651 lines
22 KiB
Java
651 lines
22 KiB
Java
/*
|
|
* Copyright (C) 2009 The Guava Authors
|
|
*
|
|
* Licensed 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 com.google.common.base;
|
|
|
|
import static com.google.common.base.Preconditions.checkArgument;
|
|
import static com.google.common.base.Preconditions.checkNotNull;
|
|
|
|
import java.util.ArrayList;
|
|
import java.util.Collections;
|
|
import java.util.Iterator;
|
|
import java.util.LinkedHashMap;
|
|
import java.util.List;
|
|
import java.util.Map;
|
|
import java.util.regex.Matcher;
|
|
import java.util.regex.Pattern;
|
|
|
|
import javax.annotation.CheckReturnValue;
|
|
|
|
import com.google.common.annotations.Beta;
|
|
import com.google.common.annotations.GwtCompatible;
|
|
import com.google.common.annotations.GwtIncompatible;
|
|
|
|
/**
|
|
* Extracts non-overlapping substrings from an input string, typically by
|
|
* recognizing appearances of a <i>separator</i> sequence. This separator can be
|
|
* specified as a single {@linkplain #on(char) character}, fixed
|
|
* {@linkplain #on(String) string}, {@linkplain #onPattern regular expression}
|
|
* or {@link #on(CharMatcher) CharMatcher} instance. Or, instead of using a
|
|
* separator at all, a splitter can extract adjacent substrings of a given
|
|
* {@linkplain #fixedLength fixed length}.
|
|
*
|
|
* <p>
|
|
* For example, this expression:
|
|
*
|
|
* <pre>
|
|
* {@code
|
|
*
|
|
* Splitter.on(',').split("foo,bar,qux")}
|
|
* </pre>
|
|
*
|
|
* ... produces an {@code Iterable} containing {@code "foo"}, {@code "bar"} and
|
|
* {@code "qux"}, in that order.
|
|
*
|
|
* <p>
|
|
* By default, {@code Splitter}'s behavior is simplistic and unassuming. The
|
|
* following expression:
|
|
*
|
|
* <pre>
|
|
* {@code
|
|
*
|
|
* Splitter.on(',').split(" foo,,, bar ,")}
|
|
* </pre>
|
|
*
|
|
* ... yields the substrings {@code [" foo", "", "", " bar ", ""]}. If this is
|
|
* not the desired behavior, use configuration methods to obtain a <i>new</i>
|
|
* splitter instance with modified behavior:
|
|
*
|
|
* <pre>
|
|
* {
|
|
* @code
|
|
*
|
|
* private static final Splitter MY_SPLITTER = Splitter.on(',').trimResults().omitEmptyStrings();
|
|
* }
|
|
* </pre>
|
|
*
|
|
* <p>
|
|
* Now {@code MY_SPLITTER.split("foo,,, bar ,")} returns just {@code ["foo",
|
|
* "bar"]}. Note that the order in which these configuration methods are called
|
|
* is never significant.
|
|
*
|
|
* <p>
|
|
* <b>Warning:</b> Splitter instances are immutable. Invoking a configuration
|
|
* method has no effect on the receiving instance; you must store and use the
|
|
* new splitter instance it returns instead.
|
|
*
|
|
* <pre>
|
|
* {
|
|
* @code
|
|
*
|
|
* // Do NOT do this
|
|
* Splitter splitter = Splitter.on('/');
|
|
* splitter.trimResults(); // does nothing!
|
|
* return splitter.split("wrong / wrong / wrong");
|
|
* }
|
|
* </pre>
|
|
*
|
|
* <p>
|
|
* For separator-based splitters that do not use {@code omitEmptyStrings}, an
|
|
* input string containing {@code n} occurrences of the separator naturally
|
|
* yields an iterable of size {@code n + 1}. So if the separator does not occur
|
|
* anywhere in the input, a single substring is returned containing the entire
|
|
* input. Consequently, all splitters split the empty string to {@code [""]}
|
|
* (note: even fixed-length splitters).
|
|
*
|
|
* <p>
|
|
* Splitter instances are thread-safe immutable, and are therefore safe to store
|
|
* as {@code static final} constants.
|
|
*
|
|
* <p>
|
|
* The {@link Joiner} class provides the inverse operation to splitting, but
|
|
* note that a round-trip between the two should be assumed to be lossy.
