All Downloads are FREE. Search and download functionalities are using the official Maven repository.

net.sf.saxon.regex.JavaRegularExpression Maven / Gradle / Ivy

There is a newer version: 12.5
Show newest version
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Copyright (c) 2018-2022 Saxonica Limited
// This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0.
// If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
// This Source Code Form is "Incompatible With Secondary Licenses", as defined by the Mozilla Public License, v. 2.0.
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

package net.sf.saxon.regex;

//import net.sf.saxon.expr.regexj.Matcher;
//import net.sf.saxon.expr.regexj.Pattern;

import net.sf.saxon.str.StringView;
import net.sf.saxon.str.UnicodeString;
import net.sf.saxon.trans.XPathException;
import net.sf.saxon.tree.iter.AtomicIterator;
import net.sf.saxon.tree.iter.EmptyIterator;

import java.util.function.Function;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
import java.util.regex.PatternSyntaxException;

/**
 * An implementation of RegularExpression that calls the JDK regular expression library directly.
 * This can be invoked by appending ";j" to the flags attribute/argument.
 *
 * 

Note that in Saxon-CS, this class is emulated by code that invokes the .NET * regex engine, which has different rules. In this case the ";n" flag activates this option.

*/ public class JavaRegularExpression implements RegularExpression { Pattern pattern; String javaRegex; int flagBits; /** * Create a regular expression, starting with an already-translated Java regex. * NOTE: this constructor is called from compiled XQuery code * * @param javaRegex the regular expression after translation to Java notation * @param flags the user-specified flags (prior to any semicolon) */ public JavaRegularExpression(UnicodeString javaRegex, String flags) throws XPathException { this.flagBits = setFlags(flags); this.javaRegex = javaRegex.toString(); try { // The Java COMMENTS option does more than the XPath "x" option. However, // on .NET, the "x" option can be emulated accurately. flagBits = flagBits & (~(Pattern.COMMENTS)); pattern = Pattern.compile(this.javaRegex, flagBits); } catch (PatternSyntaxException e) { throw new XPathException("Incorrect syntax for Java regular expression", e); } } /** * Get the Java regular expression (after translation from an XPath regex, but before compilation) * * @return the regular expression in Java notation */ public String getJavaRegularExpression() { return javaRegex; } /** * Get the flag bits as used by the Java regular expression engine * * @return the flag bits */ public int getFlagBits() { return flagBits; } /** * Use this regular expression to analyze an input string, in support of the XSLT * analyze-string instruction. The resulting RegexIterator provides both the matching and * non-matching substrings, and allows them to be distinguished. It also provides access * to matched subgroups. * @param input the string to which the regular expression is to be applied */ @Override public RegexIterator analyze(UnicodeString input) { return new JRegexIterator(input.toString(), pattern); } /** * Determine whether the regular expression contains a match for a given string * * @param input the string to match * @return true if the string matches, false otherwise */ @Override public boolean containsMatch(UnicodeString input) { return pattern.matcher(input.toString()).find(); } /** * Determine whether the regular expression matches a given string in its entirety * * @param input the string to match * @return true if the string matches, false otherwise */ @Override public boolean matches(UnicodeString input) { return pattern.matcher(input.toString()).matches(); } /** * Replace all substrings of a supplied input string that match the regular expression * with a replacement string. * * @param input the input string on which replacements are to be performed * @param replacement the replacement string in the format of the XPath replace() function * @return the result of performing the replacement * @throws net.sf.saxon.trans.XPathException * if the replacement string is invalid */ @Override public UnicodeString replace(UnicodeString input, UnicodeString replacement) throws XPathException { Matcher matcher = pattern.matcher(input.toString()); try { return StringView.tidy(matcher.replaceAll(replacement.toString())); } catch (IndexOutOfBoundsException e) { throw new XPathException(e.getMessage(), "FORX0004"); } } /** * Replace all substrings of a supplied input string that match the regular expression * with a replacement string. * * @param input the input string on which replacements are to be performed * @param replacement a function that is called once for each matching substring, and * that returns a replacement for that substring * @return the result of performing the replacement * @throws XPathException if the replacement string is invalid */ @Override public UnicodeString replaceWith(UnicodeString input, Function replacement) throws XPathException { throw new XPathException("saxon:replace-with() is not supported with the Java regex engine"); } /** * Use this regular expression to tokenize an input string. * * @param input the string to be tokenized * @return a SequenceIterator containing the resulting tokens, as objects of type StringValue */ @Override public AtomicIterator tokenize(UnicodeString input) { if (input.isEmpty()) { return EmptyIterator.ofAtomic(); } return new JTokenIterator(input.toString(), pattern); } /** * Set the Java flags from the supplied XPath flags. The flags recognized have their * Java-defined meanings rather than their XPath-defined meanings. The available flags are: *

d - UNIX_LINES

*

m - MULTILINE

*

i - CASE_INSENSITIVE

*

s - DOTALL

*

x - COMMENTS

*

u - UNICODE_CASE

*

q - LITERAL

*

c - CANON_EQ

* * @param inFlags the flags as a string, e.g. "im" * @return the flags as a bit-significant integer * @throws XPathException if the supplied value contains an unrecognized flag character * @see java.util.regex.Pattern */ public static int setFlags(/*@NotNull*/ CharSequence inFlags) throws XPathException { int flags = Pattern.UNIX_LINES; for (int i = 0; i < inFlags.length(); i++) { char c = inFlags.charAt(i); switch (c) { case 'd': flags |= Pattern.UNIX_LINES; break; case 'm': flags |= Pattern.MULTILINE; break; case 'i': flags |= Pattern.CASE_INSENSITIVE; break; case 's': flags |= Pattern.DOTALL; break; case 'x': flags |= Pattern.COMMENTS; // note, this enables comments as well as whitespace break; case 'u': flags |= Pattern.UNICODE_CASE; break; case 'q': flags |= Pattern.LITERAL; break; case 'c': flags |= Pattern.CANON_EQ; break; default: XPathException err = new XPathException("Invalid character '" + c + "' in regular expression flags"); err.setErrorCode("FORX0001"); throw err; } } return flags; } /** * Get the flags used at the time the regular expression was compiled. * * @return a string containing the flags */ @Override public String getFlags() { String flags = ""; if ((flagBits & Pattern.UNIX_LINES) != 0) { flags += 'd'; } if ((flagBits & Pattern.MULTILINE) != 0) { flags += 'm'; } if ((flagBits & Pattern.CASE_INSENSITIVE) != 0) { flags += 'i'; } if ((flagBits & Pattern.DOTALL) != 0) { flags += 's'; } if ((flagBits & Pattern.COMMENTS) != 0) { flags += 'x'; } if ((flagBits & Pattern.UNICODE_CASE) != 0) { flags += 'u'; } if ((flagBits & Pattern.LITERAL) != 0) { flags += 'q'; } if ((flagBits & Pattern.CANON_EQ) != 0) { flags += 'c'; } return flags; } }




© 2015 - 2024 Weber Informatics LLC | Privacy Policy