signature from isAlphanumeric(String) to * isAlphanumeric(CharSequence) * @since 3.0 Changed "" to return false and not true */ public static boolean isAlphanumeric(final CharSequence cs) { if (isEmpty(cs)) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!Character.isLetterOrDigit(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if the CharSequence contains only Unicode letters, digits or space * ({@code ' '}). *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code true}. *

* *
	 * StringUtils.isAlphanumericSpace(null)   = false
	 * StringUtils.isAlphanumericSpace("")     = true
	 * StringUtils.isAlphanumericSpace("  ")   = true
	 * StringUtils.isAlphanumericSpace("abc")  = true
	 * StringUtils.isAlphanumericSpace("ab c") = true
	 * StringUtils.isAlphanumericSpace("ab2c") = true
	 * StringUtils.isAlphanumericSpace("ab-c") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains letters, digits or space, and is * non-null * @since 3.0 Changed signature from isAlphanumericSpace(String) to * isAlphanumericSpace(CharSequence) */ public static boolean isAlphanumericSpace(final CharSequence cs) { if (cs == null) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { final char nowChar = cs.charAt(i); if (nowChar != ' ' && !Character.isLetterOrDigit(nowChar)) { return false; } } return true; } /** *

* Checks if the CharSequence contains only Unicode letters and space (' '). *

* *

* {@code null} will return {@code false} An empty CharSequence (length()=0) * will return {@code true}. *

* *
	 * StringUtils.isAlphaSpace(null)   = false
	 * StringUtils.isAlphaSpace("")     = true
	 * StringUtils.isAlphaSpace("  ")   = true
	 * StringUtils.isAlphaSpace("abc")  = true
	 * StringUtils.isAlphaSpace("ab c") = true
	 * StringUtils.isAlphaSpace("ab2c") = false
	 * StringUtils.isAlphaSpace("ab-c") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains letters and space, and is non-null * @since 3.0 Changed signature from isAlphaSpace(String) to * isAlphaSpace(CharSequence) */ public static boolean isAlphaSpace(final CharSequence cs) { if (cs == null) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { final char nowChar = cs.charAt(i); if (nowChar != ' ' && !Character.isLetter(nowChar)) { return false; } } return true; } /** *

* Checks if any of the CharSequences are empty ("") or null or whitespace only. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.isAnyBlank((String) null)    = true
	 * StringUtils.isAnyBlank((String[]) null)  = false
	 * StringUtils.isAnyBlank(null, "foo")      = true
	 * StringUtils.isAnyBlank(null, null)       = true
	 * StringUtils.isAnyBlank("", "bar")        = true
	 * StringUtils.isAnyBlank("bob", "")        = true
	 * StringUtils.isAnyBlank("  bob  ", null)  = true
	 * StringUtils.isAnyBlank(" ", "bar")       = true
	 * StringUtils.isAnyBlank(new String[] {})  = false
	 * StringUtils.isAnyBlank(new String[]{""}) = true
	 * StringUtils.isAnyBlank("foo", "bar")     = false
	 * 
* * @param css the CharSequences to check, may be null or empty * @return {@code true} if any of the CharSequences are empty or null or * whitespace only * @since 3.2 */ public static boolean isAnyBlank(final CharSequence... css) { if (css.length == 0) { return false; } for (final CharSequence cs : css) { if (isBlank(cs)) { return true; } } return false; } /** *

* Checks if any of the CharSequences are empty ("") or null. *

* *
	 * StringUtils.isAnyEmpty((String) null)    = true
	 * StringUtils.isAnyEmpty((String[]) null)  = false
	 * StringUtils.isAnyEmpty(null, "foo")      = true
	 * StringUtils.isAnyEmpty("", "bar")        = true
	 * StringUtils.isAnyEmpty("bob", "")        = true
	 * StringUtils.isAnyEmpty("  bob  ", null)  = true
	 * StringUtils.isAnyEmpty(" ", "bar")       = false
	 * StringUtils.isAnyEmpty("foo", "bar")     = false
	 * StringUtils.isAnyEmpty(new String[]{})   = false
	 * StringUtils.isAnyEmpty(new String[]{""}) = true
	 * 
* * @param css the CharSequences to check, may be null or empty * @return {@code true} if any of the CharSequences are empty or null * @since 3.2 */ public static boolean isAnyEmpty(final CharSequence... css) { if (css.length == 0) { return false; } for (final CharSequence cs : css) { if (isEmpty(cs)) { return true; } } return false; } /** *

* Checks if the CharSequence contains only ASCII printable characters. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code true}. *

* *
	 * StringUtils.isAsciiPrintable(null)     = false
	 * StringUtils.isAsciiPrintable("")       = true
	 * StringUtils.isAsciiPrintable(" ")      = true
	 * StringUtils.isAsciiPrintable("Ceki")   = true
	 * StringUtils.isAsciiPrintable("ab2c")   = true
	 * StringUtils.isAsciiPrintable("!ab-c~") = true
	 * StringUtils.isAsciiPrintable("\u0020") = true
	 * StringUtils.isAsciiPrintable("\u0021") = true
	 * StringUtils.isAsciiPrintable("\u007e") = true
	 * StringUtils.isAsciiPrintable("\u007f") = false
	 * StringUtils.isAsciiPrintable("Ceki G\u00fclc\u00fc") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if every character is in the range 32 thru 126 * @since 2.1 * @since 3.0 Changed signature from isAsciiPrintable(String) to * isAsciiPrintable(CharSequence) */ public static boolean isAsciiPrintable(final CharSequence cs) { if (cs == null) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!CharUtils.isAsciiPrintable(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if a CharSequence is empty (""), null or whitespace only. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.isBlank(null)      = true
	 * StringUtils.isBlank("")        = true
	 * StringUtils.isBlank(" ")       = true
	 * StringUtils.isBlank("bob")     = false
	 * StringUtils.isBlank("  bob  ") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if the CharSequence is null, empty or whitespace only * @since 2.0 * @since 3.0 Changed signature from isBlank(String) to isBlank(CharSequence) */ public static boolean isBlank(final CharSequence cs) { final int strLen = length(cs); if (strLen == 0) { return true; } for (int i = 0; i < strLen; i++) { if (!Character.isWhitespace(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if a CharSequence is empty ("") or null. *

* *
	 * StringUtils.isEmpty(null)      = true
	 * StringUtils.isEmpty("")        = true
	 * StringUtils.isEmpty(" ")       = false
	 * StringUtils.isEmpty("bob")     = false
	 * StringUtils.isEmpty("  bob  ") = false
	 * 
* *

* NOTE: This method changed in Lang version 2.0. It no longer trims the * CharSequence. That functionality is available in isBlank(). *

* * @param cs the CharSequence to check, may be null * @return {@code true} if the CharSequence is empty or null * @since 3.0 Changed signature from isEmpty(String) to isEmpty(CharSequence) */ public static boolean isEmpty(final CharSequence cs) { return cs == null || cs.length() == 0; } /** *

* Checks if the CharSequence contains mixed casing of both uppercase and * lowercase characters. *

* *

* {@code null} will return {@code false}. An empty CharSequence * ({@code length()=0}) will return {@code false}. *

* *
	 * StringUtils.isMixedCase(null)    = false
	 * StringUtils.isMixedCase("")      = false
	 * StringUtils.isMixedCase("ABC")   = false
	 * StringUtils.isMixedCase("abc")   = false
	 * StringUtils.isMixedCase("aBc")   = true
	 * StringUtils.isMixedCase("A c")   = true
	 * StringUtils.isMixedCase("A1c")   = true
	 * StringUtils.isMixedCase("a/C")   = true
	 * StringUtils.isMixedCase("aC\t")  = true
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if the CharSequence contains both uppercase and * lowercase characters * @since 3.5 */ public static boolean isMixedCase(final CharSequence cs) { if (isEmpty(cs) || cs.length() == 1) { return false; } boolean containsUppercase = false; boolean containsLowercase = false; final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (containsUppercase && containsLowercase) { return true; } else if (Character.isUpperCase(cs.charAt(i))) { containsUppercase = true; } else if (Character.isLowerCase(cs.charAt(i))) { containsLowercase = true; } } return containsUppercase && containsLowercase; } /** *

* Checks if none of the CharSequences are empty (""), null or whitespace only. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.isNoneBlank((String) null)    = false
	 * StringUtils.isNoneBlank((String[]) null)  = true
	 * StringUtils.isNoneBlank(null, "foo")      = false
	 * StringUtils.isNoneBlank(null, null)       = false
	 * StringUtils.isNoneBlank("", "bar")        = false
	 * StringUtils.isNoneBlank("bob", "")        = false
	 * StringUtils.isNoneBlank("  bob  ", null)  = false
	 * StringUtils.isNoneBlank(" ", "bar")       = false
	 * StringUtils.isNoneBlank(new String[] {})  = true
	 * StringUtils.isNoneBlank(new String[]{""}) = false
	 * StringUtils.isNoneBlank("foo", "bar")     = true
	 * 
* * @param css the CharSequences to check, may be null or empty * @return {@code true} if none of the CharSequences are empty or null or * whitespace only * @since 3.2 */ public static boolean isNoneBlank(final CharSequence... css) { return !isAnyBlank(css); } /** *

* Checks if none of the CharSequences are empty ("") or null. *

* *
	 * StringUtils.isNoneEmpty((String) null)    = false
	 * StringUtils.isNoneEmpty((String[]) null)  = true
	 * StringUtils.isNoneEmpty(null, "foo")      = false
	 * StringUtils.isNoneEmpty("", "bar")        = false
	 * StringUtils.isNoneEmpty("bob", "")        = false
	 * StringUtils.isNoneEmpty("  bob  ", null)  = false
	 * StringUtils.isNoneEmpty(new String[] {})  = true
	 * StringUtils.isNoneEmpty(new String[]{""}) = false
	 * StringUtils.isNoneEmpty(" ", "bar")       = true
	 * StringUtils.isNoneEmpty("foo", "bar")     = true
	 * 
* * @param css the CharSequences to check, may be null or empty * @return {@code true} if none of the CharSequences are empty or null * @since 3.2 */ public static boolean isNoneEmpty(final CharSequence... css) { return !isAnyEmpty(css); } /** *

* Checks if a CharSequence is not empty (""), not null and not whitespace only. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.isNotBlank(null)      = false
	 * StringUtils.isNotBlank("")        = false
	 * StringUtils.isNotBlank(" ")       = false
	 * StringUtils.isNotBlank("bob")     = true
	 * StringUtils.isNotBlank("  bob  ") = true
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if the CharSequence is not empty and not null and not * whitespace only * @since 2.0 * @since 3.0 Changed signature from isNotBlank(String) to * isNotBlank(CharSequence) */ public static boolean isNotBlank(final CharSequence cs) { return !isBlank(cs); } /** *

* Checks if a CharSequence is not empty ("") and not null. *

* *
	 * StringUtils.isNotEmpty(null)      = false
	 * StringUtils.isNotEmpty("")        = false
	 * StringUtils.isNotEmpty(" ")       = true
	 * StringUtils.isNotEmpty("bob")     = true
	 * StringUtils.isNotEmpty("  bob  ") = true
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if the CharSequence is not empty and not null * @since 3.0 Changed signature from isNotEmpty(String) to * isNotEmpty(CharSequence) */ public static boolean isNotEmpty(final CharSequence cs) { return !isEmpty(cs); } /** *

* Checks if the CharSequence contains only Unicode digits. A decimal point is * not a Unicode digit and returns false. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code false}. *

* *

* Note that the method does not allow for a leading sign, either positive or * negative. Also, if a String passes the numeric test, it may still generate a * NumberFormatException when parsed by Integer.parseInt or Long.parseLong, e.g. * if the value is outside the range for int or long respectively. *

* *
	 * StringUtils.isNumeric(null)   = false
	 * StringUtils.isNumeric("")     = false
	 * StringUtils.isNumeric("  ")   = false
	 * StringUtils.isNumeric("123")  = true
	 * StringUtils.isNumeric("\u0967\u0968\u0969")  = true
	 * StringUtils.isNumeric("12 3") = false
	 * StringUtils.isNumeric("ab2c") = false
	 * StringUtils.isNumeric("12-3") = false
	 * StringUtils.isNumeric("12.3") = false
	 * StringUtils.isNumeric("-123") = false
	 * StringUtils.isNumeric("+123") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains digits, and is non-null * @since 3.0 Changed signature from isNumeric(String) to * isNumeric(CharSequence) * @since 3.0 Changed "" to return false and not true */ public static boolean isNumeric(final CharSequence cs) { if (isEmpty(cs)) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!Character.isDigit(cs.charAt(i))) { return false; } } return true; } /** *

* Checks if the CharSequence contains only Unicode digits or space * ({@code ' '}). A decimal point is not a Unicode digit and returns false. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code true}. *

* *
	 * StringUtils.isNumericSpace(null)   = false
	 * StringUtils.isNumericSpace("")     = true
	 * StringUtils.isNumericSpace("  ")   = true
	 * StringUtils.isNumericSpace("123")  = true
	 * StringUtils.isNumericSpace("12 3") = true
	 * StringUtils.isNumericSpace("\u0967\u0968\u0969")  = true
	 * StringUtils.isNumericSpace("\u0967\u0968 \u0969")  = true
	 * StringUtils.isNumericSpace("ab2c") = false
	 * StringUtils.isNumericSpace("12-3") = false
	 * StringUtils.isNumericSpace("12.3") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains digits or space, and is non-null * @since 3.0 Changed signature from isNumericSpace(String) to * isNumericSpace(CharSequence) */ public static boolean isNumericSpace(final CharSequence cs) { if (cs == null) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { final char nowChar = cs.charAt(i); if (nowChar != ' ' && !Character.isDigit(nowChar)) { return false; } } return true; } /** *

* Checks if the CharSequence contains only whitespace. *

* *

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *

* {@code null} will return {@code false}. An empty CharSequence (length()=0) * will return {@code true}. *

* *
	 * StringUtils.isWhitespace(null)   = false
	 * StringUtils.isWhitespace("")     = true
	 * StringUtils.isWhitespace("  ")   = true
	 * StringUtils.isWhitespace("abc")  = false
	 * StringUtils.isWhitespace("ab2c") = false
	 * StringUtils.isWhitespace("ab-c") = false
	 * 
* * @param cs the CharSequence to check, may be null * @return {@code true} if only contains whitespace, and is non-null * @since 2.0 * @since 3.0 Changed signature from isWhitespace(String) to * isWhitespace(CharSequence) */ public static boolean isWhitespace(final CharSequence cs) { if (cs == null) { return false; } final int sz = cs.length(); for (int i = 0; i < sz; i++) { if (!Character.isWhitespace(cs.charAt(i))) { return false; } } return true; } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)              = null
	 * StringUtils.join([], *)                = ""
	 * StringUtils.join([null], *)            = ""
	 * StringUtils.join([false, false], ';')  = "false;false"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 3.12.0 */ public static String join(final boolean[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)                   = null
	 * StringUtils.join([], *)                     = ""
	 * StringUtils.join([null], *)                 = ""
	 * StringUtils.join([true, false, true], ';')  = "true;false;true"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.12.0 */ public static String join(final boolean[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final byte[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final byte[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final char[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final char[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final double[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final double[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final float[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final float[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param separator the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final int[] array, final char separator) { if (array == null) { return null; } return join(array, separator, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final int[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided {@code Iterable} into a single String * containing the provided elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the iteration are represented by empty strings. *

* *

* See the examples here: {@link #join(Object[],char)}. *

* * @param iterable the {@code Iterable} providing the values to join together, * may be null * @param separator the separator character to use * @return the joined String, {@code null} if null iterator input * @since 2.3 */ public static String join(final Iterable iterable, final char separator) { if (iterable == null) { return null; } return join(iterable.iterator(), separator); } /** *

* Joins the elements of the provided {@code Iterable} into a single String * containing the provided elements. *

* *

* No delimiter is added before or after the list. A {@code null} separator is * the same as an empty String (""). *

* *

* See the examples here: {@link #join(Object[],String)}. *

* * @param iterable the {@code Iterable} providing the values to join together, * may be null * @param separator the separator character to use, null treated as "" * @return the joined String, {@code null} if null iterator input * @since 2.3 */ public static String join(final Iterable iterable, final String separator) { if (iterable == null) { return null; } return join(iterable.iterator(), separator); } /** *

* Joins the elements of the provided {@code Iterator} into a single String * containing the provided elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the iteration are represented by empty strings. *

* *

* See the examples here: {@link #join(Object[],char)}. *

* * @param iterator the {@code Iterator} of values to join together, may be null * @param separator the separator character to use * @return the joined String, {@code null} if null iterator input * @since 2.0 */ public static String join(final Iterator iterator, final char separator) { // handle null, zero and one elements before building a buffer if (iterator == null) { return null; } if (!iterator.hasNext()) { return EMPTY; } final Object first = iterator.next(); if (!iterator.hasNext()) { return toStringOrEmpty(first); } // two or more elements final StringBuilder buf = new StringBuilder(STRING_BUILDER_SIZE); // Java default is 16, probably too small if (first != null) { buf.append(first); } while (iterator.hasNext()) { buf.append(separator); final Object obj = iterator.next(); if (obj != null) { buf.append(obj); } } return buf.toString(); } /** *

* Joins the elements of the provided {@code Iterator} into a single String * containing the provided elements. *

* *

* No delimiter is added before or after the list. A {@code null} separator is * the same as an empty String (""). *

* *

* See the examples here: {@link #join(Object[],String)}. *

* * @param iterator the {@code Iterator} of values to join together, may be null * @param separator the separator character to use, null treated as "" * @return the joined String, {@code null} if null iterator input */ public static String join(final Iterator iterator, final String separator) { // handle null, zero and one elements before building a buffer if (iterator == null) { return null; } if (!iterator.hasNext()) { return EMPTY; } final Object first = iterator.next(); if (!iterator.hasNext()) { return Objects.toString(first, ""); } // two or more elements final StringBuilder buf = new StringBuilder(STRING_BUILDER_SIZE); // Java default is 16, probably too small if (first != null) { buf.append(first); } while (iterator.hasNext()) { if (separator != null) { buf.append(separator); } final Object obj = iterator.next(); if (obj != null) { buf.append(obj); } } return buf.toString(); } /** *

* Joins the elements of the provided {@code List} into a single String * containing the provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join(["a", "b", "c"], ';')  = "a;b;c"
	 * StringUtils.join(["a", "b", "c"], null) = "abc"
	 * StringUtils.join([null, "", "a"], ';')  = ";;a"
	 * 
* * @param list the {@code List} of values to join together, may be null * @param separator the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the list * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the list * @return the joined String, {@code null} if null list input * @since 3.8 */ public static String join(final List list, final char separator, final int startIndex, final int endIndex) { if (list == null) { return null; } final int noOfItems = endIndex - startIndex; if (noOfItems <= 0) { return EMPTY; } final List subList = list.subList(startIndex, endIndex); return join(subList.iterator(), separator); } /** *

* Joins the elements of the provided {@code List} into a single String * containing the provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join(["a", "b", "c"], ';')  = "a;b;c"
	 * StringUtils.join(["a", "b", "c"], null) = "abc"
	 * StringUtils.join([null, "", "a"], ';')  = ";;a"
	 * 
* * @param list the {@code List} of values to join together, may be null * @param separator the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the list * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the list * @return the joined String, {@code null} if null list input * @since 3.8 */ public static String join(final List list, final String separator, final int startIndex, final int endIndex) { if (list == null) { return null; } final int noOfItems = endIndex - startIndex; if (noOfItems <= 0) { return EMPTY; } final List subList = list.subList(startIndex, endIndex); return join(subList.iterator(), separator); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param separator the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final long[] array, final char separator) { if (array == null) { return null; } return join(array, separator, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final long[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join(["a", "b", "c"], ';')  = "a;b;c"
	 * StringUtils.join(["a", "b", "c"], null) = "abc"
	 * StringUtils.join([null, "", "a"], ';')  = ";;a"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 2.0 */ public static String join(final Object[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join(["a", "b", "c"], ';')  = "a;b;c"
	 * StringUtils.join(["a", "b", "c"], null) = "abc"
	 * StringUtils.join([null, "", "a"], ';')  = ";;a"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 2.0 */ public static String join(final Object[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(toStringOrEmpty(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. A {@code null} separator is * the same as an empty String (""). Null objects or empty strings within the * array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)                = null
	 * StringUtils.join([], *)                  = ""
	 * StringUtils.join([null], *)              = ""
	 * StringUtils.join(["a", "b", "c"], "--")  = "a--b--c"
	 * StringUtils.join(["a", "b", "c"], null)  = "abc"
	 * StringUtils.join(["a", "b", "c"], "")    = "abc"
	 * StringUtils.join([null, "", "a"], ',')   = ",,a"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use, null treated as "" * @return the joined String, {@code null} if null array input */ public static String join(final Object[] array, final String delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. A {@code null} separator is * the same as an empty String (""). Null objects or empty strings within the * array are represented by empty strings. *

* *
	 * StringUtils.join(null, *, *, *)                = null
	 * StringUtils.join([], *, *, *)                  = ""
	 * StringUtils.join([null], *, *, *)              = ""
	 * StringUtils.join(["a", "b", "c"], "--", 0, 3)  = "a--b--c"
	 * StringUtils.join(["a", "b", "c"], "--", 1, 3)  = "b--c"
	 * StringUtils.join(["a", "b", "c"], "--", 2, 3)  = "c"
	 * StringUtils.join(["a", "b", "c"], "--", 2, 2)  = ""
	 * StringUtils.join(["a", "b", "c"], null, 0, 3)  = "abc"
	 * StringUtils.join(["a", "b", "c"], "", 0, 3)    = "abc"
	 * StringUtils.join([null, "", "a"], ',', 0, 3)   = ",,a"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use, null treated as "" * @param startIndex the first index to start joining from. * @param endIndex the index to stop joining from (exclusive). * @return the joined String, {@code null} if null array input; or the empty * string if {@code endIndex - startIndex <= 0}. The number of joined * entries is given by {@code endIndex - startIndex} * @throws ArrayIndexOutOfBoundsException ife
* {@code startIndex < 0} or
* {@code startIndex >= array.length()} * or
* {@code endIndex < 0} or
* {@code endIndex > array.length()} */ public static String join(final Object[] array, final String delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = new StringJoiner(toStringOrEmpty(delimiter)); for (int i = startIndex; i < endIndex; i++) { joiner.add(toStringOrEmpty(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final short[] array, final char delimiter) { if (array == null) { return null; } return join(array, delimiter, 0, array.length); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No delimiter is added before or after the list. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null, *)               = null
	 * StringUtils.join([], *)                 = ""
	 * StringUtils.join([null], *)             = ""
	 * StringUtils.join([1, 2, 3], ';')  = "1;2;3"
	 * StringUtils.join([1, 2, 3], null) = "123"
	 * 
* * @param array the array of values to join together, may be null * @param delimiter the separator character to use * @param startIndex the first index to start joining from. It is an error to * pass in a start index past the end of the array * @param endIndex the index to stop joining from (exclusive). It is an error * to pass in an end index past the end of the array * @return the joined String, {@code null} if null array input * @since 3.2 */ public static String join(final short[] array, final char delimiter, final int startIndex, final int endIndex) { if (array == null) { return null; } if (endIndex - startIndex <= 0) { return EMPTY; } final StringJoiner joiner = newStringJoiner(delimiter); for (int i = startIndex; i < endIndex; i++) { joiner.add(String.valueOf(array[i])); } return joiner.toString(); } /** *