|
|
*
|
|
* <p>
|
|
* See the Guava User Guide article on <a href=
|
|
* "http://code.google.com/p/guava-libraries/wiki/StringsExplained#Splitter">
|
|
* {@code Splitter}</a>.
|
|
*
|
|
* @author Julien Silland
|
|
* @author Jesse Wilson
|
|
* @author Kevin Bourrillion
|
|
* @author Louis Wasserman
|
|
* @since 1.0
|
|
*/
|
|
@GwtCompatible(emulated = true)
|
|
public final class Splitter {
|
|
private final CharMatcher trimmer;
|
|
private final boolean omitEmptyStrings;
|
|
private final Strategy strategy;
|
|
private final int limit;
|
|
|
|
private Splitter(Strategy strategy) {
|
|
this(strategy, false, CharMatcher.NONE, Integer.MAX_VALUE);
|
|
}
|
|
|
|
private Splitter(Strategy strategy, boolean omitEmptyStrings, CharMatcher trimmer, int limit) {
|
|
this.strategy = strategy;
|
|
this.omitEmptyStrings = omitEmptyStrings;
|
|
this.trimmer = trimmer;
|
|
this.limit = limit;
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that uses the given single-character separator. For
|
|
* example, {@code Splitter.on(',').split("foo,,bar")} returns an iterable
|
|
* containing {@code ["foo", "", "bar"]}.
|
|
*
|
|
* @param separator the character to recognize as a separator
|
|
* @return a splitter, with default settings, that recognizes that separator
|
|
*/
|
|
public static Splitter on(char separator) {
|
|
return on(CharMatcher.is(separator));
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that considers any single character matched by the given
|
|
* {@code CharMatcher} to be a separator. For example, {@code
|
|
* Splitter.on(CharMatcher.anyOf(";,")).split("foo,;bar,quux")} returns an
|
|
* iterable containing {@code ["foo", "", "bar", "quux"]}.
|
|
*
|
|
* @param separatorMatcher a {@link CharMatcher} that determines whether a
|
|
* character is a separator
|
|
* @return a splitter, with default settings, that uses this matcher
|
|
*/
|
|
public static Splitter on(final CharMatcher separatorMatcher) {
|
|
checkNotNull(separatorMatcher);
|
|
|
|
return new Splitter(new Strategy() {
|
|
@Override
|
|
public SplittingIterator iterator(Splitter splitter, final CharSequence toSplit) {
|
|
return new SplittingIterator(splitter, toSplit) {
|
|
@Override
|
|
int separatorStart(int start) {
|
|
return separatorMatcher.indexIn(toSplit, start);
|
|
}
|
|
|
|
@Override
|
|
int separatorEnd(int separatorPosition) {
|
|
return separatorPosition + 1;
|
|
}
|
|
};
|
|
}
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that uses the given fixed string as a separator. For
|
|
* example, {@code Splitter.on(", ").split("foo, bar,baz")} returns an iterable
|
|
* containing {@code ["foo", "bar,baz"]}.
|
|
*
|
|
* @param separator the literal, nonempty string to recognize as a separator
|
|
* @return a splitter, with default settings, that recognizes that separator
|
|
*/
|
|
public static Splitter on(final String separator) {
|
|
checkArgument(separator.length() != 0, "The separator may not be the empty string.");
|
|
|
|
return new Splitter(new Strategy() {
|
|
@Override
|
|
public SplittingIterator iterator(Splitter splitter, CharSequence toSplit) {
|
|
return new SplittingIterator(splitter, toSplit) {
|
|
@Override
|
|
public int separatorStart(int start) {
|
|
int separatorLength = separator.length();
|
|
|
|
positions: for (int p = start, last = toSplit.length() - separatorLength; p <= last; p++) {
|
|
for (int i = 0; i < separatorLength; i++) {
|
|
if (toSplit.charAt(i + p) != separator.charAt(i)) {
|
|
continue positions;
|
|
}
|
|
}
|
|
return p;
|
|
}
|
|
return -1;
|
|
}
|
|
|
|
@Override
|
|
public int separatorEnd(int separatorPosition) {
|
|
return separatorPosition + separator.length();
|
|
}
|
|
};
|
|
}
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that considers any subsequence matching {@code
|
|
* pattern} to be a separator. For example, {@code
|
|
* Splitter.on(Pattern.compile("\r?\n")).split(entireFile)} splits a string into
|
|
* lines whether it uses DOS-style or UNIX-style line terminators.
|
|
*
|
|
* @param separatorPattern the pattern that determines whether a subsequence is
|
|
* a separator. This pattern may not match the empty
|
|
* string.
|
|
* @return a splitter, with default settings, that uses this pattern
|
|
* @throws IllegalArgumentException if {@code separatorPattern} matches the
|
|
* empty string
|
|
*/
|
|
@GwtIncompatible("java.util.regex")
|
|
public static Splitter on(final Pattern separatorPattern) {
|
|
checkNotNull(separatorPattern);
|
|
checkArgument(!separatorPattern.matcher("").matches(), "The pattern may not match the empty string: %s",
|
|
separatorPattern);
|
|
|
|
return new Splitter(new Strategy() {
|
|
@Override
|
|
public SplittingIterator iterator(final Splitter splitter, CharSequence toSplit) {
|
|
final Matcher matcher = separatorPattern.matcher(toSplit);
|
|
return new SplittingIterator(splitter, toSplit) {
|
|
@Override
|
|
public int separatorStart(int start) {
|
|
return matcher.find(start) ? matcher.start() : -1;
|
|
}
|
|
|
|
@Override
|
|
public int separatorEnd(int separatorPosition) {
|
|
return matcher.end();
|
|
}
|
|
};
|
|
}
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that considers any subsequence matching a given pattern
|
|
* (regular expression) to be a separator. For example, {@code
|
|
* Splitter.onPattern("\r?\n").split(entireFile)} splits a string into lines
|
|
* whether it uses DOS-style or UNIX-style line terminators. This is equivalent
|
|
* to {@code Splitter.on(Pattern.compile(pattern))}.
|
|
*
|
|
* @param separatorPattern the pattern that determines whether a subsequence is
|
|
* a separator. This pattern may not match the empty
|
|
* string.
|
|
* @return a splitter, with default settings, that uses this pattern
|
|
* @throws java.util.regex.PatternSyntaxException if {@code separatorPattern} is
|
|
* a malformed expression
|
|
* @throws IllegalArgumentException if {@code separatorPattern}
|
|
* matches the empty string
|
|
*/
|
|
@GwtIncompatible("java.util.regex")
|
|
public static Splitter onPattern(String separatorPattern) {
|
|
return on(Pattern.compile(separatorPattern));
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that divides strings into pieces of the given length. For
|
|
* example, {@code Splitter.fixedLength(2).split("abcde")} returns an iterable
|
|
* containing {@code ["ab", "cd", "e"]}. The last piece can be smaller than
|
|
* {@code length} but will never be empty.
|
|
*
|
|
* <p>
|
|
* <b>Exception:</b> for consistency with separator-based splitters, {@code
|
|
* split("")} does not yield an empty iterable, but an iterable containing
|
|
* {@code ""}. This is the only case in which {@code
|
|
* Iterables.size(split(input))} does not equal {@code
|
|
* IntMath.divide(input.length(), length, CEILING)}. To avoid this behavior, use
|
|
* {@code omitEmptyStrings}.
|
|
*
|
|
* @param length the desired length of pieces after splitting, a positive
|
|
* integer
|
|
* @return a splitter, with default settings, that can split into fixed sized
|
|
* pieces
|
|
* @throws IllegalArgumentException if {@code length} is zero or negative
|
|
*/
|
|
public static Splitter fixedLength(final int length) {
|
|
checkArgument(length > 0, "The length may not be less than 1");
|
|
|
|
return new Splitter(new Strategy() {
|
|
@Override
|
|
public SplittingIterator iterator(final Splitter splitter, CharSequence toSplit) {
|
|
return new SplittingIterator(splitter, toSplit) {
|
|
@Override
|
|
public int separatorStart(int start) {
|
|
int nextChunkStart = start + length;
|
|
return (nextChunkStart < toSplit.length() ? nextChunkStart : -1);
|
|
}
|
|
|
|
@Override
|
|
public int separatorEnd(int separatorPosition) {
|
|
return separatorPosition;
|
|
}
|
|
};
|
|
}
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that behaves equivalently to {@code this} splitter, but
|
|
* automatically omits empty strings from the results. For example, {@code
|
|
* Splitter.on(',').omitEmptyStrings().split(",a,,,b,c,,")} returns an iterable
|
|
* containing only {@code ["a", "b", "c"]}.