* Joins the elements of the provided array into a single String containing the * provided list of elements. *

* *

* No separator is added to the joined String. Null objects or empty strings * within the array are represented by empty strings. *

* *
	 * StringUtils.join(null)            = null
	 * StringUtils.join([])              = ""
	 * StringUtils.join([null])          = ""
	 * StringUtils.join(["a", "b", "c"]) = "abc"
	 * StringUtils.join([null, "", "a"]) = "a"
	 * 
* * @param the specific type of values to join together * @param elements the values to join together, may be null * @return the joined String, {@code null} if null array input * @since 2.0 * @since 3.0 Changed signature to use varargs */ @SafeVarargs public static String join(final T... elements) { return join(elements, null); } /** *

* Joins the elements of the provided varargs into a single String containing * the provided elements. *

* *

* No delimiter is added before or after the list. {@code null} elements and * separator are treated as empty Strings (""). *

* *
	 * StringUtils.joinWith(",", {"a", "b"})        = "a,b"
	 * StringUtils.joinWith(",", {"a", "b",""})     = "a,b,"
	 * StringUtils.joinWith(",", {"a", null, "b"})  = "a,,b"
	 * StringUtils.joinWith(null, {"a", "b"})       = "ab"
	 * 
* * @param delimiter the separator character to use, null treated as "" * @param array the varargs providing the values to join together. * {@code null} elements are treated as "" * @return the joined String. * @throws java.lang.IllegalArgumentException if a null varargs is provided * @since 3.5 */ public static String joinWith(final String delimiter, final Object... array) { if (array == null) { throw new IllegalArgumentException("Object varargs must not be null"); } return join(array, delimiter); } /** *

* Finds the last index within a CharSequence, handling {@code null}. This * method uses {@link String#lastIndexOf(String)} if possible. *

* *

* A {@code null} CharSequence will return {@code -1}. *

* *
	 * StringUtils.lastIndexOf(null, *)          = -1
	 * StringUtils.lastIndexOf(*, null)          = -1
	 * StringUtils.lastIndexOf("", "")           = 0
	 * StringUtils.lastIndexOf("aabaabaa", "a")  = 7
	 * StringUtils.lastIndexOf("aabaabaa", "b")  = 5
	 * StringUtils.lastIndexOf("aabaabaa", "ab") = 4
	 * StringUtils.lastIndexOf("aabaabaa", "")   = 8
	 * 
* * @param seq the CharSequence to check, may be null * @param searchSeq the CharSequence to find, may be null * @return the last index of the search String, -1 if no match or {@code null} * string input * @since 2.0 * @since 3.0 Changed signature from lastIndexOf(String, String) to * lastIndexOf(CharSequence, CharSequence) */ public static int lastIndexOf(final CharSequence seq, final CharSequence searchSeq) { if (seq == null) { return INDEX_NOT_FOUND; } return CharSequenceUtils.lastIndexOf(seq, searchSeq, seq.length()); } /** *

* Finds the last index within a CharSequence, handling {@code null}. This * method uses {@link String#lastIndexOf(String, int)} if possible. *

* *

* A {@code null} CharSequence will return {@code -1}. A negative start position * returns {@code -1}. An empty ("") search CharSequence always matches unless * the start position is negative. A start position greater than the string * length searches the whole string. The search starts at the startPos and works * backwards; matches starting after the start position are ignored. *

* *
	 * StringUtils.lastIndexOf(null, *, *)          = -1
	 * StringUtils.lastIndexOf(*, null, *)          = -1
	 * StringUtils.lastIndexOf("aabaabaa", "a", 8)  = 7
	 * StringUtils.lastIndexOf("aabaabaa", "b", 8)  = 5
	 * StringUtils.lastIndexOf("aabaabaa", "ab", 8) = 4
	 * StringUtils.lastIndexOf("aabaabaa", "b", 9)  = 5
	 * StringUtils.lastIndexOf("aabaabaa", "b", -1) = -1
	 * StringUtils.lastIndexOf("aabaabaa", "a", 0)  = 0
	 * StringUtils.lastIndexOf("aabaabaa", "b", 0)  = -1
	 * StringUtils.lastIndexOf("aabaabaa", "b", 1)  = -1
	 * StringUtils.lastIndexOf("aabaabaa", "b", 2)  = 2
	 * StringUtils.lastIndexOf("aabaabaa", "ba", 2)  = 2
	 * 
* * @param seq the CharSequence to check, may be null * @param searchSeq the CharSequence to find, may be null * @param startPos the start position, negative treated as zero * @return the last index of the search CharSequence (always ≤ startPos), -1 * if no match or {@code null} string input * @since 2.0 * @since 3.0 Changed signature from lastIndexOf(String, String, int) to * lastIndexOf(CharSequence, CharSequence, int) */ public static int lastIndexOf(final CharSequence seq, final CharSequence searchSeq, final int startPos) { return CharSequenceUtils.lastIndexOf(seq, searchSeq, startPos); } /** * Returns the index within {@code seq} of the last occurrence of the specified * character. For values of {@code searchChar} in the range from 0 to 0xFFFF * (inclusive), the index (in Unicode code units) returned is the largest value * k such that:
* *
	 * this.charAt(k) == searchChar
	 * 
* *
is true. For other values of {@code searchChar}, it is the * largest value k such that:
* *
	 * this.codePointAt(k) == searchChar
	 * 
* *
is true. In either case, if no such character occurs in this * string, then {@code -1} is returned. Furthermore, a {@code null} or empty * ("") {@code CharSequence} will return {@code -1}. The {@code seq} * {@code CharSequence} object is searched backwards starting at the last * character. * *
	 * StringUtils.lastIndexOf(null, *)         = -1
	 * StringUtils.lastIndexOf("", *)           = -1
	 * StringUtils.lastIndexOf("aabaabaa", 'a') = 7
	 * StringUtils.lastIndexOf("aabaabaa", 'b') = 5
	 * 
* * @param seq the {@code CharSequence} to check, may be null * @param searchChar the character to find * @return the last index of the search character, -1 if no match or * {@code null} string input * @since 2.0 * @since 3.0 Changed signature from lastIndexOf(String, int) to * lastIndexOf(CharSequence, int) * @since 3.6 Updated {@link CharSequenceUtils} call to behave more like * {@code String} */ public static int lastIndexOf(final CharSequence seq, final int searchChar) { if (isEmpty(seq)) { return INDEX_NOT_FOUND; } return CharSequenceUtils.lastIndexOf(seq, searchChar, seq.length()); } /** * Returns the index within {@code seq} of the last occurrence of the specified * character, searching backward starting at the specified index. For values of * {@code searchChar} in the range from 0 to 0xFFFF (inclusive), the index * returned is the largest value k such that:
* *
	 * (this.charAt(k) == searchChar) && (k <= startPos)
	 * 
* *
is true. For other values of {@code searchChar}, it is the * largest value k such that:
* *
	 * (this.codePointAt(k) == searchChar) && (k <= startPos)
	 * 
* *
is true. In either case, if no such character occurs in * {@code seq} at or before position {@code startPos}, then {@code -1} is * returned. Furthermore, a {@code null} or empty ("") {@code CharSequence} will * return {@code -1}. A start position greater than the string length searches * the whole string. The search starts at the {@code startPos} and works * backwards; matches starting after the start position are ignored. * *

* All indices are specified in {@code char} values (Unicode code units). * *

	 * StringUtils.lastIndexOf(null, *, *)          = -1
	 * StringUtils.lastIndexOf("", *,  *)           = -1
	 * StringUtils.lastIndexOf("aabaabaa", 'b', 8)  = 5
	 * StringUtils.lastIndexOf("aabaabaa", 'b', 4)  = 2
	 * StringUtils.lastIndexOf("aabaabaa", 'b', 0)  = -1
	 * StringUtils.lastIndexOf("aabaabaa", 'b', 9)  = 5
	 * StringUtils.lastIndexOf("aabaabaa", 'b', -1) = -1
	 * StringUtils.lastIndexOf("aabaabaa", 'a', 0)  = 0
	 * 
* * @param seq the CharSequence to check, may be null * @param searchChar the character to find * @param startPos the start position * @return the last index of the search character (always ≤ startPos), -1 if * no match or {@code null} string input * @since 2.0 * @since 3.0 Changed signature from lastIndexOf(String, int, int) to * lastIndexOf(CharSequence, int, int) */ public static int lastIndexOf(final CharSequence seq, final int searchChar, final int startPos) { if (isEmpty(seq)) { return INDEX_NOT_FOUND; } return CharSequenceUtils.lastIndexOf(seq, searchChar, startPos); } /** *

* Find the latest index of any substring in a set of potential substrings. *

* *

* A {@code null} CharSequence will return {@code -1}. A {@code null} search * array will return {@code -1}. A {@code null} or zero length search array * entry will be ignored, but a search array containing "" will return the * length of {@code str} if {@code str} is not null. This method uses * {@link String#indexOf(String)} if possible *

* *
	 * StringUtils.lastIndexOfAny(null, *)                    = -1
	 * StringUtils.lastIndexOfAny(*, null)                    = -1
	 * StringUtils.lastIndexOfAny(*, [])                      = -1
	 * StringUtils.lastIndexOfAny(*, [null])                  = -1
	 * StringUtils.lastIndexOfAny("zzabyycdxx", ["ab", "cd"]) = 6
	 * StringUtils.lastIndexOfAny("zzabyycdxx", ["cd", "ab"]) = 6
	 * StringUtils.lastIndexOfAny("zzabyycdxx", ["mn", "op"]) = -1
	 * StringUtils.lastIndexOfAny("zzabyycdxx", ["mn", "op"]) = -1
	 * StringUtils.lastIndexOfAny("zzabyycdxx", ["mn", ""])   = 10
	 * 
* * @param str the CharSequence to check, may be null * @param searchStrs the CharSequences to search for, may be null * @return the last index of any of the CharSequences, -1 if no match * @since 3.0 Changed signature from lastIndexOfAny(String, String[]) to * lastIndexOfAny(CharSequence, CharSequence) */ public static int lastIndexOfAny(final CharSequence str, final CharSequence... searchStrs) { if (str == null || searchStrs == null) { return INDEX_NOT_FOUND; } int ret = INDEX_NOT_FOUND; int tmp = 0; for (final CharSequence search : searchStrs) { if (search == null) { continue; } tmp = CharSequenceUtils.lastIndexOf(str, search, str.length()); if (tmp > ret) { ret = tmp; } } return ret; } /** *

* Case in-sensitive find of the last index within a CharSequence. *

* *

* A {@code null} CharSequence will return {@code -1}. A negative start position * returns {@code -1}. An empty ("") search CharSequence always matches unless * the start position is negative. A start position greater than the string * length searches the whole string. *

* *
	 * StringUtils.lastIndexOfIgnoreCase(null, *)          = -1
	 * StringUtils.lastIndexOfIgnoreCase(*, null)          = -1
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "A")  = 7
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "B")  = 5
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "AB") = 4
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @return the first index of the search CharSequence, -1 if no match or * {@code null} string input * @since 2.5 * @since 3.0 Changed signature from lastIndexOfIgnoreCase(String, String) to * lastIndexOfIgnoreCase(CharSequence, CharSequence) */ public static int lastIndexOfIgnoreCase(final CharSequence str, final CharSequence searchStr) { if (str == null || searchStr == null) { return INDEX_NOT_FOUND; } return lastIndexOfIgnoreCase(str, searchStr, str.length()); } /** *

* Case in-sensitive find of the last index within a CharSequence from the * specified position. *

* *

* A {@code null} CharSequence will return {@code -1}. A negative start position * returns {@code -1}. An empty ("") search CharSequence always matches unless * the start position is negative. A start position greater than the string * length searches the whole string. The search starts at the startPos and works * backwards; matches starting after the start position are ignored. *

* *
	 * StringUtils.lastIndexOfIgnoreCase(null, *, *)          = -1
	 * StringUtils.lastIndexOfIgnoreCase(*, null, *)          = -1
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "A", 8)  = 7
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "B", 8)  = 5
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "AB", 8) = 4
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "B", 9)  = 5
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "B", -1) = -1
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "A", 0)  = 0
	 * StringUtils.lastIndexOfIgnoreCase("aabaabaa", "B", 0)  = -1
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @param startPos the start position * @return the last index of the search CharSequence (always ≤ startPos), -1 * if no match or {@code null} input * @since 2.5 * @since 3.0 Changed signature from lastIndexOfIgnoreCase(String, String, int) * to lastIndexOfIgnoreCase(CharSequence, CharSequence, int) */ public static int lastIndexOfIgnoreCase(final CharSequence str, final CharSequence searchStr, int startPos) { if (str == null || searchStr == null) { return INDEX_NOT_FOUND; } final int searchStrLength = searchStr.length(); final int strLength = str.length(); if (startPos > strLength - searchStrLength) { startPos = strLength - searchStrLength; } if (startPos < 0) { return INDEX_NOT_FOUND; } if (searchStrLength == 0) { return startPos; } for (int i = startPos; i >= 0; i--) { if (CharSequenceUtils.regionMatches(str, true, i, searchStr, 0, searchStrLength)) { return i; } } return INDEX_NOT_FOUND; } /** *

* Finds the n-th last index within a String, handling {@code null}. This method * uses {@link String#lastIndexOf(String)}. *

* *

* A {@code null} String will return {@code -1}. *

* *
	 * StringUtils.lastOrdinalIndexOf(null, *, *)          = -1
	 * StringUtils.lastOrdinalIndexOf(*, null, *)          = -1
	 * StringUtils.lastOrdinalIndexOf("", "", *)           = 0
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "a", 1)  = 7
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "a", 2)  = 6
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "b", 1)  = 5
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "b", 2)  = 2
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "ab", 1) = 4
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "ab", 2) = 1
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "", 1)   = 8
	 * StringUtils.lastOrdinalIndexOf("aabaabaa", "", 2)   = 8
	 * 
* *

* Note that 'tail(CharSequence str, int n)' may be implemented as: *

* *
	 * str.substring(lastOrdinalIndexOf(str, "\n", n) + 1)
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @param ordinal the n-th last {@code searchStr} to find * @return the n-th last index of the search CharSequence, {@code -1} * ({@code INDEX_NOT_FOUND}) if no match or {@code null} string input * @since 2.5 * @since 3.0 Changed signature from lastOrdinalIndexOf(String, String, int) to * lastOrdinalIndexOf(CharSequence, CharSequence, int) */ public static int lastOrdinalIndexOf(final CharSequence str, final CharSequence searchStr, final int ordinal) { return ordinalIndexOf(str, searchStr, ordinal, true); } /** *

* Gets the leftmost {@code len} characters of a String. *

* *

* If {@code len} characters are not available, or the String is {@code null}, * the String will be returned without an exception. An empty String is returned * if len is negative. *

* *
	 * StringUtils.left(null, *)    = null
	 * StringUtils.left(*, -ve)     = ""
	 * StringUtils.left("", *)      = ""
	 * StringUtils.left("abc", 0)   = ""
	 * StringUtils.left("abc", 2)   = "ab"
	 * StringUtils.left("abc", 4)   = "abc"
	 * 
* * @param str the String to get the leftmost characters from, may be null * @param len the length of the required String * @return the leftmost characters, {@code null} if null String input */ public static String left(final String str, final int len) { if (str == null) { return null; } if (len < 0) { return EMPTY; } if (str.length() <= len) { return str; } return str.substring(0, len); } /** *

* Left pad a String with spaces (' '). *

* *

* The String is padded to the size of {@code size}. *

* *
	 * StringUtils.leftPad(null, *)   = null
	 * StringUtils.leftPad("", 3)     = "   "
	 * StringUtils.leftPad("bat", 3)  = "bat"
	 * StringUtils.leftPad("bat", 5)  = "  bat"
	 * StringUtils.leftPad("bat", 1)  = "bat"
	 * StringUtils.leftPad("bat", -1) = "bat"
	 * 
* * @param str the String to pad out, may be null * @param size the size to pad to * @return left padded String or original String if no padding is necessary, * {@code null} if null String input */ public static String leftPad(final String str, final int size) { return leftPad(str, size, ' '); } /** *

* Left pad a String with a specified character. *

* *

* Pad to a size of {@code size}. *

* *
	 * StringUtils.leftPad(null, *, *)     = null
	 * StringUtils.leftPad("", 3, 'z')     = "zzz"
	 * StringUtils.leftPad("bat", 3, 'z')  = "bat"
	 * StringUtils.leftPad("bat", 5, 'z')  = "zzbat"
	 * StringUtils.leftPad("bat", 1, 'z')  = "bat"
	 * StringUtils.leftPad("bat", -1, 'z') = "bat"
	 * 
* * @param str the String to pad out, may be null * @param size the size to pad to * @param padChar the character to pad with * @return left padded String or original String if no padding is necessary, * {@code null} if null String input * @since 2.0 */ public static String leftPad(final String str, final int size, final char padChar) { if (str == null) { return null; } final int pads = size - str.length(); if (pads <= 0) { return str; // returns original String when possible } if (pads > PAD_LIMIT) { return leftPad(str, size, String.valueOf(padChar)); } return repeat(padChar, pads).concat(str); } /** *

* Left pad a String with a specified String. *

* *

* Pad to a size of {@code size}. *

* *
	 * StringUtils.leftPad(null, *, *)      = null
	 * StringUtils.leftPad("", 3, "z")      = "zzz"
	 * StringUtils.leftPad("bat", 3, "yz")  = "bat"
	 * StringUtils.leftPad("bat", 5, "yz")  = "yzbat"
	 * StringUtils.leftPad("bat", 8, "yz")  = "yzyzybat"
	 * StringUtils.leftPad("bat", 1, "yz")  = "bat"
	 * StringUtils.leftPad("bat", -1, "yz") = "bat"
	 * StringUtils.leftPad("bat", 5, null)  = "  bat"
	 * StringUtils.leftPad("bat", 5, "")    = "  bat"
	 * 
* * @param str the String to pad out, may be null * @param size the size to pad to * @param padStr the String to pad with, null or empty treated as single space * @return left padded String or original String if no padding is necessary, * {@code null} if null String input */ public static String leftPad(final String str, final int size, String padStr) { if (str == null) { return null; } if (isEmpty(padStr)) { padStr = SPACE; } final int padLen = padStr.length(); final int strLen = str.length(); final int pads = size - strLen; if (pads <= 0) { return str; // returns original String when possible } if (padLen == 1 && pads <= PAD_LIMIT) { return leftPad(str, size, padStr.charAt(0)); } if (pads == padLen) { return padStr.concat(str); } else if (pads < padLen) { return padStr.substring(0, pads).concat(str); } else { final char[] padding = new char[pads]; final char[] padChars = padStr.toCharArray(); for (int i = 0; i < pads; i++) { padding[i] = padChars[i % padLen]; } return new String(padding).concat(str); } } /** * Gets a CharSequence length or {@code 0} if the CharSequence is {@code null}. * * @param cs a CharSequence or {@code null} * @return CharSequence length or {@code 0} if the CharSequence is {@code null}. * @since 2.4 * @since 3.0 Changed signature from length(String) to length(CharSequence) */ public static int length(final CharSequence cs) { return cs == null ? 0 : cs.length(); } /** *

* Converts a String to lower case as per {@link String#toLowerCase()}. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.lowerCase(null)  = null
	 * StringUtils.lowerCase("")    = ""
	 * StringUtils.lowerCase("aBc") = "abc"
	 * 
* *

* Note: As described in the documentation for * {@link String#toLowerCase()}, the result of this method is affected by the * current locale. For platform-independent case transformations, the method * {@link #lowerCase(String, Locale)} should be used with a specific locale * (e.g. {@link Locale#ENGLISH}). *

* * @param str the String to lower case, may be null * @return the lower cased String, {@code null} if null String input */ public static String lowerCase(final String str) { if (str == null) { return null; } return str.toLowerCase(); } /** *

* Converts a String to lower case as per {@link String#toLowerCase(Locale)}. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.lowerCase(null, Locale.ENGLISH)  = null
	 * StringUtils.lowerCase("", Locale.ENGLISH)    = ""
	 * StringUtils.lowerCase("aBc", Locale.ENGLISH) = "abc"
	 * 
* * @param str the String to lower case, may be null * @param locale the locale that defines the case transformation rules, must not * be null * @return the lower cased String, {@code null} if null String input * @since 2.5 */ public static String lowerCase(final String str, final Locale locale) { if (str == null) { return null; } return str.toLowerCase(LocaleUtils.toLocale(locale)); } private static int[] matches(final CharSequence first, final CharSequence second) { final CharSequence max; final CharSequence min; if (first.length() > second.length()) { max = first; min = second; } else { max = second; min = first; } final int range = Math.max(max.length() / 2 - 1, 0); final int[] matchIndexes = new int[min.length()]; Arrays.fill(matchIndexes, -1); final boolean[] matchFlags = new boolean[max.length()]; int matches = 0; for (int mi = 0; mi < min.length(); mi++) { final char c1 = min.charAt(mi); for (int xi = Math.max(mi - range, 0), xn = Math.min(mi + range + 1, max.length()); xi < xn; xi++) { if (!matchFlags[xi] && c1 == max.charAt(xi)) { matchIndexes[mi] = xi; matchFlags[xi] = true; matches++; break; } } } final char[] ms1 = new char[matches]; final char[] ms2 = new char[matches]; for (int i = 0, si = 0; i < min.length(); i++) { if (matchIndexes[i] != -1) { ms1[si] = min.charAt(i); si++; } } for (int i = 0, si = 0; i < max.length(); i++) { if (matchFlags[i]) { ms2[si] = max.charAt(i); si++; } } int transpositions = 0; for (int mi = 0; mi < ms1.length; mi++) { if (ms1[mi] != ms2[mi]) { transpositions++; } } int prefix = 0; for (int mi = 0; mi < min.length(); mi++) { if (first.charAt(mi) == second.charAt(mi)) { prefix++; } else { break; } } return new int[] { matches, transpositions / 2, prefix, max.length() }; } /** *

* Gets {@code len} characters from the middle of a String. *

* *

* If {@code len} characters are not available, the remainder of the String will * be returned without an exception. If the String is {@code null}, {@code null} * will be returned. An empty String is returned if len is negative or exceeds * the length of {@code str}. *

* *
	 * StringUtils.mid(null, *, *)    = null
	 * StringUtils.mid(*, *, -ve)     = ""
	 * StringUtils.mid("", 0, *)      = ""
	 * StringUtils.mid("abc", 0, 2)   = "ab"
	 * StringUtils.mid("abc", 0, 4)   = "abc"
	 * StringUtils.mid("abc", 2, 4)   = "c"
	 * StringUtils.mid("abc", 4, 2)   = ""
	 * StringUtils.mid("abc", -2, 2)  = "ab"
	 * 
* * @param str the String to get the characters from, may be null * @param pos the position to start from, negative treated as zero * @param len the length of the required String * @return the middle characters, {@code null} if null String input */ public static String mid(final String str, int pos, final int len) { if (str == null) { return null; } if (len < 0 || pos > str.length()) { return EMPTY; } if (pos < 0) { pos = 0; } if (str.length() <= pos + len) { return str.substring(pos); } return str.substring(pos, pos + len); } private static StringJoiner newStringJoiner(final char delimiter) { return new StringJoiner(String.valueOf(delimiter)); } /** *

* Similar to http://www.w3.org/TR/xpath/#function-normalize * -space *

*

* The function returns the argument string with whitespace normalized by using * {@code {@link #trim(String)}} to remove leading and trailing whitespace and * then replacing sequences of whitespace characters by a single space. *

* In XML Whitespace characters are the same as those allowed by the * S production, which is S ::= * (#x20 | #x9 | #xD | #xA)+ *

* Java's regexp pattern \s defines whitespace as [ \t\n\x0B\f\r] * *

* For reference: *

*
    *
  • \x0B = vertical tab
  • *
  • \f = #xC = form feed
  • *
  • #x20 = space
  • *
  • #x9 = \t
  • *
  • #xA = \n
  • *
  • #xD = \r
  • *
* *

* The difference is that Java's whitespace includes vertical tab and form feed, * which this functional will also normalize. Additionally {@code {@link * #trim(String)}} removes control characters (char <= 32) from both ends of * this String. *

* * @see Pattern * @see #trim(String) * @see http://www.w3.org/TR/xpath/#function-normalize-space * @param str the source String to normalize whitespaces from, may be null * @return the modified string with whitespace normalized, {@code null} if null * String input * * @since 3.0 */ public static String normalizeSpace(final String str) { // LANG-1020: Improved performance significantly by normalizing manually instead // of using regex // See https://github.com/librucha/commons-lang-normalizespaces-benchmark for // performance test if (isEmpty(str)) { return str; } final int size = str.length(); final char[] newChars = new char[size]; int count = 0; int whitespacesCount = 0; boolean startWhitespaces = true; for (int i = 0; i < size; i++) { final char actualChar = str.charAt(i); final boolean isWhitespace = Character.isWhitespace(actualChar); if (isWhitespace) { if (whitespacesCount == 0 && !startWhitespaces) { newChars[count++] = SPACE.charAt(0); } whitespacesCount++; } else { startWhitespaces = false; newChars[count++] = (actualChar == 160 ? 32 : actualChar); whitespacesCount = 0; } } if (startWhitespaces) { return EMPTY; } return new String(newChars, 0, count - (whitespacesCount > 0 ? 1 : 0)).trim(); } /** *

* Finds the n-th index within a CharSequence, handling {@code null}. This * method uses {@link String#indexOf(String)} if possible. *

*

* Note: The code starts looking for a match at the start of the target, * incrementing the starting index by one after each successful match (unless * {@code searchStr} is an empty string in which case the position is never * incremented and {@code 0} is returned immediately). This means that matches * may overlap. *

*

* A {@code null} CharSequence will return {@code -1}. *

* *
	 * StringUtils.ordinalIndexOf(null, *, *)          = -1
	 * StringUtils.ordinalIndexOf(*, null, *)          = -1
	 * StringUtils.ordinalIndexOf("", "", *)           = 0
	 * StringUtils.ordinalIndexOf("aabaabaa", "a", 1)  = 0
	 * StringUtils.ordinalIndexOf("aabaabaa", "a", 2)  = 1
	 * StringUtils.ordinalIndexOf("aabaabaa", "b", 1)  = 2
	 * StringUtils.ordinalIndexOf("aabaabaa", "b", 2)  = 5
	 * StringUtils.ordinalIndexOf("aabaabaa", "ab", 1) = 1
	 * StringUtils.ordinalIndexOf("aabaabaa", "ab", 2) = 4
	 * StringUtils.ordinalIndexOf("aabaabaa", "", 1)   = 0
	 * StringUtils.ordinalIndexOf("aabaabaa", "", 2)   = 0
	 * 
* *

* Matches may overlap: *

* *
	 * StringUtils.ordinalIndexOf("ababab", "aba", 1)   = 0
	 * StringUtils.ordinalIndexOf("ababab", "aba", 2)   = 2
	 * StringUtils.ordinalIndexOf("ababab", "aba", 3)   = -1
	 *
	 * StringUtils.ordinalIndexOf("abababab", "abab", 1) = 0
	 * StringUtils.ordinalIndexOf("abababab", "abab", 2) = 2
	 * StringUtils.ordinalIndexOf("abababab", "abab", 3) = 4
	 * StringUtils.ordinalIndexOf("abababab", "abab", 4) = -1
	 * 
* *

* Note that 'head(CharSequence str, int n)' may be implemented as: *

* *
	 * str.substring(0, lastOrdinalIndexOf(str, "\n", n))
	 * 
* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @param ordinal the n-th {@code searchStr} to find * @return the n-th index of the search CharSequence, {@code -1} * ({@code INDEX_NOT_FOUND}) if no match or {@code null} string input * @since 2.1 * @since 3.0 Changed signature from ordinalIndexOf(String, String, int) to * ordinalIndexOf(CharSequence, CharSequence, int) */ public static int ordinalIndexOf(final CharSequence str, final CharSequence searchStr, final int ordinal) { return ordinalIndexOf(str, searchStr, ordinal, false); } /** *

* Finds the n-th index within a String, handling {@code null}. This method uses * {@link String#indexOf(String)} if possible. *

*

* Note that matches may overlap *

* *

* A {@code null} CharSequence will return {@code -1}. *

* * @param str the CharSequence to check, may be null * @param searchStr the CharSequence to find, may be null * @param ordinal the n-th {@code searchStr} to find, overlapping matches are * allowed. * @param lastIndex true if lastOrdinalIndexOf() otherwise false if * ordinalIndexOf() * @return the n-th index of the search CharSequence, {@code -1} * ({@code INDEX_NOT_FOUND}) if no match or {@code null} string input */ // Shared code between ordinalIndexOf(String, String, int) and // lastOrdinalIndexOf(String, String, int) private static int ordinalIndexOf(final CharSequence str, final CharSequence searchStr, final int ordinal, final boolean lastIndex) { if (str == null || searchStr == null || ordinal <= 0) { return INDEX_NOT_FOUND; } if (searchStr.length() == 0) { return lastIndex ? str.length() : 0; } int found = 0; // set the initial index beyond the end of the string // this is to allow for the initial index decrement/increment int index = lastIndex ? str.length() : INDEX_NOT_FOUND; do { if (lastIndex) { index = CharSequenceUtils.lastIndexOf(str, searchStr, index - 1); // step backwards thru string } else { index = CharSequenceUtils.indexOf(str, searchStr, index + 1); // step forwards through string } if (index < 0) { return index; } found++; } while (found < ordinal); return index; } /** *

* Overlays part of a String with another String. *

* *

* A {@code null} string input returns {@code null}. A negative index is treated * as zero. An index greater than the string length is treated as the string * length. The start index is always the smaller of the two indices. *

* *
	 * StringUtils.overlay(null, *, *, *)            = null
	 * StringUtils.overlay("", "abc", 0, 0)          = "abc"
	 * StringUtils.overlay("abcdef", null, 2, 4)     = "abef"
	 * StringUtils.overlay("abcdef", "", 2, 4)       = "abef"
	 * StringUtils.overlay("abcdef", "", 4, 2)       = "abef"
	 * StringUtils.overlay("abcdef", "zzzz", 2, 4)   = "abzzzzef"
	 * StringUtils.overlay("abcdef", "zzzz", 4, 2)   = "abzzzzef"
	 * StringUtils.overlay("abcdef", "zzzz", -1, 4)  = "zzzzef"
	 * StringUtils.overlay("abcdef", "zzzz", 2, 8)   = "abzzzz"
	 * StringUtils.overlay("abcdef", "zzzz", -2, -3) = "zzzzabcdef"
	 * StringUtils.overlay("abcdef", "zzzz", 8, 10)  = "abcdefzzzz"
	 * 
* * @param str the String to do overlaying in, may be null * @param overlay the String to overlay, may be null * @param start the position to start overlaying at * @param end the position to stop overlaying before * @return overlayed String, {@code null} if null String input * @since 2.0 */ public static String overlay(final String str, String overlay, int start, int end) { if (str == null) { return null; } if (overlay == null) { overlay = EMPTY; } final int len = str.length(); if (start < 0) { start = 0; } if (start > len) { start = len; } if (end < 0) { end = 0; } if (end > len) { end = len; } if (start > end) { final int temp = start; start = end; end = temp; } return str.substring(0, start) + overlay + str.substring(end); } /** * Prepends the prefix to the start of the string if the string does not already * start with any of the prefixes. * * @param str The string. * @param prefix The prefix to prepend to the start of the string. * @param ignoreCase Indicates whether the compare should ignore case. * @param prefixes Additional prefixes that are valid (optional). * * @return A new String if prefix was prepended, the same string otherwise. */ private static String prependIfMissing(final String str, final CharSequence prefix, final boolean ignoreCase, final CharSequence... prefixes) { if (str == null || isEmpty(prefix) || startsWith(str, prefix, ignoreCase)) { return str; } if (prefixes.length > 0) { for (final CharSequence p : prefixes) { if (startsWith(str, p, ignoreCase)) { return str; } } } return prefix.toString() + str; } /** * Prepends the prefix to the start of the string if the string does not already * start with any of the prefixes. * *
	 * StringUtils.prependIfMissing(null, null) = null
	 * StringUtils.prependIfMissing("abc", null) = "abc"
	 * StringUtils.prependIfMissing("", "xyz") = "xyz"
	 * StringUtils.prependIfMissing("abc", "xyz") = "xyzabc"
	 * StringUtils.prependIfMissing("xyzabc", "xyz") = "xyzabc"
	 * StringUtils.prependIfMissing("XYZabc", "xyz") = "xyzXYZabc"
	 * 
*

* With additional prefixes, *

* *
	 * StringUtils.prependIfMissing(null, null, null) = null
	 * StringUtils.prependIfMissing("abc", null, null) = "abc"
	 * StringUtils.prependIfMissing("", "xyz", null) = "xyz"
	 * StringUtils.prependIfMissing("abc", "xyz", new CharSequence[]{null}) = "xyzabc"
	 * StringUtils.prependIfMissing("abc", "xyz", "") = "abc"
	 * StringUtils.prependIfMissing("abc", "xyz", "mno") = "xyzabc"
	 * StringUtils.prependIfMissing("xyzabc", "xyz", "mno") = "xyzabc"
	 * StringUtils.prependIfMissing("mnoabc", "xyz", "mno") = "mnoabc"
	 * StringUtils.prependIfMissing("XYZabc", "xyz", "mno") = "xyzXYZabc"
	 * StringUtils.prependIfMissing("MNOabc", "xyz", "mno") = "xyzMNOabc"
	 * 
* * @param str The string. * @param prefix The prefix to prepend to the start of the string. * @param prefixes Additional prefixes that are valid. * * @return A new String if prefix was prepended, the same string otherwise. * * @since 3.2 */ public static String prependIfMissing(final String str, final CharSequence prefix, final CharSequence... prefixes) { return prependIfMissing(str, prefix, false, prefixes); } /** * Prepends the prefix to the start of the string if the string does not already * start, case insensitive, with any of the prefixes. * *
	 * StringUtils.prependIfMissingIgnoreCase(null, null) = null
	 * StringUtils.prependIfMissingIgnoreCase("abc", null) = "abc"
	 * StringUtils.prependIfMissingIgnoreCase("", "xyz") = "xyz"
	 * StringUtils.prependIfMissingIgnoreCase("abc", "xyz") = "xyzabc"
	 * StringUtils.prependIfMissingIgnoreCase("xyzabc", "xyz") = "xyzabc"
	 * StringUtils.prependIfMissingIgnoreCase("XYZabc", "xyz") = "XYZabc"
	 * 
*

* With additional prefixes, *

* *
	 * StringUtils.prependIfMissingIgnoreCase(null, null, null) = null
	 * StringUtils.prependIfMissingIgnoreCase("abc", null, null) = "abc"
	 * StringUtils.prependIfMissingIgnoreCase("", "xyz", null) = "xyz"
	 * StringUtils.prependIfMissingIgnoreCase("abc", "xyz", new CharSequence[]{null}) = "xyzabc"
	 * StringUtils.prependIfMissingIgnoreCase("abc", "xyz", "") = "abc"
	 * StringUtils.prependIfMissingIgnoreCase("abc", "xyz", "mno") = "xyzabc"
	 * StringUtils.prependIfMissingIgnoreCase("xyzabc", "xyz", "mno") = "xyzabc"
	 * StringUtils.prependIfMissingIgnoreCase("mnoabc", "xyz", "mno") = "mnoabc"
	 * StringUtils.prependIfMissingIgnoreCase("XYZabc", "xyz", "mno") = "XYZabc"
	 * StringUtils.prependIfMissingIgnoreCase("MNOabc", "xyz", "mno") = "MNOabc"
	 * 
* * @param str The string. * @param prefix The prefix to prepend to the start of the string. * @param prefixes Additional prefixes that are valid (optional). * * @return A new String if prefix was prepended, the same string otherwise. * * @since 3.2 */ public static String prependIfMissingIgnoreCase(final String str, final CharSequence prefix, final CharSequence... prefixes) { return prependIfMissing(str, prefix, true, prefixes); } /** *

* Removes all occurrences of a character from within the source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. *

* *
	 * StringUtils.remove(null, *)       = null
	 * StringUtils.remove("", *)         = ""
	 * StringUtils.remove("queued", 'u') = "qeed"
	 * StringUtils.remove("queued", 'z') = "queued"
	 * 
* * @param str the source String to search, may be null * @param remove the char to search for and remove, may be null * @return the substring with the char removed if found, {@code null} if null * String input * @since 2.1 */ public static String remove(final String str, final char remove) { if (isEmpty(str) || str.indexOf(remove) == INDEX_NOT_FOUND) { return str; } final char[] chars = str.toCharArray(); int pos = 0; for (int i = 0; i < chars.length; i++) { if (chars[i] != remove) { chars[pos++] = chars[i]; } } return new String(chars, 0, pos); } /** *

* Removes all occurrences of a substring from within the source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. A {@code null} remove string will return * the source string. An empty ("") remove string will return the source string. *

* *
	 * StringUtils.remove(null, *)        = null
	 * StringUtils.remove("", *)          = ""
	 * StringUtils.remove(*, null)        = *
	 * StringUtils.remove(*, "")          = *
	 * StringUtils.remove("queued", "ue") = "qd"
	 * StringUtils.remove("queued", "zz") = "queued"
	 * 
* * @param str the source String to search, may be null * @param remove the String to search for and remove, may be null * @return the substring with the string removed if found, {@code null} if null * String input * @since 2.1 */ public static String remove(final String str, final String remove) { if (isEmpty(str) || isEmpty(remove)) { return str; } return replace(str, remove, EMPTY, -1); } /** *

* Removes each substring of the text String that matches the given regular * expression. *

* * This method is a {@code null} safe equivalent to: *
    *
  • {@code text.replaceAll(regex, StringUtils.EMPTY)}
  • *
  • {@code Pattern.compile(regex).matcher(text).replaceAll(StringUtils.EMPTY)}
  • *
* *

* A {@code null} reference passed to this method is a no-op. *

* *

* Unlike in the {@link #removePattern(String, String)} method, the * {@link Pattern#DOTALL} option is NOT automatically added. To use the DOTALL * option prepend {@code "(?s)"} to the regex. DOTALL is also known as * single-line mode in Perl. *

* *
	 * StringUtils.removeAll(null, *)      = null
	 * StringUtils.removeAll("any", (String) null)  = "any"
	 * StringUtils.removeAll("any", "")    = "any"
	 * StringUtils.removeAll("any", ".*")  = ""
	 * StringUtils.removeAll("any", ".+")  = ""
	 * StringUtils.removeAll("abc", ".?")  = ""
	 * StringUtils.removeAll("A<__>\n<__>B", "<.*>")      = "A\nB"
	 * StringUtils.removeAll("A<__>\n<__>B", "(?s)<.*>")  = "AB"
	 * StringUtils.removeAll("ABCabc123abc", "[a-z]")     = "ABC123"
	 * 
* * @param text text to remove from, may be null * @param regex the regular expression to which this string is to be matched * @return the text with any removes processed, {@code null} if null String * input * * @throws java.util.regex.PatternSyntaxException if the regular expression's * syntax is invalid * * @see #replaceAll(String, String, String) * @see #removePattern(String, String) * @see String#replaceAll(String, String) * @see java.util.regex.Pattern * @see java.util.regex.Pattern#DOTALL * @since 3.5 * * @deprecated Moved to RegExUtils. */ @Deprecated public static String removeAll(final String text, final String regex) { return RegExUtils.removeAll(text, regex); } /** *

* Removes a substring only if it is at the end of a source string, otherwise * returns the source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. A {@code null} search string will return * the source string. *

* *
	 * StringUtils.removeEnd(null, *)      = null
	 * StringUtils.removeEnd("", *)        = ""
	 * StringUtils.removeEnd(*, null)      = *
	 * StringUtils.removeEnd("www.domain.com", ".com.")  = "www.domain.com"
	 * StringUtils.removeEnd("www.domain.com", ".com")   = "www.domain"
	 * StringUtils.removeEnd("www.domain.com", "domain") = "www.domain.com"
	 * StringUtils.removeEnd("abc", "")    = "abc"
	 * 
* * @param str the source String to search, may be null * @param remove the String to search for and remove, may be null * @return the substring with the string removed if found, {@code null} if null * String input * @since 2.1 */ public static String removeEnd(final String str, final String remove) { if (isEmpty(str) || isEmpty(remove)) { return str; } if (str.endsWith(remove)) { return str.substring(0, str.length() - remove.length()); } return str; } /** *

* Case insensitive removal of a substring if it is at the end of a source * string, otherwise returns the source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. A {@code null} search string will return * the source string. *

* *
	 * StringUtils.removeEndIgnoreCase(null, *)      = null
	 * StringUtils.removeEndIgnoreCase("", *)        = ""
	 * StringUtils.removeEndIgnoreCase(*, null)      = *
	 * StringUtils.removeEndIgnoreCase("www.domain.com", ".com.")  = "www.domain.com"
	 * StringUtils.removeEndIgnoreCase("www.domain.com", ".com")   = "www.domain"
	 * StringUtils.removeEndIgnoreCase("www.domain.com", "domain") = "www.domain.com"
	 * StringUtils.removeEndIgnoreCase("abc", "")    = "abc"
	 * StringUtils.removeEndIgnoreCase("www.domain.com", ".COM") = "www.domain")
	 * StringUtils.removeEndIgnoreCase("www.domain.COM", ".com") = "www.domain")
	 * 
* * @param str the source String to search, may be null * @param remove the String to search for (case insensitive) and remove, may be * null * @return the substring with the string removed if found, {@code null} if null * String input * @since 2.4 */ public static String removeEndIgnoreCase(final String str, final String remove) { if (isEmpty(str) || isEmpty(remove)) { return str; } if (endsWithIgnoreCase(str, remove)) { return str.substring(0, str.length() - remove.length()); } return str; } /** *

* Removes the first substring of the text string that matches the given regular * expression. *

* * This method is a {@code null} safe equivalent to: *
    *
  • {@code text.replaceFirst(regex, StringUtils.EMPTY)}
  • *
  • {@code Pattern.compile(regex).matcher(text).replaceFirst(StringUtils.EMPTY)}
  • *
* *

* A {@code null} reference passed to this method is a no-op. *

* *

* The {@link Pattern#DOTALL} option is NOT automatically added. To use the * DOTALL option prepend {@code "(?s)"} to the regex. DOTALL is also known as * single-line mode in Perl. *

* *
	 * StringUtils.removeFirst(null, *)      = null
	 * StringUtils.removeFirst("any", (String) null)  = "any"
	 * StringUtils.removeFirst("any", "")    = "any"
	 * StringUtils.removeFirst("any", ".*")  = ""
	 * StringUtils.removeFirst("any", ".+")  = ""
	 * StringUtils.removeFirst("abc", ".?")  = "bc"
	 * StringUtils.removeFirst("A<__>\n<__>B", "<.*>")      = "A\n<__>B"
	 * StringUtils.removeFirst("A<__>\n<__>B", "(?s)<.*>")  = "AB"
	 * StringUtils.removeFirst("ABCabc123", "[a-z]")          = "ABCbc123"
	 * StringUtils.removeFirst("ABCabc123abc", "[a-z]+")      = "ABC123abc"
	 * 
* * @param text text to remove from, may be null * @param regex the regular expression to which this string is to be matched * @return the text with the first replacement processed, {@code null} if null * String input * * @throws java.util.regex.PatternSyntaxException if the regular expression's * syntax is invalid * * @see #replaceFirst(String, String, String) * @see String#replaceFirst(String, String) * @see java.util.regex.Pattern * @see java.util.regex.Pattern#DOTALL * @since 3.5 * * @deprecated Moved to RegExUtils. */ @Deprecated public static String removeFirst(final String text, final String regex) { return replaceFirst(text, regex, EMPTY); } /** *

* Case insensitive removal of all occurrences of a substring from within the * source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. A {@code null} remove string will return * the source string. An empty ("") remove string will return the source string. *

* *
	 * StringUtils.removeIgnoreCase(null, *)        = null
	 * StringUtils.removeIgnoreCase("", *)          = ""
	 * StringUtils.removeIgnoreCase(*, null)        = *
	 * StringUtils.removeIgnoreCase(*, "")          = *
	 * StringUtils.removeIgnoreCase("queued", "ue") = "qd"
	 * StringUtils.removeIgnoreCase("queued", "zz") = "queued"
	 * StringUtils.removeIgnoreCase("quEUed", "UE") = "qd"
	 * StringUtils.removeIgnoreCase("queued", "zZ") = "queued"
	 * 
* * @param str the source String to search, may be null * @param remove the String to search for (case insensitive) and remove, may be * null * @return the substring with the string removed if found, {@code null} if null * String input * @since 3.5 */ public static String removeIgnoreCase(final String str, final String remove) { return replaceIgnoreCase(str, remove, EMPTY, -1); } /** *

* Removes each substring of the source String that matches the given regular * expression using the DOTALL option. *

* * This call is a {@code null} safe equivalent to: *
    *
  • {@code source.replaceAll("(?s)" + regex, StringUtils.EMPTY)}
  • *
  • {@code Pattern.compile(regex, Pattern.DOTALL).matcher(source).replaceAll(StringUtils.EMPTY)}
  • *
* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.removePattern(null, *)       = null
	 * StringUtils.removePattern("any", (String) null)   = "any"
	 * StringUtils.removePattern("A<__>\n<__>B", "<.*>")  = "AB"
	 * StringUtils.removePattern("ABCabc123", "[a-z]")    = "ABC123"
	 * 
* * @param source the source string * @param regex the regular expression to which this string is to be matched * @return The resulting {@code String} * @see #replacePattern(String, String, String) * @see String#replaceAll(String, String) * @see Pattern#DOTALL * @since 3.2 * @since 3.5 Changed {@code null} reference passed to this method is a no-op. * * @deprecated Moved to RegExUtils. */ @Deprecated public static String removePattern(final String source, final String regex) { return RegExUtils.removePattern(source, regex); } /** *

* Removes a substring only if it is at the beginning of a source string, * otherwise returns the source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. A {@code null} search string will return * the source string. *

* *
	 * StringUtils.removeStart(null, *)      = null
	 * StringUtils.removeStart("", *)        = ""
	 * StringUtils.removeStart(*, null)      = *
	 * StringUtils.removeStart("www.domain.com", "www.")   = "domain.com"
	 * StringUtils.removeStart("domain.com", "www.")       = "domain.com"
	 * StringUtils.removeStart("www.domain.com", "domain") = "www.domain.com"
	 * StringUtils.removeStart("abc", "")    = "abc"
	 * 
* * @param str the source String to search, may be null * @param remove the String to search for and remove, may be null * @return the substring with the string removed if found, {@code null} if null * String input * @since 2.1 */ public static String removeStart(final String str, final String remove) { if (isEmpty(str) || isEmpty(remove)) { return str; } if (str.startsWith(remove)) { return str.substring(remove.length()); } return str; } /** *