|
|
*
|
|
* <p>
|
|
* If either {@code trimResults} option is also specified when creating a
|
|
* splitter, that splitter always trims results first before checking for
|
|
* emptiness. So, for example, {@code
|
|
* Splitter.on(':').omitEmptyStrings().trimResults().split(": : : ")} returns an
|
|
* empty iterable.
|
|
*
|
|
* <p>
|
|
* Note that it is ordinarily not possible for {@link #split(CharSequence)} to
|
|
* return an empty iterable, but when using this option, it can (if the input
|
|
* sequence consists of nothing but separators).
|
|
*
|
|
* @return a splitter with the desired configuration
|
|
*/
|
|
@CheckReturnValue
|
|
public Splitter omitEmptyStrings() {
|
|
return new Splitter(strategy, true, trimmer, limit);
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that behaves equivalently to {@code this} splitter but
|
|
* stops splitting after it reaches the limit. The limit defines the maximum
|
|
* number of items returned by the iterator.
|
|
*
|
|
* <p>
|
|
* For example, {@code Splitter.on(',').limit(3).split("a,b,c,d")} returns an
|
|
* iterable containing {@code ["a", "b", "c,d"]}. When omitting empty strings,
|
|
* the omitted strings do no count. Hence,
|
|
* {@code Splitter.on(',').limit(3).omitEmptyStrings().split("a,,,b,,,c,d")}
|
|
* returns an iterable containing {@code ["a", "b", "c,d"}. When trim is
|
|
* requested, all entries, including the last are trimmed. Hence
|
|
* {@code Splitter.on(',').limit(3).trimResults().split(" a , b , c , d ")}
|
|
* results in @{code ["a", "b", "c , d"]}.
|
|
*
|
|
* @param limit the maximum number of items returns
|
|
* @return a splitter with the desired configuration
|
|
* @since 9.0
|
|
*/
|
|
@CheckReturnValue
|
|
public Splitter limit(int limit) {
|
|
checkArgument(limit > 0, "must be greater than zero: %s", limit);
|
|
return new Splitter(strategy, omitEmptyStrings, trimmer, limit);
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that behaves equivalently to {@code this} splitter, but
|
|
* automatically removes leading and trailing {@linkplain CharMatcher#WHITESPACE
|
|
* whitespace} from each returned substring; equivalent to
|
|
* {@code trimResults(CharMatcher.WHITESPACE)}. For example, {@code
|
|
* Splitter.on(',').trimResults().split(" a, b ,c ")} returns an iterable
|
|
* containing {@code ["a", "b", "c"]}.
|
|
*
|
|
* @return a splitter with the desired configuration
|
|
*/
|
|
@CheckReturnValue
|
|
public Splitter trimResults() {
|
|
return trimResults(CharMatcher.WHITESPACE);
|
|
}
|
|
|
|
/**
|
|
* Returns a splitter that behaves equivalently to {@code this} splitter, but
|
|
* removes all leading or trailing characters matching the given {@code
|
|
* CharMatcher} from each returned substring. For example, {@code
|
|
* Splitter.on(',').trimResults(CharMatcher.is('_')).split("_a ,_b_ ,c__")}
|
|
* returns an iterable containing {@code ["a ", "b_ ", "c"]}.
|
|
*
|
|
* @param trimmer a {@link CharMatcher} that determines whether a character
|
|
* should be removed from the beginning/end of a subsequence
|
|
* @return a splitter with the desired configuration
|
|
*/
|
|
// TODO(kevinb): throw if a trimmer was already specified!
|
|
@CheckReturnValue
|
|
public Splitter trimResults(CharMatcher trimmer) {
|
|
checkNotNull(trimmer);
|
|
return new Splitter(strategy, omitEmptyStrings, trimmer, limit);
|
|
}
|
|
|
|
/**
|
|
* Splits {@code sequence} into string components and makes them available
|
|
* through an {@link Iterator}, which may be lazily evaluated. If you want an
|
|
* eagerly computed {@link List}, use {@link #splitToList(CharSequence)}.
|
|
*
|
|
* @param sequence the sequence of characters to split
|
|
* @return an iteration over the segments split from the parameter.