* Case insensitive removal of a substring if it is at the beginning of a source * string, otherwise returns the source string. *

* *

* A {@code null} source string will return {@code null}. An empty ("") source * string will return the empty string. A {@code null} search string will return * the source string. *

* *
	 * StringUtils.removeStartIgnoreCase(null, *)      = null
	 * StringUtils.removeStartIgnoreCase("", *)        = ""
	 * StringUtils.removeStartIgnoreCase(*, null)      = *
	 * StringUtils.removeStartIgnoreCase("www.domain.com", "www.")   = "domain.com"
	 * StringUtils.removeStartIgnoreCase("www.domain.com", "WWW.")   = "domain.com"
	 * StringUtils.removeStartIgnoreCase("domain.com", "www.")       = "domain.com"
	 * StringUtils.removeStartIgnoreCase("www.domain.com", "domain") = "www.domain.com"
	 * StringUtils.removeStartIgnoreCase("abc", "")    = "abc"
	 * 
* * @param str the source String to search, may be null * @param remove the String to search for (case insensitive) and remove, may be * null * @return the substring with the string removed if found, {@code null} if null * String input * @since 2.4 */ public static String removeStartIgnoreCase(final String str, final String remove) { if (str != null && startsWithIgnoreCase(str, remove)) { return str.substring(length(remove)); } return str; } /** *

* Returns padding using the specified delimiter repeated to a given length. *

* *
	 * StringUtils.repeat('e', 0)  = ""
	 * StringUtils.repeat('e', 3)  = "eee"
	 * StringUtils.repeat('e', -2) = ""
	 * 
* *

* Note: this method does not support padding with * Unicode * Supplementary Characters as they require a pair of {@code char}s to be * represented. If you are needing to support full I18N of your applications * consider using {@link #repeat(String, int)} instead. *

* * @param ch character to repeat * @param repeat number of times to repeat char, negative treated as zero * @return String with repeated character * @see #repeat(String, int) */ public static String repeat(final char ch, final int repeat) { if (repeat <= 0) { return EMPTY; } final char[] buf = new char[repeat]; Arrays.fill(buf, ch); return new String(buf); } /** *

* Repeat a String {@code repeat} times to form a new String. *

* *
	 * StringUtils.repeat(null, 2) = null
	 * StringUtils.repeat("", 0)   = ""
	 * StringUtils.repeat("", 2)   = ""
	 * StringUtils.repeat("a", 3)  = "aaa"
	 * StringUtils.repeat("ab", 2) = "abab"
	 * StringUtils.repeat("a", -2) = ""
	 * 
* * @param str the String to repeat, may be null * @param repeat number of times to repeat str, negative treated as zero * @return a new String consisting of the original String repeated, {@code null} * if null String input */ public static String repeat(final String str, final int repeat) { // Performance tuned for 2.0 (JDK1.4) if (str == null) { return null; } if (repeat <= 0) { return EMPTY; } final int inputLength = str.length(); if (repeat == 1 || inputLength == 0) { return str; } if (inputLength == 1 && repeat <= PAD_LIMIT) { return repeat(str.charAt(0), repeat); } final int outputLength = inputLength * repeat; switch (inputLength) { case 1: return repeat(str.charAt(0), repeat); case 2: final char ch0 = str.charAt(0); final char ch1 = str.charAt(1); final char[] output2 = new char[outputLength]; for (int i = repeat * 2 - 2; i >= 0; i--, i--) { output2[i] = ch0; output2[i + 1] = ch1; } return new String(output2); default: final StringBuilder buf = new StringBuilder(outputLength); for (int i = 0; i < repeat; i++) { buf.append(str); } return buf.toString(); } } /** *

* Repeat a String {@code repeat} times to form a new String, with a String * separator injected each time. *

* *
	 * StringUtils.repeat(null, null, 2) = null
	 * StringUtils.repeat(null, "x", 2)  = null
	 * StringUtils.repeat("", null, 0)   = ""
	 * StringUtils.repeat("", "", 2)     = ""
	 * StringUtils.repeat("", "x", 3)    = "xxx"
	 * StringUtils.repeat("?", ", ", 3)  = "?, ?, ?"
	 * 
* * @param str the String to repeat, may be null * @param separator the String to inject, may be null * @param repeat number of times to repeat str, negative treated as zero * @return a new String consisting of the original String repeated, {@code null} * if null String input * @since 2.5 */ public static String repeat(final String str, final String separator, final int repeat) { if (str == null || separator == null) { return repeat(str, repeat); } // given that repeat(String, int) is quite optimized, better to rely on it than // try and splice this into it final String result = repeat(str + separator, repeat); return removeEnd(result, separator); } /** *

* Replaces all occurrences of a String within another String. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replace(null, *, *)        = null
	 * StringUtils.replace("", *, *)          = ""
	 * StringUtils.replace("any", null, *)    = "any"
	 * StringUtils.replace("any", *, null)    = "any"
	 * StringUtils.replace("any", "", *)      = "any"
	 * StringUtils.replace("aba", "a", null)  = "aba"
	 * StringUtils.replace("aba", "a", "")    = "b"
	 * StringUtils.replace("aba", "a", "z")   = "zbz"
	 * 
* * @see #replace(String text, String searchString, String replacement, int max) * @param text text to search and replace in, may be null * @param searchString the String to search for, may be null * @param replacement the String to replace it with, may be null * @return the text with any replacements processed, {@code null} if null String * input */ public static String replace(final String text, final String searchString, final String replacement) { return replace(text, searchString, replacement, -1); } /** *

* Replaces a String with another String inside a larger String, for the first * {@code max} values of the search String. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replace(null, *, *, *)         = null
	 * StringUtils.replace("", *, *, *)           = ""
	 * StringUtils.replace("any", null, *, *)     = "any"
	 * StringUtils.replace("any", *, null, *)     = "any"
	 * StringUtils.replace("any", "", *, *)       = "any"
	 * StringUtils.replace("any", *, *, 0)        = "any"
	 * StringUtils.replace("abaa", "a", null, -1) = "abaa"
	 * StringUtils.replace("abaa", "a", "", -1)   = "b"
	 * StringUtils.replace("abaa", "a", "z", 0)   = "abaa"
	 * StringUtils.replace("abaa", "a", "z", 1)   = "zbaa"
	 * StringUtils.replace("abaa", "a", "z", 2)   = "zbza"
	 * StringUtils.replace("abaa", "a", "z", -1)  = "zbzz"
	 * 
* * @param text text to search and replace in, may be null * @param searchString the String to search for, may be null * @param replacement the String to replace it with, may be null * @param max maximum number of values to replace, or {@code -1} if no * maximum * @return the text with any replacements processed, {@code null} if null String * input */ public static String replace(final String text, final String searchString, final String replacement, final int max) { return replace(text, searchString, replacement, max, false); } /** *

* Replaces a String with another String inside a larger String, for the first * {@code max} values of the search String, case sensitively/insensitively based * on {@code ignoreCase} value. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replace(null, *, *, *, false)         = null
	 * StringUtils.replace("", *, *, *, false)           = ""
	 * StringUtils.replace("any", null, *, *, false)     = "any"
	 * StringUtils.replace("any", *, null, *, false)     = "any"
	 * StringUtils.replace("any", "", *, *, false)       = "any"
	 * StringUtils.replace("any", *, *, 0, false)        = "any"
	 * StringUtils.replace("abaa", "a", null, -1, false) = "abaa"
	 * StringUtils.replace("abaa", "a", "", -1, false)   = "b"
	 * StringUtils.replace("abaa", "a", "z", 0, false)   = "abaa"
	 * StringUtils.replace("abaa", "A", "z", 1, false)   = "abaa"
	 * StringUtils.replace("abaa", "A", "z", 1, true)   = "zbaa"
	 * StringUtils.replace("abAa", "a", "z", 2, true)   = "zbza"
	 * StringUtils.replace("abAa", "a", "z", -1, true)  = "zbzz"
	 * 
* * @param text text to search and replace in, may be null * @param searchString the String to search for (case insensitive), may be null * @param replacement the String to replace it with, may be null * @param max maximum number of values to replace, or {@code -1} if no * maximum * @param ignoreCase if true replace is case insensitive, otherwise case * sensitive * @return the text with any replacements processed, {@code null} if null String * input */ private static String replace(final String text, String searchString, final String replacement, int max, final boolean ignoreCase) { if (isEmpty(text) || isEmpty(searchString) || replacement == null || max == 0) { return text; } if (ignoreCase) { searchString = searchString.toLowerCase(); } int start = 0; int end = ignoreCase ? indexOfIgnoreCase(text, searchString, start) : indexOf(text, searchString, start); if (end == INDEX_NOT_FOUND) { return text; } final int replLength = searchString.length(); int increase = Math.max(replacement.length() - replLength, 0); increase *= max < 0 ? 16 : Math.min(max, 64); final StringBuilder buf = new StringBuilder(text.length() + increase); while (end != INDEX_NOT_FOUND) { buf.append(text, start, end).append(replacement); start = end + replLength; if (--max == 0) { break; } end = ignoreCase ? indexOfIgnoreCase(text, searchString, start) : indexOf(text, searchString, start); } buf.append(text, start, text.length()); return buf.toString(); } /** *

* Replaces each substring of the text String that matches the given regular * expression with the given replacement. *

* * This method is a {@code null} safe equivalent to: *
    *
  • {@code text.replaceAll(regex, replacement)}
  • *
  • {@code Pattern.compile(regex).matcher(text).replaceAll(replacement)}
  • *
* *

* A {@code null} reference passed to this method is a no-op. *

* *

* Unlike in the {@link #replacePattern(String, String, String)} method, the * {@link Pattern#DOTALL} option is NOT automatically added. To use the DOTALL * option prepend {@code "(?s)"} to the regex. DOTALL is also known as * single-line mode in Perl. *

* *
	 * StringUtils.replaceAll(null, *, *)       = null
	 * StringUtils.replaceAll("any", (String) null, *)   = "any"
	 * StringUtils.replaceAll("any", *, null)   = "any"
	 * StringUtils.replaceAll("", "", "zzz")    = "zzz"
	 * StringUtils.replaceAll("", ".*", "zzz")  = "zzz"
	 * StringUtils.replaceAll("", ".+", "zzz")  = ""
	 * StringUtils.replaceAll("abc", "", "ZZ")  = "ZZaZZbZZcZZ"
	 * StringUtils.replaceAll("<__>\n<__>", "<.*>", "z")      = "z\nz"
	 * StringUtils.replaceAll("<__>\n<__>", "(?s)<.*>", "z")  = "z"
	 * StringUtils.replaceAll("ABCabc123", "[a-z]", "_")       = "ABC___123"
	 * StringUtils.replaceAll("ABCabc123", "[^A-Z0-9]+", "_")  = "ABC_123"
	 * StringUtils.replaceAll("ABCabc123", "[^A-Z0-9]+", "")   = "ABC123"
	 * StringUtils.replaceAll("Lorem ipsum  dolor   sit", "( +)([a-z]+)", "_$2")  = "Lorem_ipsum_dolor_sit"
	 * 
* * @param text text to search and replace in, may be null * @param regex the regular expression to which this string is to be * matched * @param replacement the string to be substituted for each match * @return the text with any replacements processed, {@code null} if null String * input * * @throws java.util.regex.PatternSyntaxException if the regular expression's * syntax is invalid * * @see #replacePattern(String, String, String) * @see String#replaceAll(String, String) * @see java.util.regex.Pattern * @see java.util.regex.Pattern#DOTALL * @since 3.5 * * @deprecated Moved to RegExUtils. */ @Deprecated public static String replaceAll(final String text, final String regex, final String replacement) { return RegExUtils.replaceAll(text, regex, replacement); } /** *

* Replaces all occurrences of a character in a String with another. This is a * null-safe version of {@link String#replace(char, char)}. *

* *

* A {@code null} string input returns {@code null}. An empty ("") string input * returns an empty string. *

* *
	 * StringUtils.replaceChars(null, *, *)        = null
	 * StringUtils.replaceChars("", *, *)          = ""
	 * StringUtils.replaceChars("abcba", 'b', 'y') = "aycya"
	 * StringUtils.replaceChars("abcba", 'z', 'y') = "abcba"
	 * 
* * @param str String to replace characters in, may be null * @param searchChar the character to search for, may be null * @param replaceChar the character to replace, may be null * @return modified String, {@code null} if null string input * @since 2.0 */ public static String replaceChars(final String str, final char searchChar, final char replaceChar) { if (str == null) { return null; } return str.replace(searchChar, replaceChar); } /** *

* Replaces multiple characters in a String in one go. This method can also be * used to delete characters. *

* *

* For example:
* {@code replaceChars("hello", "ho", "jy") = jelly}. *

* *

* A {@code null} string input returns {@code null}. An empty ("") string input * returns an empty string. A null or empty set of search characters returns the * input string. *

* *

* The length of the search characters should normally equal the length of the * replace characters. If the search characters is longer, then the extra search * characters are deleted. If the search characters is shorter, then the extra * replace characters are ignored. *

* *
	 * StringUtils.replaceChars(null, *, *)           = null
	 * StringUtils.replaceChars("", *, *)             = ""
	 * StringUtils.replaceChars("abc", null, *)       = "abc"
	 * StringUtils.replaceChars("abc", "", *)         = "abc"
	 * StringUtils.replaceChars("abc", "b", null)     = "ac"
	 * StringUtils.replaceChars("abc", "b", "")       = "ac"
	 * StringUtils.replaceChars("abcba", "bc", "yz")  = "ayzya"
	 * StringUtils.replaceChars("abcba", "bc", "y")   = "ayya"
	 * StringUtils.replaceChars("abcba", "bc", "yzx") = "ayzya"
	 * 
* * @param str String to replace characters in, may be null * @param searchChars a set of characters to search for, may be null * @param replaceChars a set of characters to replace, may be null * @return modified String, {@code null} if null string input * @since 2.0 */ public static String replaceChars(final String str, final String searchChars, String replaceChars) { if (isEmpty(str) || isEmpty(searchChars)) { return str; } if (replaceChars == null) { replaceChars = EMPTY; } boolean modified = false; final int replaceCharsLength = replaceChars.length(); final int strLength = str.length(); final StringBuilder buf = new StringBuilder(strLength); for (int i = 0; i < strLength; i++) { final char ch = str.charAt(i); final int index = searchChars.indexOf(ch); if (index >= 0) { modified = true; if (index < replaceCharsLength) { buf.append(replaceChars.charAt(index)); } } else { buf.append(ch); } } if (modified) { return buf.toString(); } return str; } /** *

* Replaces all occurrences of Strings within another String. *

* *

* A {@code null} reference passed to this method is a no-op, or if any "search * string" or "string to replace" is null, that replace will be ignored. This * will not repeat. For repeating replaces, call the overloaded method. *

* *
	 *  StringUtils.replaceEach(null, *, *)        = null
	 *  StringUtils.replaceEach("", *, *)          = ""
	 *  StringUtils.replaceEach("aba", null, null) = "aba"
	 *  StringUtils.replaceEach("aba", new String[0], null) = "aba"
	 *  StringUtils.replaceEach("aba", null, new String[0]) = "aba"
	 *  StringUtils.replaceEach("aba", new String[]{"a"}, null)  = "aba"
	 *  StringUtils.replaceEach("aba", new String[]{"a"}, new String[]{""})  = "b"
	 *  StringUtils.replaceEach("aba", new String[]{null}, new String[]{"a"})  = "aba"
	 *  StringUtils.replaceEach("abcde", new String[]{"ab", "d"}, new String[]{"w", "t"})  = "wcte"
	 *  (example of how it does not repeat)
	 *  StringUtils.replaceEach("abcde", new String[]{"ab", "d"}, new String[]{"d", "t"})  = "dcte"
	 * 
* * @param text text to search and replace in, no-op if null * @param searchList the Strings to search for, no-op if null * @param replacementList the Strings to replace them with, no-op if null * @return the text with any replacements processed, {@code null} if null String * input * @throws IllegalArgumentException if the lengths of the arrays are not the * same (null is ok, and/or size 0) * @since 2.4 */ public static String replaceEach(final String text, final String[] searchList, final String[] replacementList) { return replaceEach(text, searchList, replacementList, false, 0); } /** *

* Replace all occurrences of Strings within another String. This is a private * recursive helper method for * {@link #replaceEachRepeatedly(String, String[], String[])} and * {@link #replaceEach(String, String[], String[])} *

* *

* A {@code null} reference passed to this method is a no-op, or if any "search * string" or "string to replace" is null, that replace will be ignored. *

* *
	 *  StringUtils.replaceEach(null, *, *, *, *) = null
	 *  StringUtils.replaceEach("", *, *, *, *) = ""
	 *  StringUtils.replaceEach("aba", null, null, *, *) = "aba"
	 *  StringUtils.replaceEach("aba", new String[0], null, *, *) = "aba"
	 *  StringUtils.replaceEach("aba", null, new String[0], *, *) = "aba"
	 *  StringUtils.replaceEach("aba", new String[]{"a"}, null, *, *) = "aba"
	 *  StringUtils.replaceEach("aba", new String[]{"a"}, new String[]{""}, *, >=0) = "b"
	 *  StringUtils.replaceEach("aba", new String[]{null}, new String[]{"a"}, *, >=0) = "aba"
	 *  StringUtils.replaceEach("abcde", new String[]{"ab", "d"}, new String[]{"w", "t"}, *, >=0) = "wcte"
	 *  (example of how it repeats)
	 *  StringUtils.replaceEach("abcde", new String[]{"ab", "d"}, new String[]{"d", "t"}, false, >=0) = "dcte"
	 *  StringUtils.replaceEach("abcde", new String[]{"ab", "d"}, new String[]{"d", "t"}, true, >=2) = "tcte"
	 *  StringUtils.replaceEach("abcde", new String[]{"ab", "d"}, new String[]{"d", "ab"}, *, *) = IllegalStateException
	 * 
* * @param text text to search and replace in, no-op if null * @param searchList the Strings to search for, no-op if null * @param replacementList the Strings to replace them with, no-op if null * @param repeat if true, then replace repeatedly until there are no * more possible replacements or timeToLive < 0 * @param timeToLive if less than 0 then there is a circular reference and * endless loop * @return the text with any replacements processed, {@code null} if null String * input * @throws IllegalStateException if the search is repeating and there is an * endless loop due to outputs of one being * inputs to another * @throws IllegalArgumentException if the lengths of the arrays are not the * same (null is ok, and/or size 0) * @since 2.4 */ private static String replaceEach(final String text, final String[] searchList, final String[] replacementList, final boolean repeat, final int timeToLive) { // mchyzer Performance note: This creates very few new objects (one major goal) // let me know if there are performance requests, we can create a harness to // measure // if recursing, this shouldn't be less than 0 if (timeToLive < 0) { final Set searchSet = new HashSet<>(Arrays.asList(searchList)); final Set replacementSet = new HashSet<>(Arrays.asList(replacementList)); searchSet.retainAll(replacementSet); if (!searchSet.isEmpty()) { throw new IllegalStateException("Aborting to protect against StackOverflowError - " + "output of one loop is the input of another"); } } if (isEmpty(text) || searchList.length == 0 || replacementList.length == 0 || (searchList.length > 0 && timeToLive == -1)) { return text; } final int searchLength = searchList.length; final int replacementLength = replacementList.length; // make sure lengths are ok, these need to be equal if (searchLength != replacementLength) { throw new IllegalArgumentException( "Search and Replace array lengths don't match: " + searchLength + " vs " + replacementLength); } // keep track of which still have matches final boolean[] noMoreMatchesForReplIndex = new boolean[searchLength]; // index on index that the match was found int textIndex = -1; int replaceIndex = -1; int tempIndex = -1; // index of replace array that will replace the search string found // NOTE: logic duplicated below START for (int i = 0; i < searchLength; i++) { if (noMoreMatchesForReplIndex[i] || isEmpty(searchList[i]) || replacementList[i] == null) { continue; } tempIndex = text.indexOf(searchList[i]); // see if we need to keep searching for this if (tempIndex == -1) { noMoreMatchesForReplIndex[i] = true; } else if (textIndex == -1 || tempIndex < textIndex) { textIndex = tempIndex; replaceIndex = i; } } // NOTE: logic mostly below END // no search strings found, we are done if (textIndex == -1) { return text; } int start = 0; // get a good guess on the size of the result buffer so it doesn't have to // double if it goes over a bit int increase = 0; // count the replacement text elements that are larger than their corresponding // text being replaced for (int i = 0; i < searchList.length; i++) { if (searchList[i] == null || replacementList[i] == null) { continue; } final int greater = replacementList[i].length() - searchList[i].length(); if (greater > 0) { increase += 3 * greater; // assume 3 matches } } // have upper-bound at 20% increase, then let Java take over increase = Math.min(increase, text.length() / 5); final StringBuilder buf = new StringBuilder(text.length() + increase); while (textIndex != -1) { for (int i = start; i < textIndex; i++) { buf.append(text.charAt(i)); } buf.append(replacementList[replaceIndex]); start = textIndex + searchList[replaceIndex].length(); textIndex = -1; replaceIndex = -1; // find the next earliest match // NOTE: logic mostly duplicated above START for (int i = 0; i < searchLength; i++) { if (noMoreMatchesForReplIndex[i] || searchList[i] == null || searchList[i].isEmpty() || replacementList[i] == null) { continue; } tempIndex = text.indexOf(searchList[i], start); // see if we need to keep searching for this if (tempIndex == -1) { noMoreMatchesForReplIndex[i] = true; } else if (textIndex == -1 || tempIndex < textIndex) { textIndex = tempIndex; replaceIndex = i; } } // NOTE: logic duplicated above END } final int textLength = text.length(); for (int i = start; i < textLength; i++) { buf.append(text.charAt(i)); } final String result = buf.toString(); if (!repeat) { return result; } return replaceEach(result, searchList, replacementList, repeat, timeToLive - 1); } /** *

* Replaces all occurrences of Strings within another String. *

* *

* A {@code null} reference passed to this method is a no-op, or if any "search * string" or "string to replace" is null, that replace will be ignored. *

* *
	 *  StringUtils.replaceEachRepeatedly(null, *, *) = null
	 *  StringUtils.replaceEachRepeatedly("", *, *) = ""
	 *  StringUtils.replaceEachRepeatedly("aba", null, null) = "aba"
	 *  StringUtils.replaceEachRepeatedly("aba", new String[0], null) = "aba"
	 *  StringUtils.replaceEachRepeatedly("aba", null, new String[0]) = "aba"
	 *  StringUtils.replaceEachRepeatedly("aba", new String[]{"a"}, null) = "aba"
	 *  StringUtils.replaceEachRepeatedly("aba", new String[]{"a"}, new String[]{""}) = "b"
	 *  StringUtils.replaceEachRepeatedly("aba", new String[]{null}, new String[]{"a"}) = "aba"
	 *  StringUtils.replaceEachRepeatedly("abcde", new String[]{"ab", "d"}, new String[]{"w", "t"}) = "wcte"
	 *  (example of how it repeats)
	 *  StringUtils.replaceEachRepeatedly("abcde", new String[]{"ab", "d"}, new String[]{"d", "t"}) = "tcte"
	 *  StringUtils.replaceEachRepeatedly("abcde", new String[]{"ab", "d"}, new String[]{"d", "ab"}) = IllegalStateException
	 * 
* * @param text text to search and replace in, no-op if null * @param searchList the Strings to search for, no-op if null * @param replacementList the Strings to replace them with, no-op if null * @return the text with any replacements processed, {@code null} if null String * input * @throws IllegalStateException if the search is repeating and there is an * endless loop due to outputs of one being * inputs to another * @throws IllegalArgumentException if the lengths of the arrays are not the * same (null is ok, and/or size 0) * @since 2.4 */ public static String replaceEachRepeatedly(final String text, final String[] searchList, final String[] replacementList) { // timeToLive should be 0 if not used or nothing to replace, else it's // the length of the replace array final int timeToLive = searchList == null ? 0 : searchList.length; return replaceEach(text, searchList, replacementList, true, timeToLive); } /** *

* Replaces the first substring of the text string that matches the given * regular expression with the given replacement. *

* * This method is a {@code null} safe equivalent to: *
    *
  • {@code text.replaceFirst(regex, replacement)}
  • *
  • {@code Pattern.compile(regex).matcher(text).replaceFirst(replacement)}
  • *
* *

* A {@code null} reference passed to this method is a no-op. *

* *

* The {@link Pattern#DOTALL} option is NOT automatically added. To use the * DOTALL option prepend {@code "(?s)"} to the regex. DOTALL is also known as * single-line mode in Perl. *

* *
	 * StringUtils.replaceFirst(null, *, *)       = null
	 * StringUtils.replaceFirst("any", (String) null, *)   = "any"
	 * StringUtils.replaceFirst("any", *, null)   = "any"
	 * StringUtils.replaceFirst("", "", "zzz")    = "zzz"
	 * StringUtils.replaceFirst("", ".*", "zzz")  = "zzz"
	 * StringUtils.replaceFirst("", ".+", "zzz")  = ""
	 * StringUtils.replaceFirst("abc", "", "ZZ")  = "ZZabc"
	 * StringUtils.replaceFirst("<__>\n<__>", "<.*>", "z")      = "z\n<__>"
	 * StringUtils.replaceFirst("<__>\n<__>", "(?s)<.*>", "z")  = "z"
	 * StringUtils.replaceFirst("ABCabc123", "[a-z]", "_")          = "ABC_bc123"
	 * StringUtils.replaceFirst("ABCabc123abc", "[^A-Z0-9]+", "_")  = "ABC_123abc"
	 * StringUtils.replaceFirst("ABCabc123abc", "[^A-Z0-9]+", "")   = "ABC123abc"
	 * StringUtils.replaceFirst("Lorem ipsum  dolor   sit", "( +)([a-z]+)", "_$2")  = "Lorem_ipsum  dolor   sit"
	 * 
* * @param text text to search and replace in, may be null * @param regex the regular expression to which this string is to be * matched * @param replacement the string to be substituted for the first match * @return the text with the first replacement processed, {@code null} if null * String input * * @throws java.util.regex.PatternSyntaxException if the regular expression's * syntax is invalid * * @see String#replaceFirst(String, String) * @see java.util.regex.Pattern * @see java.util.regex.Pattern#DOTALL * @since 3.5 * * @deprecated Moved to RegExUtils. */ @Deprecated public static String replaceFirst(final String text, final String regex, final String replacement) { return RegExUtils.replaceFirst(text, regex, replacement); } /** *

* Case insensitively replaces all occurrences of a String within another * String. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replaceIgnoreCase(null, *, *)        = null
	 * StringUtils.replaceIgnoreCase("", *, *)          = ""
	 * StringUtils.replaceIgnoreCase("any", null, *)    = "any"
	 * StringUtils.replaceIgnoreCase("any", *, null)    = "any"
	 * StringUtils.replaceIgnoreCase("any", "", *)      = "any"
	 * StringUtils.replaceIgnoreCase("aba", "a", null)  = "aba"
	 * StringUtils.replaceIgnoreCase("abA", "A", "")    = "b"
	 * StringUtils.replaceIgnoreCase("aba", "A", "z")   = "zbz"
	 * 
* * @see #replaceIgnoreCase(String text, String searchString, String replacement, * int max) * @param text text to search and replace in, may be null * @param searchString the String to search for (case insensitive), may be null * @param replacement the String to replace it with, may be null * @return the text with any replacements processed, {@code null} if null String * input * @since 3.5 */ public static String replaceIgnoreCase(final String text, final String searchString, final String replacement) { return replaceIgnoreCase(text, searchString, replacement, -1); } /** *

* Case insensitively replaces a String with another String inside a larger * String, for the first {@code max} values of the search String. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replaceIgnoreCase(null, *, *, *)         = null
	 * StringUtils.replaceIgnoreCase("", *, *, *)           = ""
	 * StringUtils.replaceIgnoreCase("any", null, *, *)     = "any"
	 * StringUtils.replaceIgnoreCase("any", *, null, *)     = "any"
	 * StringUtils.replaceIgnoreCase("any", "", *, *)       = "any"
	 * StringUtils.replaceIgnoreCase("any", *, *, 0)        = "any"
	 * StringUtils.replaceIgnoreCase("abaa", "a", null, -1) = "abaa"
	 * StringUtils.replaceIgnoreCase("abaa", "a", "", -1)   = "b"
	 * StringUtils.replaceIgnoreCase("abaa", "a", "z", 0)   = "abaa"
	 * StringUtils.replaceIgnoreCase("abaa", "A", "z", 1)   = "zbaa"
	 * StringUtils.replaceIgnoreCase("abAa", "a", "z", 2)   = "zbza"
	 * StringUtils.replaceIgnoreCase("abAa", "a", "z", -1)  = "zbzz"
	 * 
* * @param text text to search and replace in, may be null * @param searchString the String to search for (case insensitive), may be null * @param replacement the String to replace it with, may be null * @param max maximum number of values to replace, or {@code -1} if no * maximum * @return the text with any replacements processed, {@code null} if null String * input * @since 3.5 */ public static String replaceIgnoreCase(final String text, final String searchString, final String replacement, final int max) { return replace(text, searchString, replacement, max, true); } /** *

* Replaces a String with another String inside a larger String, once. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replaceOnce(null, *, *)        = null
	 * StringUtils.replaceOnce("", *, *)          = ""
	 * StringUtils.replaceOnce("any", null, *)    = "any"
	 * StringUtils.replaceOnce("any", *, null)    = "any"
	 * StringUtils.replaceOnce("any", "", *)      = "any"
	 * StringUtils.replaceOnce("aba", "a", null)  = "aba"
	 * StringUtils.replaceOnce("aba", "a", "")    = "ba"
	 * StringUtils.replaceOnce("aba", "a", "z")   = "zba"
	 * 
* * @see #replace(String text, String searchString, String replacement, int max) * @param text text to search and replace in, may be null * @param searchString the String to search for, may be null * @param replacement the String to replace with, may be null * @return the text with any replacements processed, {@code null} if null String * input */ public static String replaceOnce(final String text, final String searchString, final String replacement) { return replace(text, searchString, replacement, 1); } /** *

* Case insensitively replaces a String with another String inside a larger * String, once. *

* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replaceOnceIgnoreCase(null, *, *)        = null
	 * StringUtils.replaceOnceIgnoreCase("", *, *)          = ""
	 * StringUtils.replaceOnceIgnoreCase("any", null, *)    = "any"
	 * StringUtils.replaceOnceIgnoreCase("any", *, null)    = "any"
	 * StringUtils.replaceOnceIgnoreCase("any", "", *)      = "any"
	 * StringUtils.replaceOnceIgnoreCase("aba", "a", null)  = "aba"
	 * StringUtils.replaceOnceIgnoreCase("aba", "a", "")    = "ba"
	 * StringUtils.replaceOnceIgnoreCase("aba", "a", "z")   = "zba"
	 * StringUtils.replaceOnceIgnoreCase("FoOFoofoo", "foo", "") = "Foofoo"
	 * 
* * @see #replaceIgnoreCase(String text, String searchString, String replacement, * int max) * @param text text to search and replace in, may be null * @param searchString the String to search for (case insensitive), may be null * @param replacement the String to replace with, may be null * @return the text with any replacements processed, {@code null} if null String * input * @since 3.5 */ public static String replaceOnceIgnoreCase(final String text, final String searchString, final String replacement) { return replaceIgnoreCase(text, searchString, replacement, 1); } /** *

* Replaces each substring of the source String that matches the given regular * expression with the given replacement using the {@link Pattern#DOTALL} * option. DOTALL is also known as single-line mode in Perl. *

* * This call is a {@code null} safe equivalent to: *
    *
  • {@code source.replaceAll("(?s)" + regex, replacement)}
  • *
  • {@code Pattern.compile(regex, Pattern.DOTALL).matcher(source).replaceAll(replacement)}
  • *
* *

* A {@code null} reference passed to this method is a no-op. *

* *
	 * StringUtils.replacePattern(null, *, *)       = null
	 * StringUtils.replacePattern("any", (String) null, *)   = "any"
	 * StringUtils.replacePattern("any", *, null)   = "any"
	 * StringUtils.replacePattern("", "", "zzz")    = "zzz"
	 * StringUtils.replacePattern("", ".*", "zzz")  = "zzz"
	 * StringUtils.replacePattern("", ".+", "zzz")  = ""
	 * StringUtils.replacePattern("<__>\n<__>", "<.*>", "z")       = "z"
	 * StringUtils.replacePattern("ABCabc123", "[a-z]", "_")       = "ABC___123"
	 * StringUtils.replacePattern("ABCabc123", "[^A-Z0-9]+", "_")  = "ABC_123"
	 * StringUtils.replacePattern("ABCabc123", "[^A-Z0-9]+", "")   = "ABC123"
	 * StringUtils.replacePattern("Lorem ipsum  dolor   sit", "( +)([a-z]+)", "_$2")  = "Lorem_ipsum_dolor_sit"
	 * 
* * @param source the source string * @param regex the regular expression to which this string is to be * matched * @param replacement the string to be substituted for each match * @return The resulting {@code String} * @see #replaceAll(String, String, String) * @see String#replaceAll(String, String) * @see Pattern#DOTALL * @since 3.2 * @since 3.5 Changed {@code null} reference passed to this method is a no-op. * * @deprecated Moved to RegExUtils. */ @Deprecated public static String replacePattern(final String source, final String regex, final String replacement) { return RegExUtils.replacePattern(source, regex, replacement); } /** *

* Reverses a String as per {@link StringBuilder#reverse()}. *

* *

* A {@code null} String returns {@code null}. *

* *
	 * StringUtils.reverse(null)  = null
	 * StringUtils.reverse("")    = ""
	 * StringUtils.reverse("bat") = "tab"
	 * 
* * @param str the String to reverse, may be null * @return the reversed String, {@code null} if null String input */ public static String reverse(final String str) { if (str == null) { return null; } return new StringBuilder(str).reverse().toString(); } /** *

* Reverses a String that is delimited by a specific character. *

* *

* The Strings between the delimiters are not reversed. Thus java.lang.String * becomes String.lang.java (if the delimiter is {@code '.'}). *

* *
	 * StringUtils.reverseDelimited(null, *)      = null
	 * StringUtils.reverseDelimited("", *)        = ""
	 * StringUtils.reverseDelimited("a.b.c", 'x') = "a.b.c"
	 * StringUtils.reverseDelimited("a.b.c", ".") = "c.b.a"
	 * 
* * @param str the String to reverse, may be null * @param separatorChar the separator character to use * @return the reversed String, {@code null} if null String input * @since 2.0 */ public static String reverseDelimited(final String str, final char separatorChar) { if (str == null) { return null; } // could implement manually, but simple way is to reuse other, // probably slower, methods. /* * final String[] strs = split(str, separatorChar); ArrayUtils.reverse(strs); * return join(strs, separatorChar); */ throw new UnsupportedOperationException("Not supported in TeaVM"); } /** *

* Gets the rightmost {@code len} characters of a String. *

* *

* If {@code len} characters are not available, or the String is {@code null}, * the String will be returned without an an exception. An empty String is * returned if len is negative. *

* *
	 * StringUtils.right(null, *)    = null
	 * StringUtils.right(*, -ve)     = ""
	 * StringUtils.right("", *)      = ""
	 * StringUtils.right("abc", 0)   = ""
	 * StringUtils.right("abc", 2)   = "bc"
	 * StringUtils.right("abc", 4)   = "abc"
	 * 
* * @param str the String to get the rightmost characters from, may be null * @param len the length of the required String * @return the rightmost characters, {@code null} if null String input */ public static String right(final String str, final int len) { if (str == null) { return null; } if (len < 0) { return EMPTY; } if (str.length() <= len) { return str; } return str.substring(str.length() - len); } /** *

* Right pad a String with spaces (' '). *

* *

* The String is padded to the size of {@code size}. *

* *
	 * StringUtils.rightPad(null, *)   = null
	 * StringUtils.rightPad("", 3)     = "   "
	 * StringUtils.rightPad("bat", 3)  = "bat"
	 * StringUtils.rightPad("bat", 5)  = "bat  "
	 * StringUtils.rightPad("bat", 1)  = "bat"
	 * StringUtils.rightPad("bat", -1) = "bat"
	 * 
* * @param str the String to pad out, may be null * @param size the size to pad to * @return right padded String or original String if no padding is necessary, * {@code null} if null String input */ public static String rightPad(final String str, final int size) { return rightPad(str, size, ' '); } /** *

* Right pad a String with a specified character. *

* *

* The String is padded to the size of {@code size}. *

* *
	 * StringUtils.rightPad(null, *, *)     = null
	 * StringUtils.rightPad("", 3, 'z')     = "zzz"
	 * StringUtils.rightPad("bat", 3, 'z')  = "bat"
	 * StringUtils.rightPad("bat", 5, 'z')  = "batzz"
	 * StringUtils.rightPad("bat", 1, 'z')  = "bat"
	 * StringUtils.rightPad("bat", -1, 'z') = "bat"
	 * 
* * @param str the String to pad out, may be null * @param size the size to pad to * @param padChar the character to pad with * @return right padded String or original String if no padding is necessary, * {@code null} if null String input * @since 2.0 */ public static String rightPad(final String str, final int size, final char padChar) { if (str == null) { return null; } final int pads = size - str.length(); if (pads <= 0) { return str; // returns original String when possible } if (pads > PAD_LIMIT) { return rightPad(str, size, String.valueOf(padChar)); } return str.concat(repeat(padChar, pads)); } /** *

* Right pad a String with a specified String. *

* *

* The String is padded to the size of {@code size}. *

* *
	 * StringUtils.rightPad(null, *, *)      = null
	 * StringUtils.rightPad("", 3, "z")      = "zzz"
	 * StringUtils.rightPad("bat", 3, "yz")  = "bat"
	 * StringUtils.rightPad("bat", 5, "yz")  = "batyz"
	 * StringUtils.rightPad("bat", 8, "yz")  = "batyzyzy"
	 * StringUtils.rightPad("bat", 1, "yz")  = "bat"
	 * StringUtils.rightPad("bat", -1, "yz") = "bat"
	 * StringUtils.rightPad("bat", 5, null)  = "bat  "
	 * StringUtils.rightPad("bat", 5, "")    = "bat  "
	 * 
* * @param str the String to pad out, may be null * @param size the size to pad to * @param padStr the String to pad with, null or empty treated as single space * @return right padded String or original String if no padding is necessary, * {@code null} if null String input */ public static String rightPad(final String str, final int size, String padStr) { if (str == null) { return null; } if (isEmpty(padStr)) { padStr = SPACE; } final int padLen = padStr.length(); final int strLen = str.length(); final int pads = size - strLen; if (pads <= 0) { return str; // returns original String when possible } if (padLen == 1 && pads <= PAD_LIMIT) { return rightPad(str, size, padStr.charAt(0)); } if (pads == padLen) { return str.concat(padStr); } else if (pads < padLen) { return str.concat(padStr.substring(0, pads)); } else { final char[] padding = new char[pads]; final char[] padChars = padStr.toCharArray(); for (int i = 0; i < pads; i++) { padding[i] = padChars[i % padLen]; } return str.concat(new String(padding)); } } /** *

* Rotate (circular shift) a String of {@code shift} characters. *

*
    *
  • If {@code shift > 0}, right circular shift (ex : ABCDEF => * FABCDE)
  • *
  • If {@code shift < 0}, left circular shift (ex : ABCDEF => BCDEFA)
  • *
* *
	 * StringUtils.rotate(null, *)        = null
	 * StringUtils.rotate("", *)          = ""
	 * StringUtils.rotate("abcdefg", 0)   = "abcdefg"
	 * StringUtils.rotate("abcdefg", 2)   = "fgabcde"
	 * StringUtils.rotate("abcdefg", -2)  = "cdefgab"
	 * StringUtils.rotate("abcdefg", 7)   = "abcdefg"
	 * StringUtils.rotate("abcdefg", -7)  = "abcdefg"
	 * StringUtils.rotate("abcdefg", 9)   = "fgabcde"
	 * StringUtils.rotate("abcdefg", -9)  = "cdefgab"
	 * 
* * @param str the String to rotate, may be null * @param shift number of time to shift (positive : right shift, negative : left * shift) * @return the rotated String, or the original String if {@code shift == 0}, or * {@code null} if null String input * @since 3.5 */ public static String rotate(final String str, final int shift) { if (str == null) { return null; } final int strLen = str.length(); if (shift == 0 || strLen == 0 || shift % strLen == 0) { return str; } final StringBuilder builder = new StringBuilder(strLen); final int offset = -(shift % strLen); builder.append(substring(str, offset)); builder.append(substring(str, 0, offset)); return builder.toString(); } /** *

* Splits the provided text into an array, using whitespace as the separator. * Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as one separator. For more control over the split use * the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.split(null)       = null
	 * StringUtils.split("")         = []
	 * StringUtils.split("abc def")  = ["abc", "def"]
	 * StringUtils.split("abc  def") = ["abc", "def"]
	 * StringUtils.split(" abc ")    = ["abc"]
	 * 
* * @param str the String to parse, may be null * @return an array of parsed Strings, {@code null} if null String input */ public static String[] split(final String str) { return split(str, null, -1); } /** *

* Splits the provided text into an array, separator specified. This is an * alternative to using StringTokenizer. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as one separator. For more control over the split use * the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.split(null, *)         = null
	 * StringUtils.split("", *)           = []
	 * StringUtils.split("a.b.c", '.')    = ["a", "b", "c"]
	 * StringUtils.split("a..b.c", '.')   = ["a", "b", "c"]
	 * StringUtils.split("a:b:c", '.')    = ["a:b:c"]
	 * StringUtils.split("a b c", ' ')    = ["a", "b", "c"]
	 * 
* * @param str the String to parse, may be null * @param separatorChar the character used as the delimiter * @return an array of parsed Strings, {@code null} if null String input * @since 2.0 */ public static String[] split(final String str, final char separatorChar) { return splitWorker(str, separatorChar, false); } /** *

* Splits the provided text into an array, separators specified. This is an * alternative to using StringTokenizer. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as one separator. For more control over the split use * the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. A {@code null} * separatorChars splits on whitespace. *

* *
	 * StringUtils.split(null, *)         = null
	 * StringUtils.split("", *)           = []
	 * StringUtils.split("abc def", null) = ["abc", "def"]
	 * StringUtils.split("abc def", " ")  = ["abc", "def"]
	 * StringUtils.split("abc  def", " ") = ["abc", "def"]
	 * StringUtils.split("ab:cd:ef", ":") = ["ab", "cd", "ef"]
	 * 
* * @param str the String to parse, may be null * @param separatorChars the characters used as the delimiters, {@code null} * splits on whitespace * @return an array of parsed Strings, {@code null} if null String input */ public static String[] split(final String str, final String separatorChars) { return splitWorker(str, separatorChars, -1, false); } /** *

* Splits the provided text into an array with a maximum length, separators * specified. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as one separator. *

* *

* A {@code null} input String returns {@code null}. A {@code null} * separatorChars splits on whitespace. *

* *

* If more than {@code max} delimited substrings are found, the last returned * string includes all characters after the first {@code max - 1} returned * strings (including separator characters). *

* *
	 * StringUtils.split(null, *, *)            = null
	 * StringUtils.split("", *, *)              = []
	 * StringUtils.split("ab cd ef", null, 0)   = ["ab", "cd", "ef"]
	 * StringUtils.split("ab   cd ef", null, 0) = ["ab", "cd", "ef"]
	 * StringUtils.split("ab:cd:ef", ":", 0)    = ["ab", "cd", "ef"]
	 * StringUtils.split("ab:cd:ef", ":", 2)    = ["ab", "cd:ef"]
	 * 
* * @param str the String to parse, may be null * @param separatorChars the characters used as the delimiters, {@code null} * splits on whitespace * @param max the maximum number of elements to include in the array. * A zero or negative value implies no limit * @return an array of parsed Strings, {@code null} if null String input */ public static String[] split(final String str, final String separatorChars, final int max) { return splitWorker(str, separatorChars, max, false); } /** *

* Splits a String by Character type as returned by * {@code java.lang.Character.getType(char)}. Groups of contiguous characters of * the same type are returned as complete tokens. * *

	 * StringUtils.splitByCharacterType(null)         = null
	 * StringUtils.splitByCharacterType("")           = []
	 * StringUtils.splitByCharacterType("ab de fg")   = ["ab", " ", "de", " ", "fg"]
	 * StringUtils.splitByCharacterType("ab   de fg") = ["ab", "   ", "de", " ", "fg"]
	 * StringUtils.splitByCharacterType("ab:cd:ef")   = ["ab", ":", "cd", ":", "ef"]
	 * StringUtils.splitByCharacterType("number5")    = ["number", "5"]
	 * StringUtils.splitByCharacterType("fooBar")     = ["foo", "B", "ar"]
	 * StringUtils.splitByCharacterType("foo200Bar")  = ["foo", "200", "B", "ar"]
	 * StringUtils.splitByCharacterType("ASFRules")   = ["ASFR", "ules"]
	 * 
* * @param str the String to split, may be {@code null} * @return an array of parsed Strings, {@code null} if null String input * @since 2.4 */ public static String[] splitByCharacterType(final String str) { return splitByCharacterType(str, false); } /** *

* Splits a String by Character type as returned by * {@code java.lang.Character.getType(char)}. Groups of contiguous characters of * the same type are returned as complete tokens, with the following exception: * if {@code camelCase} is {@code true}, the character of type * {@code Character.UPPERCASE_LETTER}, if any, immediately preceding a token of * type {@code Character.LOWERCASE_LETTER} will belong to the following token * rather than to the preceding, if any, {@code Character.UPPERCASE_LETTER} * token. * * @param str the String to split, may be {@code null} * @param camelCase whether to use so-called "camel-case" for letter types * @return an array of parsed Strings, {@code null} if null String input * @since 2.4 */ private static String[] splitByCharacterType(final String str, final boolean camelCase) { if (str == null) { return null; } if (str.isEmpty()) { return new String[0]; } final char[] c = str.toCharArray(); final List list = new ArrayList<>(); int tokenStart = 0; int currentType = Character.getType(c[tokenStart]); for (int pos = tokenStart + 1; pos < c.length; pos++) { final int type = Character.getType(c[pos]); if (type == currentType) { continue; } if (camelCase && type == Character.LOWERCASE_LETTER && currentType == Character.UPPERCASE_LETTER) { final int newTokenStart = pos - 1; if (newTokenStart != tokenStart) { list.add(new String(c, tokenStart, newTokenStart - tokenStart)); tokenStart = newTokenStart; } } else { list.add(new String(c, tokenStart, pos - tokenStart)); tokenStart = pos; } currentType = type; } list.add(new String(c, tokenStart, c.length - tokenStart)); return list.toArray(new String[0]); } /** *