|
|
*/
|
|
public Iterable<String> split(final CharSequence sequence) {
|
|
checkNotNull(sequence);
|
|
|
|
return new Iterable<String>() {
|
|
@Override
|
|
public Iterator<String> iterator() {
|
|
return splittingIterator(sequence);
|
|
}
|
|
|
|
@Override
|
|
public String toString() {
|
|
return Joiner.on(", ").appendTo(new StringBuilder().append('['), this).append(']').toString();
|
|
}
|
|
};
|
|
}
|
|
|
|
private Iterator<String> splittingIterator(CharSequence sequence) {
|
|
return strategy.iterator(this, sequence);
|
|
}
|
|
|
|
/**
|
|
* Splits {@code sequence} into string components and returns them as an
|
|
* immutable list. If you want an {@link Iterable} which may be lazily
|
|
* evaluated, use {@link #split(CharSequence)}.
|
|
*
|
|
* @param sequence the sequence of characters to split
|
|
* @return an immutable list of the segments split from the parameter
|
|
* @since 15.0
|
|
*/
|
|
@Beta
|
|
public List<String> splitToList(CharSequence sequence) {
|
|
checkNotNull(sequence);
|
|
|
|
Iterator<String> iterator = splittingIterator(sequence);
|
|
List<String> result = new ArrayList<String>();
|
|
|
|
while (iterator.hasNext()) {
|
|
result.add(iterator.next());
|
|
}
|
|
|
|
return Collections.unmodifiableList(result);
|
|
}
|
|
|
|
/**
|
|
* Returns a {@code MapSplitter} which splits entries based on this splitter,
|
|
* and splits entries into keys and values using the specified separator.
|
|
*
|
|
* @since 10.0
|
|
*/
|
|
@CheckReturnValue
|
|
@Beta
|
|
public MapSplitter withKeyValueSeparator(String separator) {
|
|
return withKeyValueSeparator(on(separator));
|
|
}
|
|
|
|
/**
|
|
* Returns a {@code MapSplitter} which splits entries based on this splitter,
|
|
* and splits entries into keys and values using the specified separator.
|
|
*
|
|
* @since 14.0
|
|
*/
|
|
@CheckReturnValue
|
|
@Beta
|
|
public MapSplitter withKeyValueSeparator(char separator) {
|
|
return withKeyValueSeparator(on(separator));
|
|
}
|
|
|
|
/**
|
|
* Returns a {@code MapSplitter} which splits entries based on this splitter,
|
|
* and splits entries into keys and values using the specified key-value
|
|
* splitter.
|
|
*
|
|
* @since 10.0
|
|
*/
|
|
@CheckReturnValue
|
|
@Beta
|
|
public MapSplitter withKeyValueSeparator(Splitter keyValueSplitter) {
|
|
return new MapSplitter(this, keyValueSplitter);
|
|
}
|
|
|
|
/**
|
|
* An object that splits strings into maps as {@code Splitter} splits iterables
|
|
* and lists. Like {@code Splitter}, it is thread-safe and immutable.
|
|
*
|
|
* @since 10.0
|
|
*/
|
|
@Beta
|
|
public static final class MapSplitter {
|
|
private static final String INVALID_ENTRY_MESSAGE = "Chunk [%s] is not a valid entry";
|
|
private final Splitter outerSplitter;
|
|
private final Splitter entrySplitter;
|
|
|
|
private MapSplitter(Splitter outerSplitter, Splitter entrySplitter) {
|
|
this.outerSplitter = outerSplitter; // only "this" is passed
|
|
this.entrySplitter = checkNotNull(entrySplitter);
|
|
}
|
|
|
|
/**
|
|
* Splits {@code sequence} into substrings, splits each substring into an entry,
|
|
* and returns an unmodifiable map with each of the entries. For example, <code>
|
|
* Splitter.on(';').trimResults().withKeyValueSeparator("=>")
|
|
* .split("a=>b ; c=>b")
|
|
* </code> will return a mapping from {@code "a"} to {@code "b"} and {@code "c"}
|
|
* to {@code b}.
|
|
*
|
|
* <p>
|
|
* The returned map preserves the order of the entries from {@code sequence}.