* Splits a String by Character type as returned by * {@code java.lang.Character.getType(char)}. Groups of contiguous characters of * the same type are returned as complete tokens, with the following exception: * the character of type {@code Character.UPPERCASE_LETTER}, if any, immediately * preceding a token of type {@code Character.LOWERCASE_LETTER} will belong to * the following token rather than to the preceding, if any, * {@code Character.UPPERCASE_LETTER} token. * *

	 * StringUtils.splitByCharacterTypeCamelCase(null)         = null
	 * StringUtils.splitByCharacterTypeCamelCase("")           = []
	 * StringUtils.splitByCharacterTypeCamelCase("ab de fg")   = ["ab", " ", "de", " ", "fg"]
	 * StringUtils.splitByCharacterTypeCamelCase("ab   de fg") = ["ab", "   ", "de", " ", "fg"]
	 * StringUtils.splitByCharacterTypeCamelCase("ab:cd:ef")   = ["ab", ":", "cd", ":", "ef"]
	 * StringUtils.splitByCharacterTypeCamelCase("number5")    = ["number", "5"]
	 * StringUtils.splitByCharacterTypeCamelCase("fooBar")     = ["foo", "Bar"]
	 * StringUtils.splitByCharacterTypeCamelCase("foo200Bar")  = ["foo", "200", "Bar"]
	 * StringUtils.splitByCharacterTypeCamelCase("ASFRules")   = ["ASF", "Rules"]
	 * 
* * @param str the String to split, may be {@code null} * @return an array of parsed Strings, {@code null} if null String input * @since 2.4 */ public static String[] splitByCharacterTypeCamelCase(final String str) { return splitByCharacterType(str, true); } /** *

* Splits the provided text into an array, separator string specified. *

* *

* The separator(s) will not be included in the returned String array. Adjacent * separators are treated as one separator. *

* *

* A {@code null} input String returns {@code null}. A {@code null} separator * splits on whitespace. *

* *
	 * StringUtils.splitByWholeSeparator(null, *)               = null
	 * StringUtils.splitByWholeSeparator("", *)                 = []
	 * StringUtils.splitByWholeSeparator("ab de fg", null)      = ["ab", "de", "fg"]
	 * StringUtils.splitByWholeSeparator("ab   de fg", null)    = ["ab", "de", "fg"]
	 * StringUtils.splitByWholeSeparator("ab:cd:ef", ":")       = ["ab", "cd", "ef"]
	 * StringUtils.splitByWholeSeparator("ab-!-cd-!-ef", "-!-") = ["ab", "cd", "ef"]
	 * 
* * @param str the String to parse, may be null * @param separator String containing the String to be used as a delimiter, * {@code null} splits on whitespace * @return an array of parsed Strings, {@code null} if null String was input */ public static String[] splitByWholeSeparator(final String str, final String separator) { return splitByWholeSeparatorWorker(str, separator, -1, false); } /** *

* Splits the provided text into an array, separator string specified. Returns a * maximum of {@code max} substrings. *

* *

* The separator(s) will not be included in the returned String array. Adjacent * separators are treated as one separator. *

* *

* A {@code null} input String returns {@code null}. A {@code null} separator * splits on whitespace. *

* *
	 * StringUtils.splitByWholeSeparator(null, *, *)               = null
	 * StringUtils.splitByWholeSeparator("", *, *)                 = []
	 * StringUtils.splitByWholeSeparator("ab de fg", null, 0)      = ["ab", "de", "fg"]
	 * StringUtils.splitByWholeSeparator("ab   de fg", null, 0)    = ["ab", "de", "fg"]
	 * StringUtils.splitByWholeSeparator("ab:cd:ef", ":", 2)       = ["ab", "cd:ef"]
	 * StringUtils.splitByWholeSeparator("ab-!-cd-!-ef", "-!-", 5) = ["ab", "cd", "ef"]
	 * StringUtils.splitByWholeSeparator("ab-!-cd-!-ef", "-!-", 2) = ["ab", "cd-!-ef"]
	 * 
* * @param str the String to parse, may be null * @param separator String containing the String to be used as a delimiter, * {@code null} splits on whitespace * @param max the maximum number of elements to include in the returned * array. A zero or negative value implies no limit. * @return an array of parsed Strings, {@code null} if null String was input */ public static String[] splitByWholeSeparator(final String str, final String separator, final int max) { return splitByWholeSeparatorWorker(str, separator, max, false); } /** *

* Splits the provided text into an array, separator string specified. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as separators for empty tokens. For more control over * the split use the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. A {@code null} separator * splits on whitespace. *

* *
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens(null, *)               = null
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("", *)                 = []
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab de fg", null)      = ["ab", "de", "fg"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab   de fg", null)    = ["ab", "", "", "de", "fg"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab:cd:ef", ":")       = ["ab", "cd", "ef"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab-!-cd-!-ef", "-!-") = ["ab", "cd", "ef"]
	 * 
* * @param str the String to parse, may be null * @param separator String containing the String to be used as a delimiter, * {@code null} splits on whitespace * @return an array of parsed Strings, {@code null} if null String was input * @since 2.4 */ public static String[] splitByWholeSeparatorPreserveAllTokens(final String str, final String separator) { return splitByWholeSeparatorWorker(str, separator, -1, true); } /** *

* Splits the provided text into an array, separator string specified. Returns a * maximum of {@code max} substrings. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as separators for empty tokens. For more control over * the split use the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. A {@code null} separator * splits on whitespace. *

* *
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens(null, *, *)               = null
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("", *, *)                 = []
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab de fg", null, 0)      = ["ab", "de", "fg"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab   de fg", null, 0)    = ["ab", "", "", "de", "fg"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab:cd:ef", ":", 2)       = ["ab", "cd:ef"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab-!-cd-!-ef", "-!-", 5) = ["ab", "cd", "ef"]
	 * StringUtils.splitByWholeSeparatorPreserveAllTokens("ab-!-cd-!-ef", "-!-", 2) = ["ab", "cd-!-ef"]
	 * 
* * @param str the String to parse, may be null * @param separator String containing the String to be used as a delimiter, * {@code null} splits on whitespace * @param max the maximum number of elements to include in the returned * array. A zero or negative value implies no limit. * @return an array of parsed Strings, {@code null} if null String was input * @since 2.4 */ public static String[] splitByWholeSeparatorPreserveAllTokens(final String str, final String separator, final int max) { return splitByWholeSeparatorWorker(str, separator, max, true); } /** * Performs the logic for the {@code splitByWholeSeparatorPreserveAllTokens} * methods. * * @param str the String to parse, may be {@code null} * @param separator String containing the String to be used as a * delimiter, {@code null} splits on whitespace * @param max the maximum number of elements to include in the * returned array. A zero or negative value implies no * limit. * @param preserveAllTokens if {@code true}, adjacent separators are treated as * empty token separators; if {@code false}, adjacent * separators are treated as one separator. * @return an array of parsed Strings, {@code null} if null String input * @since 2.4 */ private static String[] splitByWholeSeparatorWorker(final String str, final String separator, final int max, final boolean preserveAllTokens) { if (str == null) { return null; } final int len = str.length(); if (len == 0) { return new String[0]; } if (separator == null || EMPTY.equals(separator)) { // Split on whitespace. return splitWorker(str, null, max, preserveAllTokens); } final int separatorLength = separator.length(); final ArrayList substrings = new ArrayList<>(); int numberOfSubstrings = 0; int beg = 0; int end = 0; while (end < len) { end = str.indexOf(separator, beg); if (end > -1) { if (end > beg) { numberOfSubstrings += 1; if (numberOfSubstrings == max) { end = len; substrings.add(str.substring(beg)); } else { // The following is OK, because String.substring( beg, end ) excludes // the character at the position 'end'. substrings.add(str.substring(beg, end)); // Set the starting point for the next search. // The following is equivalent to beg = end + (separatorLength - 1) + 1, // which is the right calculation: beg = end + separatorLength; } } else { // We found a consecutive occurrence of the separator, so skip it. if (preserveAllTokens) { numberOfSubstrings += 1; if (numberOfSubstrings == max) { end = len; substrings.add(str.substring(beg)); } else { substrings.add(EMPTY); } } beg = end + separatorLength; } } else { // String.substring( beg ) goes from 'beg' to the end of the String. substrings.add(str.substring(beg)); end = len; } } return substrings.toArray(new String[0]); } /** *

* Splits the provided text into an array, using whitespace as the separator, * preserving all tokens, including empty tokens created by adjacent separators. * This is an alternative to using StringTokenizer. Whitespace is defined by * {@link Character#isWhitespace(char)}. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as separators for empty tokens. For more control over * the split use the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.splitPreserveAllTokens(null)       = null
	 * StringUtils.splitPreserveAllTokens("")         = []
	 * StringUtils.splitPreserveAllTokens("abc def")  = ["abc", "def"]
	 * StringUtils.splitPreserveAllTokens("abc  def") = ["abc", "", "def"]
	 * StringUtils.splitPreserveAllTokens(" abc ")    = ["", "abc", ""]
	 * 
* * @param str the String to parse, may be {@code null} * @return an array of parsed Strings, {@code null} if null String input * @since 2.1 */ public static String[] splitPreserveAllTokens(final String str) { return splitWorker(str, null, -1, true); } /** *

* Splits the provided text into an array, separator specified, preserving all * tokens, including empty tokens created by adjacent separators. This is an * alternative to using StringTokenizer. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as separators for empty tokens. For more control over * the split use the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.splitPreserveAllTokens(null, *)         = null
	 * StringUtils.splitPreserveAllTokens("", *)           = []
	 * StringUtils.splitPreserveAllTokens("a.b.c", '.')    = ["a", "b", "c"]
	 * StringUtils.splitPreserveAllTokens("a..b.c", '.')   = ["a", "", "b", "c"]
	 * StringUtils.splitPreserveAllTokens("a:b:c", '.')    = ["a:b:c"]
	 * StringUtils.splitPreserveAllTokens("a\tb\nc", null) = ["a", "b", "c"]
	 * StringUtils.splitPreserveAllTokens("a b c", ' ')    = ["a", "b", "c"]
	 * StringUtils.splitPreserveAllTokens("a b c ", ' ')   = ["a", "b", "c", ""]
	 * StringUtils.splitPreserveAllTokens("a b c  ", ' ')   = ["a", "b", "c", "", ""]
	 * StringUtils.splitPreserveAllTokens(" a b c", ' ')   = ["", a", "b", "c"]
	 * StringUtils.splitPreserveAllTokens("  a b c", ' ')  = ["", "", a", "b", "c"]
	 * StringUtils.splitPreserveAllTokens(" a b c ", ' ')  = ["", a", "b", "c", ""]
	 * 
* * @param str the String to parse, may be {@code null} * @param separatorChar the character used as the delimiter, {@code null} splits * on whitespace * @return an array of parsed Strings, {@code null} if null String input * @since 2.1 */ public static String[] splitPreserveAllTokens(final String str, final char separatorChar) { return splitWorker(str, separatorChar, true); } /** *

* Splits the provided text into an array, separators specified, preserving all * tokens, including empty tokens created by adjacent separators. This is an * alternative to using StringTokenizer. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as separators for empty tokens. For more control over * the split use the StrTokenizer class. *

* *

* A {@code null} input String returns {@code null}. A {@code null} * separatorChars splits on whitespace. *

* *
	 * StringUtils.splitPreserveAllTokens(null, *)           = null
	 * StringUtils.splitPreserveAllTokens("", *)             = []
	 * StringUtils.splitPreserveAllTokens("abc def", null)   = ["abc", "def"]
	 * StringUtils.splitPreserveAllTokens("abc def", " ")    = ["abc", "def"]
	 * StringUtils.splitPreserveAllTokens("abc  def", " ")   = ["abc", "", def"]
	 * StringUtils.splitPreserveAllTokens("ab:cd:ef", ":")   = ["ab", "cd", "ef"]
	 * StringUtils.splitPreserveAllTokens("ab:cd:ef:", ":")  = ["ab", "cd", "ef", ""]
	 * StringUtils.splitPreserveAllTokens("ab:cd:ef::", ":") = ["ab", "cd", "ef", "", ""]
	 * StringUtils.splitPreserveAllTokens("ab::cd:ef", ":")  = ["ab", "", cd", "ef"]
	 * StringUtils.splitPreserveAllTokens(":cd:ef", ":")     = ["", cd", "ef"]
	 * StringUtils.splitPreserveAllTokens("::cd:ef", ":")    = ["", "", cd", "ef"]
	 * StringUtils.splitPreserveAllTokens(":cd:ef:", ":")    = ["", cd", "ef", ""]
	 * 
* * @param str the String to parse, may be {@code null} * @param separatorChars the characters used as the delimiters, {@code null} * splits on whitespace * @return an array of parsed Strings, {@code null} if null String input * @since 2.1 */ public static String[] splitPreserveAllTokens(final String str, final String separatorChars) { return splitWorker(str, separatorChars, -1, true); } /** *

* Splits the provided text into an array with a maximum length, separators * specified, preserving all tokens, including empty tokens created by adjacent * separators. *

* *

* The separator is not included in the returned String array. Adjacent * separators are treated as separators for empty tokens. Adjacent separators * are treated as one separator. *

* *

* A {@code null} input String returns {@code null}. A {@code null} * separatorChars splits on whitespace. *

* *

* If more than {@code max} delimited substrings are found, the last returned * string includes all characters after the first {@code max - 1} returned * strings (including separator characters). *

* *
	 * StringUtils.splitPreserveAllTokens(null, *, *)            = null
	 * StringUtils.splitPreserveAllTokens("", *, *)              = []
	 * StringUtils.splitPreserveAllTokens("ab de fg", null, 0)   = ["ab", "de", "fg"]
	 * StringUtils.splitPreserveAllTokens("ab   de fg", null, 0) = ["ab", "", "", "de", "fg"]
	 * StringUtils.splitPreserveAllTokens("ab:cd:ef", ":", 0)    = ["ab", "cd", "ef"]
	 * StringUtils.splitPreserveAllTokens("ab:cd:ef", ":", 2)    = ["ab", "cd:ef"]
	 * StringUtils.splitPreserveAllTokens("ab   de fg", null, 2) = ["ab", "  de fg"]
	 * StringUtils.splitPreserveAllTokens("ab   de fg", null, 3) = ["ab", "", " de fg"]
	 * StringUtils.splitPreserveAllTokens("ab   de fg", null, 4) = ["ab", "", "", "de fg"]
	 * 
* * @param str the String to parse, may be {@code null} * @param separatorChars the characters used as the delimiters, {@code null} * splits on whitespace * @param max the maximum number of elements to include in the array. * A zero or negative value implies no limit * @return an array of parsed Strings, {@code null} if null String input * @since 2.1 */ public static String[] splitPreserveAllTokens(final String str, final String separatorChars, final int max) { return splitWorker(str, separatorChars, max, true); } /** * Performs the logic for the {@code split} and {@code splitPreserveAllTokens} * methods that do not return a maximum array length. * * @param str the String to parse, may be {@code null} * @param separatorChar the separate character * @param preserveAllTokens if {@code true}, adjacent separators are treated as * empty token separators; if {@code false}, adjacent * separators are treated as one separator. * @return an array of parsed Strings, {@code null} if null String input */ private static String[] splitWorker(final String str, final char separatorChar, final boolean preserveAllTokens) { // Performance tuned for 2.0 (JDK1.4) if (str == null) { return null; } final int len = str.length(); if (len == 0) { return new String[0]; } final List list = new ArrayList<>(); int i = 0; int start = 0; boolean match = false; boolean lastMatch = false; while (i < len) { if (str.charAt(i) == separatorChar) { if (match || preserveAllTokens) { list.add(str.substring(start, i)); match = false; lastMatch = true; } start = ++i; continue; } lastMatch = false; match = true; i++; } if (match || preserveAllTokens && lastMatch) { list.add(str.substring(start, i)); } return list.toArray(new String[0]); } /** * Performs the logic for the {@code split} and {@code splitPreserveAllTokens} * methods that return a maximum array length. * * @param str the String to parse, may be {@code null} * @param separatorChars the separate character * @param max the maximum number of elements to include in the * array. A zero or negative value implies no limit. * @param preserveAllTokens if {@code true}, adjacent separators are treated as * empty token separators; if {@code false}, adjacent * separators are treated as one separator. * @return an array of parsed Strings, {@code null} if null String input */ private static String[] splitWorker(final String str, final String separatorChars, final int max, final boolean preserveAllTokens) { // Performance tuned for 2.0 (JDK1.4) // Direct code is quicker than StringTokenizer. // Also, StringTokenizer uses isSpace() not isWhitespace() if (str == null) { return null; } final int len = str.length(); if (len == 0) { return new String[0]; } final List list = new ArrayList<>(); int sizePlus1 = 1; int i = 0; int start = 0; boolean match = false; boolean lastMatch = false; if (separatorChars == null) { // Null separator means use whitespace while (i < len) { if (Character.isWhitespace(str.charAt(i))) { if (match || preserveAllTokens) { lastMatch = true; if (sizePlus1++ == max) { i = len; lastMatch = false; } list.add(str.substring(start, i)); match = false; } start = ++i; continue; } lastMatch = false; match = true; i++; } } else if (separatorChars.length() == 1) { // Optimise 1 character case final char sep = separatorChars.charAt(0); while (i < len) { if (str.charAt(i) == sep) { if (match || preserveAllTokens) { lastMatch = true; if (sizePlus1++ == max) { i = len; lastMatch = false; } list.add(str.substring(start, i)); match = false; } start = ++i; continue; } lastMatch = false; match = true; i++; } } else { // standard case while (i < len) { if (separatorChars.indexOf(str.charAt(i)) >= 0) { if (match || preserveAllTokens) { lastMatch = true; if (sizePlus1++ == max) { i = len; lastMatch = false; } list.add(str.substring(start, i)); match = false; } start = ++i; continue; } lastMatch = false; match = true; i++; } } if (match || preserveAllTokens && lastMatch) { list.add(str.substring(start, i)); } return list.toArray(new String[0]); } /** *

* Check if a CharSequence starts with a specified prefix. *

* *

* {@code null}s are handled without exceptions. Two {@code null} references are * considered to be equal. The comparison is case sensitive. *

* *
	 * StringUtils.startsWith(null, null)      = true
	 * StringUtils.startsWith(null, "abc")     = false
	 * StringUtils.startsWith("abcdef", null)  = false
	 * StringUtils.startsWith("abcdef", "abc") = true
	 * StringUtils.startsWith("ABCDEF", "abc") = false
	 * 
* * @see java.lang.String#startsWith(String) * @param str the CharSequence to check, may be null * @param prefix the prefix to find, may be null * @return {@code true} if the CharSequence starts with the prefix, case * sensitive, or both {@code null} * @since 2.4 * @since 3.0 Changed signature from startsWith(String, String) to * startsWith(CharSequence, CharSequence) */ public static boolean startsWith(final CharSequence str, final CharSequence prefix) { return startsWith(str, prefix, false); } /** *

* Check if a CharSequence starts with a specified prefix (optionally case * insensitive). *

* * @see java.lang.String#startsWith(String) * @param str the CharSequence to check, may be null * @param prefix the prefix to find, may be null * @param ignoreCase indicates whether the compare should ignore case (case * insensitive) or not. * @return {@code true} if the CharSequence starts with the prefix or both * {@code null} */ private static boolean startsWith(final CharSequence str, final CharSequence prefix, final boolean ignoreCase) { if (str == null || prefix == null) { return str == prefix; } // Get length once instead of twice in the unlikely case that it changes. final int preLen = prefix.length(); if (preLen > str.length()) { return false; } return CharSequenceUtils.regionMatches(str, ignoreCase, 0, prefix, 0, preLen); } /** *

* Check if a CharSequence starts with any of the provided case-sensitive * prefixes. *

* *
	 * StringUtils.startsWithAny(null, null)      = false
	 * StringUtils.startsWithAny(null, new String[] {"abc"})  = false
	 * StringUtils.startsWithAny("abcxyz", null)     = false
	 * StringUtils.startsWithAny("abcxyz", new String[] {""}) = true
	 * StringUtils.startsWithAny("abcxyz", new String[] {"abc"}) = true
	 * StringUtils.startsWithAny("abcxyz", new String[] {null, "xyz", "abc"}) = true
	 * StringUtils.startsWithAny("abcxyz", null, "xyz", "ABCX") = false
	 * StringUtils.startsWithAny("ABCXYZ", null, "xyz", "abc") = false
	 * 
* * @param sequence the CharSequence to check, may be null * @param searchStrings the case-sensitive CharSequence prefixes, may be empty * or contain {@code null} * @see StringUtils#startsWith(CharSequence, CharSequence) * @return {@code true} if the input {@code sequence} is {@code null} AND no * {@code searchStrings} are provided, or the input {@code sequence} * begins with any of the provided case-sensitive {@code searchStrings}. * @since 2.5 * @since 3.0 Changed signature from startsWithAny(String, String[]) to * startsWithAny(CharSequence, CharSequence...) */ public static boolean startsWithAny(final CharSequence sequence, final CharSequence... searchStrings) { if (isEmpty(sequence) || searchStrings.length == 0) { return false; } for (final CharSequence searchString : searchStrings) { if (startsWith(sequence, searchString)) { return true; } } return false; } /** *

* Case insensitive check if a CharSequence starts with a specified prefix. *

* *

* {@code null}s are handled without exceptions. Two {@code null} references are * considered to be equal. The comparison is case insensitive. *

* *
	 * StringUtils.startsWithIgnoreCase(null, null)      = true
	 * StringUtils.startsWithIgnoreCase(null, "abc")     = false
	 * StringUtils.startsWithIgnoreCase("abcdef", null)  = false
	 * StringUtils.startsWithIgnoreCase("abcdef", "abc") = true
	 * StringUtils.startsWithIgnoreCase("ABCDEF", "abc") = true
	 * 
* * @see java.lang.String#startsWith(String) * @param str the CharSequence to check, may be null * @param prefix the prefix to find, may be null * @return {@code true} if the CharSequence starts with the prefix, case * insensitive, or both {@code null} * @since 2.4 * @since 3.0 Changed signature from startsWithIgnoreCase(String, String) to * startsWithIgnoreCase(CharSequence, CharSequence) */ public static boolean startsWithIgnoreCase(final CharSequence str, final CharSequence prefix) { return startsWith(str, prefix, true); } /** *

* Strips whitespace from the start and end of a String. *

* *

* This is similar to {@link #trim(String)} but removes whitespace. Whitespace * is defined by {@link Character#isWhitespace(char)}. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.strip(null)     = null
	 * StringUtils.strip("")       = ""
	 * StringUtils.strip("   ")    = ""
	 * StringUtils.strip("abc")    = "abc"
	 * StringUtils.strip("  abc")  = "abc"
	 * StringUtils.strip("abc  ")  = "abc"
	 * StringUtils.strip(" abc ")  = "abc"
	 * StringUtils.strip(" ab c ") = "ab c"
	 * 
* * @param str the String to remove whitespace from, may be null * @return the stripped String, {@code null} if null String input */ public static String strip(final String str) { return strip(str, null); } /** *

* Strips any of a set of characters from the start and end of a String. This is * similar to {@link String#trim()} but allows the characters to be stripped to * be controlled. *

* *

* A {@code null} input String returns {@code null}. An empty string ("") input * returns the empty string. *

* *

* If the stripChars String is {@code null}, whitespace is stripped as defined * by {@link Character#isWhitespace(char)}. Alternatively use * {@link #strip(String)}. *

* *
	 * StringUtils.strip(null, *)          = null
	 * StringUtils.strip("", *)            = ""
	 * StringUtils.strip("abc", null)      = "abc"
	 * StringUtils.strip("  abc", null)    = "abc"
	 * StringUtils.strip("abc  ", null)    = "abc"
	 * StringUtils.strip(" abc ", null)    = "abc"
	 * StringUtils.strip("  abcyx", "xyz") = "  abc"
	 * 
* * @param str the String to remove characters from, may be null * @param stripChars the characters to remove, null treated as whitespace * @return the stripped String, {@code null} if null String input */ public static String strip(String str, final String stripChars) { str = stripStart(str, stripChars); return stripEnd(str, stripChars); } /** *

* Removes diacritics (~= accents) from a string. The case will not be altered. *

*

* For instance, 'à' will be replaced by 'a'. *

*

* Note that ligatures will be left as is. *

* *
	 * StringUtils.stripAccents(null)                = null
	 * StringUtils.stripAccents("")                  = ""
	 * StringUtils.stripAccents("control")           = "control"
	 * StringUtils.stripAccents("éclair")     = "eclair"
	 * 
* * @param input String to be stripped * @return input text with diacritics removed * * @since 3.0 */ // See also Lucene's ASCIIFoldingFilter (Lucene 2.9) that replaces accented // characters by their unaccented equivalent (and uncommitted bug fix: // https://issues.apache.org/jira/browse/LUCENE-1343?focusedCommentId=12858907&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#action_12858907). public static String stripAccents(final String input) { if (input == null) { return null; } final StringBuilder decomposed = new StringBuilder(Normalizer.normalize(input, Normalizer.Form.NFD)); convertRemainingAccentCharacters(decomposed); // Note that this doesn't correctly remove ligatures... return STRIP_ACCENTS_PATTERN.matcher(decomposed).replaceAll(EMPTY); } /** *

* Strips whitespace from the start and end of every String in an array. * Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *

* A new array is returned each time, except for length zero. A {@code null} * array will return {@code null}. An empty array will return itself. A * {@code null} array entry will be ignored. *

* *
	 * StringUtils.stripAll(null)             = null
	 * StringUtils.stripAll([])               = []
	 * StringUtils.stripAll(["abc", "  abc"]) = ["abc", "abc"]
	 * StringUtils.stripAll(["abc  ", null])  = ["abc", null]
	 * 
* * @param strs the array to remove whitespace from, may be null * @return the stripped Strings, {@code null} if null array input */ public static String[] stripAll(final String... strs) { return stripAll(strs, null); } /** *

* Strips any of a set of characters from the start and end of every String in * an array. *

*

* Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *

* A new array is returned each time, except for length zero. A {@code null} * array will return {@code null}. An empty array will return itself. A * {@code null} array entry will be ignored. A {@code null} stripChars will * strip whitespace as defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.stripAll(null, *)                = null
	 * StringUtils.stripAll([], *)                  = []
	 * StringUtils.stripAll(["abc", "  abc"], null) = ["abc", "abc"]
	 * StringUtils.stripAll(["abc  ", null], null)  = ["abc", null]
	 * StringUtils.stripAll(["abc  ", null], "yz")  = ["abc  ", null]
	 * StringUtils.stripAll(["yabcz", null], "yz")  = ["abc", null]
	 * 
* * @param strs the array to remove characters from, may be null * @param stripChars the characters to remove, null treated as whitespace * @return the stripped Strings, {@code null} if null array input */ public static String[] stripAll(final String[] strs, final String stripChars) { final int strsLen = strs.length; if (strsLen == 0) { return strs; } final String[] newArr = new String[strsLen]; for (int i = 0; i < strsLen; i++) { newArr[i] = strip(strs[i], stripChars); } return newArr; } /** *

* Strips any of a set of characters from the end of a String. *

* *

* A {@code null} input String returns {@code null}. An empty string ("") input * returns the empty string. *

* *

* If the stripChars String is {@code null}, whitespace is stripped as defined * by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.stripEnd(null, *)          = null
	 * StringUtils.stripEnd("", *)            = ""
	 * StringUtils.stripEnd("abc", "")        = "abc"
	 * StringUtils.stripEnd("abc", null)      = "abc"
	 * StringUtils.stripEnd("  abc", null)    = "  abc"
	 * StringUtils.stripEnd("abc  ", null)    = "abc"
	 * StringUtils.stripEnd(" abc ", null)    = " abc"
	 * StringUtils.stripEnd("  abcyx", "xyz") = "  abc"
	 * StringUtils.stripEnd("120.00", ".0")   = "12"
	 * 
* * @param str the String to remove characters from, may be null * @param stripChars the set of characters to remove, null treated as whitespace * @return the stripped String, {@code null} if null String input */ public static String stripEnd(final String str, final String stripChars) { int end = length(str); if (end == 0) { return str; } if (stripChars == null) { while (end != 0 && Character.isWhitespace(str.charAt(end - 1))) { end--; } } else if (stripChars.isEmpty()) { return str; } else { while (end != 0 && stripChars.indexOf(str.charAt(end - 1)) != INDEX_NOT_FOUND) { end--; } } return str.substring(0, end); } /** *

* Strips any of a set of characters from the start of a String. *

* *

* A {@code null} input String returns {@code null}. An empty string ("") input * returns the empty string. *

* *

* If the stripChars String is {@code null}, whitespace is stripped as defined * by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.stripStart(null, *)          = null
	 * StringUtils.stripStart("", *)            = ""
	 * StringUtils.stripStart("abc", "")        = "abc"
	 * StringUtils.stripStart("abc", null)      = "abc"
	 * StringUtils.stripStart("  abc", null)    = "abc"
	 * StringUtils.stripStart("abc  ", null)    = "abc  "
	 * StringUtils.stripStart(" abc ", null)    = "abc "
	 * StringUtils.stripStart("yxabc  ", "xyz") = "abc  "
	 * 
* * @param str the String to remove characters from, may be null * @param stripChars the characters to remove, null treated as whitespace * @return the stripped String, {@code null} if null String input */ public static String stripStart(final String str, final String stripChars) { final int strLen = length(str); if (strLen == 0) { return str; } int start = 0; if (stripChars == null) { while (start != strLen && Character.isWhitespace(str.charAt(start))) { start++; } } else if (stripChars.isEmpty()) { return str; } else { while (start != strLen && stripChars.indexOf(str.charAt(start)) != INDEX_NOT_FOUND) { start++; } } return str.substring(start); } /** *

* Strips whitespace from the start and end of a String returning an empty * String if {@code null} input. *

* *

* This is similar to {@link #trimToEmpty(String)} but removes whitespace. * Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.stripToEmpty(null)     = ""
	 * StringUtils.stripToEmpty("")       = ""
	 * StringUtils.stripToEmpty("   ")    = ""
	 * StringUtils.stripToEmpty("abc")    = "abc"
	 * StringUtils.stripToEmpty("  abc")  = "abc"
	 * StringUtils.stripToEmpty("abc  ")  = "abc"
	 * StringUtils.stripToEmpty(" abc ")  = "abc"
	 * StringUtils.stripToEmpty(" ab c ") = "ab c"
	 * 
* * @param str the String to be stripped, may be null * @return the trimmed String, or an empty String if {@code null} input * @since 2.0 */ public static String stripToEmpty(final String str) { return str == null ? EMPTY : strip(str, null); } /** *

* Strips whitespace from the start and end of a String returning {@code null} * if the String is empty ("") after the strip. *

* *

* This is similar to {@link #trimToNull(String)} but removes whitespace. * Whitespace is defined by {@link Character#isWhitespace(char)}. *

* *
	 * StringUtils.stripToNull(null)     = null
	 * StringUtils.stripToNull("")       = null
	 * StringUtils.stripToNull("   ")    = null
	 * StringUtils.stripToNull("abc")    = "abc"
	 * StringUtils.stripToNull("  abc")  = "abc"
	 * StringUtils.stripToNull("abc  ")  = "abc"
	 * StringUtils.stripToNull(" abc ")  = "abc"
	 * StringUtils.stripToNull(" ab c ") = "ab c"
	 * 
* * @param str the String to be stripped, may be null * @return the stripped String, {@code null} if whitespace, empty or null String * input * @since 2.0 */ public static String stripToNull(String str) { if (str == null) { return null; } str = strip(str, null); return str.isEmpty() ? null : str; // NOSONARLINT str cannot be null here } /** *

* Gets a substring from the specified String avoiding exceptions. *

* *

* A negative start position can be used to start {@code n} characters from the * end of the String. *

* *

* A {@code null} String will return {@code null}. An empty ("") String will * return "". *

* *
	 * StringUtils.substring(null, *)   = null
	 * StringUtils.substring("", *)     = ""
	 * StringUtils.substring("abc", 0)  = "abc"
	 * StringUtils.substring("abc", 2)  = "c"
	 * StringUtils.substring("abc", 4)  = ""
	 * StringUtils.substring("abc", -2) = "bc"
	 * StringUtils.substring("abc", -4) = "abc"
	 * 
* * @param str the String to get the substring from, may be null * @param start the position to start from, negative means count back from the * end of the String by this many characters * @return substring from start position, {@code null} if null String input */ public static String substring(final String str, int start) { if (str == null) { return null; } // handle negatives, which means last n characters if (start < 0) { start = str.length() + start; // remember start is negative } if (start < 0) { start = 0; } if (start > str.length()) { return EMPTY; } return str.substring(start); } /** *

* Gets a substring from the specified String avoiding exceptions. *

* *

* A negative start position can be used to start/end {@code n} characters from * the end of the String. *

* *

* The returned substring starts with the character in the {@code start} * position and ends before the {@code end} position. All position counting is * zero-based -- i.e., to start at the beginning of the string use * {@code start = 0}. Negative start and end positions can be used to specify * offsets relative to the end of the String. *

* *

* If {@code start} is not strictly to the left of {@code end}, "" is returned. *

* *
	 * StringUtils.substring(null, *, *)    = null
	 * StringUtils.substring("", * ,  *)    = "";
	 * StringUtils.substring("abc", 0, 2)   = "ab"
	 * StringUtils.substring("abc", 2, 0)   = ""
	 * StringUtils.substring("abc", 2, 4)   = "c"
	 * StringUtils.substring("abc", 4, 6)   = ""
	 * StringUtils.substring("abc", 2, 2)   = ""
	 * StringUtils.substring("abc", -2, -1) = "b"
	 * StringUtils.substring("abc", -4, 2)  = "ab"
	 * 
* * @param str the String to get the substring from, may be null * @param start the position to start from, negative means count back from the * end of the String by this many characters * @param end the position to end at (exclusive), negative means count back * from the end of the String by this many characters * @return substring from start position to end position, {@code null} if null * String input */ public static String substring(final String str, int start, int end) { if (str == null) { return null; } // handle negatives if (end < 0) { end = str.length() + end; // remember end is negative } if (start < 0) { start = str.length() + start; // remember start is negative } // check length next if (end > str.length()) { end = str.length(); } // if start is greater than end, return "" if (start > end) { return EMPTY; } if (start < 0) { start = 0; } if (end < 0) { end = 0; } return str.substring(start, end); } /** *

* Gets the substring after the first occurrence of a separator. The separator * is not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. * *

* If nothing is found, the empty string is returned. *

* *
	 * StringUtils.substringAfter(null, *)      = null
	 * StringUtils.substringAfter("", *)        = ""
	 * StringUtils.substringAfter("abc", 'a')   = "bc"
	 * StringUtils.substringAfter("abcba", 'b') = "cba"
	 * StringUtils.substringAfter("abc", 'c')   = ""
	 * StringUtils.substringAfter("abc", 'd')   = ""
	 * StringUtils.substringAfter(" abc", 32)   = "abc"
	 * 
* * @param str the String to get a substring from, may be null * @param separator the character to search. * @return the substring after the first occurrence of the separator, * {@code null} if null String input * @since 3.11 */ public static String substringAfter(final String str, final int separator) { if (isEmpty(str)) { return str; } final int pos = str.indexOf(separator); if (pos == INDEX_NOT_FOUND) { return EMPTY; } return str.substring(pos + 1); } /** *

* Gets the substring after the first occurrence of a separator. The separator * is not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. A {@code null} separator will return the * empty string if the input string is not {@code null}. *

* *

* If nothing is found, the empty string is returned. *

* *
	 * StringUtils.substringAfter(null, *)      = null
	 * StringUtils.substringAfter("", *)        = ""
	 * StringUtils.substringAfter(*, null)      = ""
	 * StringUtils.substringAfter("abc", "a")   = "bc"
	 * StringUtils.substringAfter("abcba", "b") = "cba"
	 * StringUtils.substringAfter("abc", "c")   = ""
	 * StringUtils.substringAfter("abc", "d")   = ""
	 * StringUtils.substringAfter("abc", "")    = "abc"
	 * 
* * @param str the String to get a substring from, may be null * @param separator the String to search for, may be null * @return the substring after the first occurrence of the separator, * {@code null} if null String input * @since 2.0 */ public static String substringAfter(final String str, final String separator) { if (isEmpty(str)) { return str; } if (separator == null) { return EMPTY; } final int pos = str.indexOf(separator); if (pos == INDEX_NOT_FOUND) { return EMPTY; } return str.substring(pos + separator.length()); } /** *

* Gets the substring after the last occurrence of a separator. The separator is * not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. * *

* If nothing is found, the empty string is returned. *

* *
	 * StringUtils.substringAfterLast(null, *)      = null
	 * StringUtils.substringAfterLast("", *)        = ""
	 * StringUtils.substringAfterLast("abc", 'a')   = "bc"
	 * StringUtils.substringAfterLast(" bc", 32)    = "bc"
	 * StringUtils.substringAfterLast("abcba", 'b') = "a"
	 * StringUtils.substringAfterLast("abc", 'c')   = ""
	 * StringUtils.substringAfterLast("a", 'a')     = ""
	 * StringUtils.substringAfterLast("a", 'z')     = ""
	 * 
* * @param str the String to get a substring from, may be null * @param separator the String to search for, may be null * @return the substring after the last occurrence of the separator, * {@code null} if null String input * @since 3.11 */ public static String substringAfterLast(final String str, final int separator) { if (isEmpty(str)) { return str; } final int pos = str.lastIndexOf(separator); if (pos == INDEX_NOT_FOUND || pos == str.length() - 1) { return EMPTY; } return str.substring(pos + 1); } /** *

* Gets the substring after the last occurrence of a separator. The separator is * not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. An empty or {@code null} separator will * return the empty string if the input string is not {@code null}. *

* *

* If nothing is found, the empty string is returned. *

* *
	 * StringUtils.substringAfterLast(null, *)      = null
	 * StringUtils.substringAfterLast("", *)        = ""
	 * StringUtils.substringAfterLast(*, "")        = ""
	 * StringUtils.substringAfterLast(*, null)      = ""
	 * StringUtils.substringAfterLast("abc", "a")   = "bc"
	 * StringUtils.substringAfterLast("abcba", "b") = "a"
	 * StringUtils.substringAfterLast("abc", "c")   = ""
	 * StringUtils.substringAfterLast("a", "a")     = ""
	 * StringUtils.substringAfterLast("a", "z")     = ""
	 * 
* * @param str the String to get a substring from, may be null * @param separator the String to search for, may be null * @return the substring after the last occurrence of the separator, * {@code null} if null String input * @since 2.0 */ public static String substringAfterLast(final String str, final String separator) { if (isEmpty(str)) { return str; } if (isEmpty(separator)) { return EMPTY; } final int pos = str.lastIndexOf(separator); if (pos == INDEX_NOT_FOUND || pos == str.length() - separator.length()) { return EMPTY; } return str.substring(pos + separator.length()); } /** *

* Gets the substring before the first occurrence of a separator. The separator * is not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. *

* *

* If nothing is found, the string input is returned. *

* *
	 * StringUtils.substringBefore(null, *)      = null
	 * StringUtils.substringBefore("", *)        = ""
	 * StringUtils.substringBefore("abc", 'a')   = ""
	 * StringUtils.substringBefore("abcba", 'b') = "a"
	 * StringUtils.substringBefore("abc", 'c')   = "ab"
	 * StringUtils.substringBefore("abc", 'd')   = "abc"
	 * 
* * @param str the String to get a substring from, may be null * @param separator the String to search for, may be null * @return the substring before the first occurrence of the separator, * {@code null} if null String input * @since 3.12.0 */ public static String substringBefore(final String str, final int separator) { if (isEmpty(str)) { return str; } final int pos = str.indexOf(separator); if (pos == INDEX_NOT_FOUND) { return str; } return str.substring(0, pos); } /** *

* Gets the substring before the first occurrence of a separator. The separator * is not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. A {@code null} separator will return the * input string. *

* *

* If nothing is found, the string input is returned. *

* *
	 * StringUtils.substringBefore(null, *)      = null
	 * StringUtils.substringBefore("", *)        = ""
	 * StringUtils.substringBefore("abc", "a")   = ""
	 * StringUtils.substringBefore("abcba", "b") = "a"
	 * StringUtils.substringBefore("abc", "c")   = "ab"
	 * StringUtils.substringBefore("abc", "d")   = "abc"
	 * StringUtils.substringBefore("abc", "")    = ""
	 * StringUtils.substringBefore("abc", null)  = "abc"
	 * 
* * @param str the String to get a substring from, may be null * @param separator the String to search for, may be null * @return the substring before the first occurrence of the separator, * {@code null} if null String input * @since 2.0 */ public static String substringBefore(final String str, final String separator) { if (isEmpty(str) || separator == null) { return str; } if (separator.isEmpty()) { return EMPTY; } final int pos = str.indexOf(separator); if (pos == INDEX_NOT_FOUND) { return str; } return str.substring(0, pos); } /** *

* Gets the substring before the last occurrence of a separator. The separator * is not returned. *

* *

* A {@code null} string input will return {@code null}. An empty ("") string * input will return the empty string. An empty or {@code null} separator will * return the input string. *

* *

* If nothing is found, the string input is returned. *

* *
	 * StringUtils.substringBeforeLast(null, *)      = null
	 * StringUtils.substringBeforeLast("", *)        = ""
	 * StringUtils.substringBeforeLast("abcba", "b") = "abc"
	 * StringUtils.substringBeforeLast("abc", "c")   = "ab"
	 * StringUtils.substringBeforeLast("a", "a")     = ""
	 * StringUtils.substringBeforeLast("a", "z")     = "a"
	 * StringUtils.substringBeforeLast("a", null)    = "a"
	 * StringUtils.substringBeforeLast("a", "")      = "a"
	 * 
* * @param str the String to get a substring from, may be null * @param separator the String to search for, may be null * @return the substring before the last occurrence of the separator, * {@code null} if null String input * @since 2.0 */ public static String substringBeforeLast(final String str, final String separator) { if (isEmpty(str) || isEmpty(separator)) { return str; } final int pos = str.lastIndexOf(separator); if (pos == INDEX_NOT_FOUND) { return str; } return str.substring(0, pos); } /** *

* Gets the String that is nested in between two instances of the same String. *

* *

* A {@code null} input String returns {@code null}. A {@code null} tag returns * {@code null}. *

* *
	 * StringUtils.substringBetween(null, *)            = null
	 * StringUtils.substringBetween("", "")             = ""
	 * StringUtils.substringBetween("", "tag")          = null
	 * StringUtils.substringBetween("tagabctag", null)  = null
	 * StringUtils.substringBetween("tagabctag", "")    = ""
	 * StringUtils.substringBetween("tagabctag", "tag") = "abc"
	 * 
* * @param str the String containing the substring, may be null * @param tag the String before and after the substring, may be null * @return the substring, {@code null} if no match * @since 2.0 */ public static String substringBetween(final String str, final String tag) { return substringBetween(str, tag, tag); } /** *

* Gets the String that is nested in between two Strings. Only the first match * is returned. *

* *

* A {@code null} input String returns {@code null}. A {@code null} open/close * returns {@code null} (no match). An empty ("") open and close returns an * empty string. *

* *
	 * StringUtils.substringBetween("wx[b]yz", "[", "]") = "b"
	 * StringUtils.substringBetween(null, *, *)          = null
	 * StringUtils.substringBetween(*, null, *)          = null
	 * StringUtils.substringBetween(*, *, null)          = null
	 * StringUtils.substringBetween("", "", "")          = ""
	 * StringUtils.substringBetween("", "", "]")         = null
	 * StringUtils.substringBetween("", "[", "]")        = null
	 * StringUtils.substringBetween("yabcz", "", "")     = ""
	 * StringUtils.substringBetween("yabcz", "y", "z")   = "abc"
	 * StringUtils.substringBetween("yabczyabcz", "y", "z")   = "abc"
	 * 
* * @param str the String containing the substring, may be null * @param open the String before the substring, may be null * @param close the String after the substring, may be null * @return the substring, {@code null} if no match * @since 2.0 */ public static String substringBetween(final String str, final String open, final String close) { if (!ObjectUtils.allNotNull(str, open, close)) { return null; } final int start = str.indexOf(open); if (start != INDEX_NOT_FOUND) { final int end = str.indexOf(close, start + open.length()); if (end != INDEX_NOT_FOUND) { return str.substring(start + open.length(), end); } } return null; } /** *

* Searches a String for substrings delimited by a start and end tag, returning * all matching substrings in an array. *

* *

* A {@code null} input String returns {@code null}. A {@code null} open/close * returns {@code null} (no match). An empty ("") open/close returns * {@code null} (no match). *

* *
	 * StringUtils.substringsBetween("[a][b][c]", "[", "]") = ["a","b","c"]
	 * StringUtils.substringsBetween(null, *, *)            = null
	 * StringUtils.substringsBetween(*, null, *)            = null
	 * StringUtils.substringsBetween(*, *, null)            = null
	 * StringUtils.substringsBetween("", "[", "]")          = []
	 * 
* * @param str the String containing the substrings, null returns null, empty * returns empty * @param open the String identifying the start of the substring, empty returns * null * @param close the String identifying the end of the substring, empty returns * null * @return a String Array of substrings, or {@code null} if no match * @since 2.3 */ public static String[] substringsBetween(final String str, final String open, final String close) { if (str == null || isEmpty(open) || isEmpty(close)) { return null; } final int strLen = str.length(); if (strLen == 0) { return new String[0]; } final int closeLen = close.length(); final int openLen = open.length(); final List list = new ArrayList<>(); int pos = 0; while (pos < strLen - closeLen) { int start = str.indexOf(open, pos); if (start < 0) { break; } start += openLen; final int end = str.indexOf(close, start); if (end < 0) { break; } list.add(str.substring(start, end)); pos = end + closeLen; } if (list.isEmpty()) { return null; } return list.toArray(new String[0]); } /** *

* Swaps the case of a String changing upper and title case to lower case, and * lower case to upper case. *

* *
    *
  • Upper case character converts to Lower case
  • *
  • Title case character converts to Lower case
  • *
  • Lower case character converts to Upper case
  • *
* *

* For a word based algorithm, see * {@link org.apache.commons.lang3.text.WordUtils#swapCase(String)}. A * {@code null} input String returns {@code null}. *

* *
	 * StringUtils.swapCase(null)                 = null
	 * StringUtils.swapCase("")                   = ""
	 * StringUtils.swapCase("The dog has a BONE") = "tHE DOG HAS A bone"
	 * 
* *