|
|
*
|
|
* @throws IllegalArgumentException if the specified sequence does not split
|
|
* into valid map entries, or if there are
|
|
* duplicate keys
|
|
*/
|
|
public Map<String, String> split(CharSequence sequence) {
|
|
Map<String, String> map = new LinkedHashMap<String, String>();
|
|
for (String entry : outerSplitter.split(sequence)) {
|
|
Iterator<String> entryFields = entrySplitter.splittingIterator(entry);
|
|
|
|
checkArgument(entryFields.hasNext(), INVALID_ENTRY_MESSAGE, entry);
|
|
String key = entryFields.next();
|
|
checkArgument(!map.containsKey(key), "Duplicate key [%s] found.", key);
|
|
|
|
checkArgument(entryFields.hasNext(), INVALID_ENTRY_MESSAGE, entry);
|
|
String value = entryFields.next();
|
|
map.put(key, value);
|
|
|
|
checkArgument(!entryFields.hasNext(), INVALID_ENTRY_MESSAGE, entry);
|
|
}
|
|
return Collections.unmodifiableMap(map);
|
|
}
|
|
}
|
|
|
|
private interface Strategy {
|
|
Iterator<String> iterator(Splitter splitter, CharSequence toSplit);
|
|
}
|
|
|
|
private abstract static class SplittingIterator extends AbstractIterator<String> {
|
|
final CharSequence toSplit;
|
|
final CharMatcher trimmer;
|
|
final boolean omitEmptyStrings;
|
|
|
|
/**
|
|
* Returns the first index in {@code toSplit} at or after {@code start} that
|
|
* contains the separator.
|
|
*/
|
|
abstract int separatorStart(int start);
|
|
|
|
/**
|
|
* Returns the first index in {@code toSplit} after {@code
|
|
* separatorPosition} that does not contain a separator. This method is only
|
|
* invoked after a call to {@code separatorStart}.
|
|
*/
|
|
abstract int separatorEnd(int separatorPosition);
|
|
|
|
int offset = 0;
|
|
int limit;
|
|
|
|
protected SplittingIterator(Splitter splitter, CharSequence toSplit) {
|
|
this.trimmer = splitter.trimmer;
|
|
this.omitEmptyStrings = splitter.omitEmptyStrings;
|
|
this.limit = splitter.limit;
|
|
this.toSplit = toSplit;
|
|
}
|
|
|
|
@Override
|
|
protected String computeNext() {
|
|
/*
|
|
* The returned string will be from the end of the last match to the beginning
|
|
* of the next one. nextStart is the start position of the returned substring,
|
|
* while offset is the place to start looking for a separator.
|
|
*/
|
|
int nextStart = offset;
|
|
while (offset != -1) {
|
|
int start = nextStart;
|
|
int end;
|
|
|
|
int separatorPosition = separatorStart(offset);
|
|
if (separatorPosition == -1) {
|
|
end = toSplit.length();
|
|
offset = -1;
|
|
} else {
|
|
end = separatorPosition;
|
|
offset = separatorEnd(separatorPosition);
|
|
}
|
|
if (offset == nextStart) {
|
|
/*
|
|
* This occurs when some pattern has an empty match, even if it doesn't match
|
|
* the empty string -- for example, if it requires lookahead or the like. The
|
|
* offset must be increased to look for separators beyond this point, without
|
|
* changing the start position of the next returned substring -- so nextStart
|
|
* stays the same.
|
|
*/
|
|
offset++;
|
|
if (offset >= toSplit.length()) {
|
|
offset = -1;
|
|
}
|
|
continue;
|
|
}
|
|
|
|
while (start < end && trimmer.matches(toSplit.charAt(start))) {
|
|
start++;
|
|
}
|
|
while (end > start && trimmer.matches(toSplit.charAt(end - 1))) {
|
|
end--;
|
|
}
|
|
|
|
if (omitEmptyStrings && start == end) {
|
|
// Don't include the (unused) separator in next split string.
|
|
nextStart = offset;
|
|
continue;
|
|
}
|
|
|
|
if (limit == 1) {
|
|
// The limit has been reached, return the rest of the string as the
|
|
// final item. This is tested after empty string removal so that
|
|
// empty strings do not count towards the limit.
|
|
end = toSplit.length();
|
|
offset = -1;
|
|
// Since we may have changed the end, we need to trim it again.
|
|
while (end > start && trimmer.matches(toSplit.charAt(end - 1))) {
|
|
end--;
|
|
}
|
|
} else {
|
|
limit--;
|
|
}
|
|
|
|
return toSplit.subSequence(start, end).toString();
|
|
}
|
|
return endOfData();
|
|
}
|
|
}
|
|
}
|