* NOTE: This method changed in Lang version 2.0. It no longer performs a word * based algorithm. If you only use ASCII, you will notice no change. That * functionality is available in org.apache.commons.lang3.text.WordUtils. *

* * @param str the String to swap case, may be null * @return the changed String, {@code null} if null String input */ public static String swapCase(final String str) { if (isEmpty(str)) { return str; } final int strLen = str.length(); final int[] newCodePoints = new int[strLen]; // cannot be longer than the char array int outOffset = 0; for (int i = 0; i < strLen;) { final int oldCodepoint = str.codePointAt(i); final int newCodePoint; if (Character.isUpperCase(oldCodepoint) || Character.isTitleCase(oldCodepoint)) { newCodePoint = Character.toLowerCase(oldCodepoint); } else if (Character.isLowerCase(oldCodepoint)) { newCodePoint = Character.toUpperCase(oldCodepoint); } else { newCodePoint = oldCodepoint; } newCodePoints[outOffset++] = newCodePoint; i += Character.charCount(newCodePoint); } return new String(newCodePoints, 0, outOffset); } /** *

* Converts a {@code CharSequence} into an array of code points. *

* *

* Valid pairs of surrogate code units will be converted into a single * supplementary code point. Isolated surrogate code units (i.e. a high * surrogate not followed by a low surrogate or a low surrogate not preceded by * a high surrogate) will be returned as-is. *

* *
	 * StringUtils.toCodePoints(null)   =  null
	 * StringUtils.toCodePoints("")     =  []  // empty array
	 * 
* * @param cs the character sequence to convert * @return an array of code points * @since 3.6 */ public static int[] toCodePoints(final CharSequence cs) { if (cs == null) { return null; } if (cs.length() == 0) { return new int[0]; } final String s = cs.toString(); final int[] result = new int[s.codePointCount(0, s.length())]; int index = 0; for (int i = 0; i < result.length; i++) { result[i] = s.codePointAt(index); index += Character.charCount(result[i]); } return result; } /** * Converts a {@code byte[]} to a String using the specified character encoding. * * @param bytes the byte array to read from * @param charset the encoding to use, if null then use the platform default * @return a new String * @throws NullPointerException if {@code bytes} is null * @since 3.2 * @since 3.3 No longer throws {@link UnsupportedEncodingException}. */ public static String toEncodedString(final byte[] bytes, final Charset charset) { return new String(bytes, Charsets.toCharset(charset)); } /** * Converts the given source String as a lower-case using the * {@link Locale#ROOT} locale in a null-safe manner. * * @param source A source String or null. * @return the given source String as a lower-case using the {@link Locale#ROOT} * locale or null. * @since 3.10 */ public static String toRootLowerCase(final String source) { return source == null ? null : source.toLowerCase(Locale.ROOT); } /** * Converts the given source String as a upper-case using the * {@link Locale#ROOT} locale in a null-safe manner. * * @param source A source String or null. * @return the given source String as a upper-case using the {@link Locale#ROOT} * locale or null. * @since 3.10 */ public static String toRootUpperCase(final String source) { return source == null ? null : source.toUpperCase(Locale.ROOT); } /** * Converts a {@code byte[]} to a String using the specified character encoding. * * @param bytes the byte array to read from * @param charsetName the encoding to use, if null then use the platform default * @return a new String * @throws UnsupportedEncodingException If the named charset is not supported * @throws NullPointerException if the input is null * @deprecated use {@link StringUtils#toEncodedString(byte[], Charset)} instead * of String constants in your code * @since 3.1 */ @Deprecated public static String toString(final byte[] bytes, final String charsetName) throws UnsupportedEncodingException { return new String(bytes, Charsets.toCharset(charsetName)); } private static String toStringOrEmpty(final Object obj) { return Objects.toString(obj, EMPTY); } /** *

* Removes control characters (char <= 32) from both ends of this String, * handling {@code null} by returning {@code null}. *

* *

* The String is trimmed using {@link String#trim()}. Trim removes start and end * characters <= 32. To strip whitespace use {@link #strip(String)}. *

* *

* To trim your choice of characters, use the {@link #strip(String, String)} * methods. *

* *
	 * StringUtils.trim(null)          = null
	 * StringUtils.trim("")            = ""
	 * StringUtils.trim("     ")       = ""
	 * StringUtils.trim("abc")         = "abc"
	 * StringUtils.trim("    abc    ") = "abc"
	 * 
* * @param str the String to be trimmed, may be null * @return the trimmed string, {@code null} if null String input */ public static String trim(final String str) { return str == null ? null : str.trim(); } /** *

* Removes control characters (char <= 32) from both ends of this String * returning an empty String ("") if the String is empty ("") after the trim or * if it is {@code null}. * *

* The String is trimmed using {@link String#trim()}. Trim removes start and end * characters <= 32. To strip whitespace use {@link #stripToEmpty(String)}. *

* *
	 * StringUtils.trimToEmpty(null)          = ""
	 * StringUtils.trimToEmpty("")            = ""
	 * StringUtils.trimToEmpty("     ")       = ""
	 * StringUtils.trimToEmpty("abc")         = "abc"
	 * StringUtils.trimToEmpty("    abc    ") = "abc"
	 * 
* * @param str the String to be trimmed, may be null * @return the trimmed String, or an empty String if {@code null} input * @since 2.0 */ public static String trimToEmpty(final String str) { return str == null ? EMPTY : str.trim(); } /** *

* Removes control characters (char <= 32) from both ends of this String * returning {@code null} if the String is empty ("") after the trim or if it is * {@code null}. * *

* The String is trimmed using {@link String#trim()}. Trim removes start and end * characters <= 32. To strip whitespace use {@link #stripToNull(String)}. *

* *
	 * StringUtils.trimToNull(null)          = null
	 * StringUtils.trimToNull("")            = null
	 * StringUtils.trimToNull("     ")       = null
	 * StringUtils.trimToNull("abc")         = "abc"
	 * StringUtils.trimToNull("    abc    ") = "abc"
	 * 
* * @param str the String to be trimmed, may be null * @return the trimmed String, {@code null} if only chars <= 32, empty or * null String input * @since 2.0 */ public static String trimToNull(final String str) { final String ts = trim(str); return isEmpty(ts) ? null : ts; } /** *

* Truncates a String. This will turn "Now is the time for all good men" into * "Now is the time for". *

* *

* Specifically: *

*
    *
  • If {@code str} is less than {@code maxWidth} characters long, return * it.
  • *
  • Else truncate it to {@code substring(str, 0, maxWidth)}.
  • *
  • If {@code maxWidth} is less than {@code 0}, throw an * {@code IllegalArgumentException}.
  • *
  • In no case will it return a String of length greater than * {@code maxWidth}.
  • *
* *
	 * StringUtils.truncate(null, 0)       = null
	 * StringUtils.truncate(null, 2)       = null
	 * StringUtils.truncate("", 4)         = ""
	 * StringUtils.truncate("abcdefg", 4)  = "abcd"
	 * StringUtils.truncate("abcdefg", 6)  = "abcdef"
	 * StringUtils.truncate("abcdefg", 7)  = "abcdefg"
	 * StringUtils.truncate("abcdefg", 8)  = "abcdefg"
	 * StringUtils.truncate("abcdefg", -1) = throws an IllegalArgumentException
	 * 
* * @param str the String to truncate, may be null * @param maxWidth maximum length of result String, must be positive * @return truncated String, {@code null} if null String input * @throws IllegalArgumentException If {@code maxWidth} is less than {@code 0} * @since 3.5 */ public static String truncate(final String str, final int maxWidth) { return truncate(str, 0, maxWidth); } /** *

* Truncates a String. This will turn "Now is the time for all good men" into * "is the time for all". *

* *

* Works like {@code truncate(String, int)}, but allows you to specify a "left * edge" offset. * *

* Specifically: *

*
    *
  • If {@code str} is less than {@code maxWidth} characters long, return * it.
  • *
  • Else truncate it to {@code substring(str, offset, maxWidth)}.
  • *
  • If {@code maxWidth} is less than {@code 0}, throw an * {@code IllegalArgumentException}.
  • *
  • If {@code offset} is less than {@code 0}, throw an * {@code IllegalArgumentException}.
  • *
  • In no case will it return a String of length greater than * {@code maxWidth}.
  • *
* *
	 * StringUtils.truncate(null, 0, 0) = null
	 * StringUtils.truncate(null, 2, 4) = null
	 * StringUtils.truncate("", 0, 10) = ""
	 * StringUtils.truncate("", 2, 10) = ""
	 * StringUtils.truncate("abcdefghij", 0, 3) = "abc"
	 * StringUtils.truncate("abcdefghij", 5, 6) = "fghij"
	 * StringUtils.truncate("raspberry peach", 10, 15) = "peach"
	 * StringUtils.truncate("abcdefghijklmno", 0, 10) = "abcdefghij"
	 * StringUtils.truncate("abcdefghijklmno", -1, 10) = throws an IllegalArgumentException
	 * StringUtils.truncate("abcdefghijklmno", Integer.MIN_VALUE, 10) = throws an IllegalArgumentException
	 * StringUtils.truncate("abcdefghijklmno", Integer.MIN_VALUE, Integer.MAX_VALUE) = throws an IllegalArgumentException
	 * StringUtils.truncate("abcdefghijklmno", 0, Integer.MAX_VALUE) = "abcdefghijklmno"
	 * StringUtils.truncate("abcdefghijklmno", 1, 10) = "bcdefghijk"
	 * StringUtils.truncate("abcdefghijklmno", 2, 10) = "cdefghijkl"
	 * StringUtils.truncate("abcdefghijklmno", 3, 10) = "defghijklm"
	 * StringUtils.truncate("abcdefghijklmno", 4, 10) = "efghijklmn"
	 * StringUtils.truncate("abcdefghijklmno", 5, 10) = "fghijklmno"
	 * StringUtils.truncate("abcdefghijklmno", 5, 5) = "fghij"
	 * StringUtils.truncate("abcdefghijklmno", 5, 3) = "fgh"
	 * StringUtils.truncate("abcdefghijklmno", 10, 3) = "klm"
	 * StringUtils.truncate("abcdefghijklmno", 10, Integer.MAX_VALUE) = "klmno"
	 * StringUtils.truncate("abcdefghijklmno", 13, 1) = "n"
	 * StringUtils.truncate("abcdefghijklmno", 13, Integer.MAX_VALUE) = "no"
	 * StringUtils.truncate("abcdefghijklmno", 14, 1) = "o"
	 * StringUtils.truncate("abcdefghijklmno", 14, Integer.MAX_VALUE) = "o"
	 * StringUtils.truncate("abcdefghijklmno", 15, 1) = ""
	 * StringUtils.truncate("abcdefghijklmno", 15, Integer.MAX_VALUE) = ""
	 * StringUtils.truncate("abcdefghijklmno", Integer.MAX_VALUE, Integer.MAX_VALUE) = ""
	 * StringUtils.truncate("abcdefghij", 3, -1) = throws an IllegalArgumentException
	 * StringUtils.truncate("abcdefghij", -2, 4) = throws an IllegalArgumentException
	 * 
* * @param str the String to truncate, may be null * @param offset left edge of source String * @param maxWidth maximum length of result String, must be positive * @return truncated String, {@code null} if null String input * @throws IllegalArgumentException If {@code offset} or {@code maxWidth} is * less than {@code 0} * @since 3.5 */ public static String truncate(final String str, final int offset, final int maxWidth) { if (offset < 0) { throw new IllegalArgumentException("offset cannot be negative"); } if (maxWidth < 0) { throw new IllegalArgumentException("maxWith cannot be negative"); } if (str == null) { return null; } if (offset > str.length()) { return EMPTY; } if (str.length() > maxWidth) { final int ix = Math.min(offset + maxWidth, str.length()); return str.substring(offset, ix); } return str.substring(offset); } /** *

* Uncapitalizes a String, changing the first character to lower case as per * {@link Character#toLowerCase(int)}. No other characters are changed. *

* *

* For a word based algorithm, see * {@link org.apache.commons.lang3.text.WordUtils#uncapitalize(String)}. A * {@code null} input String returns {@code null}. *

* *
	 * StringUtils.uncapitalize(null)  = null
	 * StringUtils.uncapitalize("")    = ""
	 * StringUtils.uncapitalize("cat") = "cat"
	 * StringUtils.uncapitalize("Cat") = "cat"
	 * StringUtils.uncapitalize("CAT") = "cAT"
	 * 
* * @param str the String to uncapitalize, may be null * @return the uncapitalized String, {@code null} if null String input * @see org.apache.commons.lang3.text.WordUtils#uncapitalize(String) * @see #capitalize(String) * @since 2.0 */ public static String uncapitalize(final String str) { final int strLen = length(str); if (strLen == 0) { return str; } final int firstCodepoint = str.codePointAt(0); final int newCodePoint = Character.toLowerCase(firstCodepoint); if (firstCodepoint == newCodePoint) { // already capitalized return str; } final int[] newCodePoints = new int[strLen]; // cannot be longer than the char array int outOffset = 0; newCodePoints[outOffset++] = newCodePoint; // copy the first codepoint for (int inOffset = Character.charCount(firstCodepoint); inOffset < strLen;) { final int codepoint = str.codePointAt(inOffset); newCodePoints[outOffset++] = codepoint; // copy the remaining ones inOffset += Character.charCount(codepoint); } return new String(newCodePoints, 0, outOffset); } /** *

* Unwraps a given string from a character. *

* *
	 * StringUtils.unwrap(null, null)         = null
	 * StringUtils.unwrap(null, '\0')         = null
	 * StringUtils.unwrap(null, '1')          = null
	 * StringUtils.unwrap("a", 'a')           = "a"
	 * StringUtils.unwrap("aa", 'a')           = ""
	 * StringUtils.unwrap("\'abc\'", '\'')    = "abc"
	 * StringUtils.unwrap("AABabcBAA", 'A')   = "ABabcBA"
	 * StringUtils.unwrap("A", '#')           = "A"
	 * StringUtils.unwrap("#A", '#')          = "#A"
	 * StringUtils.unwrap("A#", '#')          = "A#"
	 * 
* * @param str the String to be unwrapped, can be null * @param wrapChar the character used to unwrap * @return unwrapped String or the original string if it is not quoted properly * with the wrapChar * @since 3.6 */ public static String unwrap(final String str, final char wrapChar) { if (isEmpty(str) || wrapChar == CharUtils.NUL || str.length() == 1) { return str; } if (str.charAt(0) == wrapChar && str.charAt(str.length() - 1) == wrapChar) { final int startIndex = 0; final int endIndex = str.length() - 1; return str.substring(startIndex + 1, endIndex); } return str; } /** *

* Unwraps a given string from anther string. *

* *
	 * StringUtils.unwrap(null, null)         = null
	 * StringUtils.unwrap(null, "")           = null
	 * StringUtils.unwrap(null, "1")          = null
	 * StringUtils.unwrap("a", "a")           = "a"
	 * StringUtils.unwrap("aa", "a")          = ""
	 * StringUtils.unwrap("\'abc\'", "\'")    = "abc"
	 * StringUtils.unwrap("\"abc\"", "\"")    = "abc"
	 * StringUtils.unwrap("AABabcBAA", "AA")  = "BabcB"
	 * StringUtils.unwrap("A", "#")           = "A"
	 * StringUtils.unwrap("#A", "#")          = "#A"
	 * StringUtils.unwrap("A#", "#")          = "A#"
	 * 
* * @param str the String to be unwrapped, can be null * @param wrapToken the String used to unwrap * @return unwrapped String or the original string if it is not quoted properly * with the wrapToken * @since 3.6 */ public static String unwrap(final String str, final String wrapToken) { if (isEmpty(str) || isEmpty(wrapToken) || str.length() < 2 * wrapToken.length()) { return str; } if (startsWith(str, wrapToken) && endsWith(str, wrapToken)) { final int startIndex = str.indexOf(wrapToken); final int endIndex = str.lastIndexOf(wrapToken); final int wrapLength = wrapToken.length(); if (startIndex != -1 && endIndex != -1) { return str.substring(startIndex + wrapLength, endIndex); } } return str; } /** *

* Converts a String to upper case as per {@link String#toUpperCase()}. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.upperCase(null)  = null
	 * StringUtils.upperCase("")    = ""
	 * StringUtils.upperCase("aBc") = "ABC"
	 * 
* *

* Note: As described in the documentation for * {@link String#toUpperCase()}, the result of this method is affected by the * current locale. For platform-independent case transformations, the method * {@link #lowerCase(String, Locale)} should be used with a specific locale * (e.g. {@link Locale#ENGLISH}). *

* * @param str the String to upper case, may be null * @return the upper cased String, {@code null} if null String input */ public static String upperCase(final String str) { if (str == null) { return null; } return str.toUpperCase(); } /** *

* Converts a String to upper case as per {@link String#toUpperCase(Locale)}. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.upperCase(null, Locale.ENGLISH)  = null
	 * StringUtils.upperCase("", Locale.ENGLISH)    = ""
	 * StringUtils.upperCase("aBc", Locale.ENGLISH) = "ABC"
	 * 
* * @param str the String to upper case, may be null * @param locale the locale that defines the case transformation rules, must not * be null * @return the upper cased String, {@code null} if null String input * @since 2.5 */ public static String upperCase(final String str, final Locale locale) { if (str == null) { return null; } return str.toUpperCase(LocaleUtils.toLocale(locale)); } /** * Returns the string representation of the {@code char} array or null. * * @param value the character array. * @return a String or null * @see String#valueOf(char[]) * @since 3.9 */ public static String valueOf(final char[] value) { return value == null ? null : String.valueOf(value); } /** *

* Wraps a string with a char. *

* *
	 * StringUtils.wrap(null, *)        = null
	 * StringUtils.wrap("", *)          = ""
	 * StringUtils.wrap("ab", '\0')     = "ab"
	 * StringUtils.wrap("ab", 'x')      = "xabx"
	 * StringUtils.wrap("ab", '\'')     = "'ab'"
	 * StringUtils.wrap("\"ab\"", '\"') = "\"\"ab\"\""
	 * 
* * @param str the string to be wrapped, may be {@code null} * @param wrapWith the char that will wrap {@code str} * @return the wrapped string, or {@code null} if {@code str==null} * @since 3.4 */ public static String wrap(final String str, final char wrapWith) { if (isEmpty(str) || wrapWith == CharUtils.NUL) { return str; } return wrapWith + str + wrapWith; } /** *

* Wraps a String with another String. *

* *

* A {@code null} input String returns {@code null}. *

* *
	 * StringUtils.wrap(null, *)         = null
	 * StringUtils.wrap("", *)           = ""
	 * StringUtils.wrap("ab", null)      = "ab"
	 * StringUtils.wrap("ab", "x")       = "xabx"
	 * StringUtils.wrap("ab", "\"")      = "\"ab\""
	 * StringUtils.wrap("\"ab\"", "\"")  = "\"\"ab\"\""
	 * StringUtils.wrap("ab", "'")       = "'ab'"
	 * StringUtils.wrap("'abcd'", "'")   = "''abcd''"
	 * StringUtils.wrap("\"abcd\"", "'") = "'\"abcd\"'"
	 * StringUtils.wrap("'abcd'", "\"")  = "\"'abcd'\""
	 * 
* * @param str the String to be wrapper, may be null * @param wrapWith the String that will wrap str * @return wrapped String, {@code null} if null String input * @since 3.4 */ public static String wrap(final String str, final String wrapWith) { if (isEmpty(str) || isEmpty(wrapWith)) { return str; } return wrapWith.concat(str).concat(wrapWith); } /** *

* Wraps a string with a char if that char is missing from the start or end of * the given string. *

* *

* A new {@code String} will not be created if {@code str} is already wrapped. *

* *
	 * StringUtils.wrapIfMissing(null, *)        = null
	 * StringUtils.wrapIfMissing("", *)          = ""
	 * StringUtils.wrapIfMissing("ab", '\0')     = "ab"
	 * StringUtils.wrapIfMissing("ab", 'x')      = "xabx"
	 * StringUtils.wrapIfMissing("ab", '\'')     = "'ab'"
	 * StringUtils.wrapIfMissing("\"ab\"", '\"') = "\"ab\""
	 * StringUtils.wrapIfMissing("/", '/')  = "/"
	 * StringUtils.wrapIfMissing("a/b/c", '/')  = "/a/b/c/"
	 * StringUtils.wrapIfMissing("/a/b/c", '/')  = "/a/b/c/"
	 * StringUtils.wrapIfMissing("a/b/c/", '/')  = "/a/b/c/"
	 * 
* * @param str the string to be wrapped, may be {@code null} * @param wrapWith the char that will wrap {@code str} * @return the wrapped string, or {@code null} if {@code str==null} * @since 3.5 */ public static String wrapIfMissing(final String str, final char wrapWith) { if (isEmpty(str) || wrapWith == CharUtils.NUL) { return str; } final boolean wrapStart = str.charAt(0) != wrapWith; final boolean wrapEnd = str.charAt(str.length() - 1) != wrapWith; if (!wrapStart && !wrapEnd) { return str; } final StringBuilder builder = new StringBuilder(str.length() + 2); if (wrapStart) { builder.append(wrapWith); } builder.append(str); if (wrapEnd) { builder.append(wrapWith); } return builder.toString(); } /** *

* Wraps a string with a string if that string is missing from the start or end * of the given string. *

* *

* A new {@code String} will not be created if {@code str} is already wrapped. *

* *
	 * StringUtils.wrapIfMissing(null, *)         = null
	 * StringUtils.wrapIfMissing("", *)           = ""
	 * StringUtils.wrapIfMissing("ab", null)      = "ab"
	 * StringUtils.wrapIfMissing("ab", "x")       = "xabx"
	 * StringUtils.wrapIfMissing("ab", "\"")      = "\"ab\""
	 * StringUtils.wrapIfMissing("\"ab\"", "\"")  = "\"ab\""
	 * StringUtils.wrapIfMissing("ab", "'")       = "'ab'"
	 * StringUtils.wrapIfMissing("'abcd'", "'")   = "'abcd'"
	 * StringUtils.wrapIfMissing("\"abcd\"", "'") = "'\"abcd\"'"
	 * StringUtils.wrapIfMissing("'abcd'", "\"")  = "\"'abcd'\""
	 * StringUtils.wrapIfMissing("/", "/")  = "/"
	 * StringUtils.wrapIfMissing("a/b/c", "/")  = "/a/b/c/"
	 * StringUtils.wrapIfMissing("/a/b/c", "/")  = "/a/b/c/"
	 * StringUtils.wrapIfMissing("a/b/c/", "/")  = "/a/b/c/"
	 * 
* * @param str the string to be wrapped, may be {@code null} * @param wrapWith the string that will wrap {@code str} * @return the wrapped string, or {@code null} if {@code str==null} * @since 3.5 */ public static String wrapIfMissing(final String str, final String wrapWith) { if (isEmpty(str) || isEmpty(wrapWith)) { return str; } final boolean wrapStart = !str.startsWith(wrapWith); final boolean wrapEnd = !str.endsWith(wrapWith); if (!wrapStart && !wrapEnd) { return str; } final StringBuilder builder = new StringBuilder(str.length() + wrapWith.length() + wrapWith.length()); if (wrapStart) { builder.append(wrapWith); } builder.append(str); if (wrapEnd) { builder.append(wrapWith); } return builder.toString(); } /** *

* {@code StringUtils} instances should NOT be constructed in standard * programming. Instead, the class should be used as * {@code StringUtils.trim(" foo ");}. *

* *

* This constructor is public to permit tools that require a JavaBean instance * to operate. *

*/ public StringUtils() { } }