RE-flex

A high-performance C++ regex library and lexical analyzer generator with Unicode support.

Three example use cases:

1. A RE/flex-generated tokenizer is used by the Tiger Compiler.
2. The RE/flex C++ regex engines are used by ugrep.
3. As a build dependency for the GCC 3.3 portion of a 2026 port of the open source PSPSDK to GCC 3.3.6, a common version used in the official Sony Playstation Portable development kit.

The RE/flex lexical analyzer generator extends Flex++ with Unicode support, indent/dedent anchors, POSIX regex lazy quantifiers, word boundaries, functions for lex and syntax error reporting, lexer rule execution performance profiling, and other new features.

Only RE/flex supports backtrack-free regex lazy matching in linear time using an advanced DFA transformation algorithm (invented by Dr. Robert van Engelen.)

RE/flex is faster than Flex and much faster than regex libraries such as Boost.Regex, C++11 std::regex, PCRE2 and RE2. For example, tokenizing a 2 KB representative C source code file into 244 tokens takes only 8.7 microseconds:

Command / Function	Software	Time (μs)
reflex --fast --noindent	RE/flex 3.4.1	8.7
reflex --fast	RE/flex 3.4.1	8.9
flex -+ --full	Flex 2.5.35	9.8
boost::spirit::lex::lexertl::actor_lexer::iterator_type	Boost.Spirit.Lex 1.82.0	10.7
reflex --full	RE/flex 3.4.1	20.6
pcre2_jit_match()	PCRE2 (jit) 10.42	60.8
hs_compile_multi(), hs_scan()	Hyperscan 5.4.2	129
reflex -m=boost-perl	Boost.Regex 1.82.0	205
RE2::Consume()	RE2 (pre-compiled) 2023-09-01	218
reflex -m=boost	Boost.Regex POSIX 1.82.0	392
pcre2_match()	PCRE2 10.42	500
RE2::Consume()	RE2 POSIX (pre-compiled) 2023-09-01	534
flex -+	Flex 2.5.35	3759
pcre2_dfa_match()	PCRE2 POSIX (dfa) 10.42	4029
regcomp(), regexec()	GNU C POSIX.2 regex	4932
std::cregex_iterator()	C++11 std::regex	6490

Note: performance in elapsed time (lower is better) in microseconds for 1000 to 10000 benchmark runs using Mac OS X 12.6.9 with clang 12.0.0 -O2, 2.9 GHz Intel Core i7, 16 GB 2133 MHz LPDDR3. Hyperscan disqualifies as a scanner due to its “All matches reported” semantics resulting in 1915 matches for this test, and due to its event handler requirements. Download the tests

The performance table is indicative of the impact on performance when using PCRE2 and Boost.Regex with RE/flex. PCRE2 and Boost.Regex are optional libraries integrated with RE/flex for Perl matching because of their efficiency. By default, RE/flex uses DFA-based extended regular expression matching in linear time, the fastest method (as shown in the table).

The RE/flex matcher tracks line numbers, column numbers, and indentations, whereas Lex and Flex do not (option noyylineno) and neither do the other regex matchers in the table (except PCRE2 and Boost.Regex when used with RE/flex). Tracking this information incurs some overhead. RE/flex also automatically decodes UTF-8/16/32 input and accepts std::istream, strings, and wide strings as input.

Features

• Includes many examples, such as a mini C compiler to Java bytecode, a tokenizer for C/C++ source code, a tokenizer for Python source code, a tokenizer for Java source code, Lua, JSON, XML, YAML, and more.
• Compatible with Flex and Bison to eliminate a learning curve, making a transition from Flex++ to RE/flex frustration-free.
• Auto-generates code that integrates seamlessly with Bison Reentrant, Bison-Bridge, Bison-Locations, Bison 3.0 C++ interface %skeleton "lalr1.cc" and Bison Complete Symbols.
• Generates code and includes methods for lexical and syntax error reporting and recovery.
• The generated scanner source code is structured and easy to understand.
• Fully supports Unicode and Unicode properties \p{C}, including Unicode identifier matching for C++11, Java, C#, and Python source code.
• Auto-detects UTF-8/16/32 input to match Unicode patterns.
• Supports file encodings ISO-8859-1 through ISO-8859-15, CP 1250 through 1258, CP 437, CP 850, CP 858, KOI8, MACROMAN, EBCDIC, and custom code pages.
• Generates scanners for lexical analysis on files, C++ streams, (wide) strings, and memory such as mmap files.
• Indent/nodent/dedent anchors to match indentation levels to tokenize.
• Lazy quantifiers for POSIX regex matching, i.e. no hacks are needed to work around greedy repetitions.
• Word boundary anchors.
• Freespace mode option to improve readability of lexer specifications.
• %class and %init to customize the generated Lexer classes.
• %include to modularize lexer specifications.
• Multiple lexer classes can be combined and used in one application, e.g. by multiple threads in a thread-safe manner.
• Configurable Lexer class generation to customize the interface for various parsers, including Yacc and Bison.
• Generates Graphviz files to visualize FSMs with the Graphviz dot tool.
• Includes an extensible hierarchy of pattern matcher engines, with a choice of regex engines, including the RE/flex regex engine, PCRE2, and Boost.Regex.
• The RE/flex regex library makes C++11 std::regex, PCRE2, and Boost.Regex much easier to use for pattern matching on (wide) strings, files, and streams.
• IEEE POSIX P1003.2 standard compliant like Lex and Flex (but generates C++).
• Extensive documentation in the online User Guide.
• Lots of other improvements over Flex++, such as yypush_buffer_state saves the scanner state (line, column, and indentation positions), not just the input buffer; no input buffer length limit (Flex has a 16KB limit); line() returns the current line (e.g. for error reporting).

Note: PCRE2 and Boost.Regex are not dependencies, they can be used as optional regex engines in addition to the RE/flex regex engine.

Installation

Windows

Use reflex/bin/reflex.exe from the command line or add a Custom Build Step in MSVC++ as follows:

1. select the project name in Solution Explorer then Property Pages from the Project menu (see also custom-build steps in Visual Studio);

2. add an extra path to the reflex/include folder in the Include Directories under VC++ Directories, which should look like $(VC_IncludePath);$(WindowsSDK_IncludePath);C:\Users\YourUserName\Documents\reflex\include (this assumes the reflex source package is in your Documents folder).

3. enter "C:\Users\YourUserName\Documents\reflex\bin\win32\reflex.exe" --header-file "C:\Users\YourUserName\Documents\mylexer.l" in the Command Line property under Custom Build Step (this assumes mylexer.l is in your Documents folder);

4. enter lex.yy.h lex.yy.cpp in the Outputs property;

5. specify Execute Before as PreBuildEvent.

If you are using specific reflex options such as --flex then add these in step 3.

Before compiling your program with MSVC++, drag the folders reflex/lib and reflex/unicode to the Source Files in the Solution Explorer panel of your project. Next, run reflex.exe simply by compiling your project (which may fail, but that is OK for now as long as we executed the custom build step to run reflex.exe). Drag the generated lex.yy.h (if present) and lex.yy.cpp files to the Source Files. Now you are all set!

In addition, the reflex/vs directory contains batch scripts to build projects with MS Visual Studio C++.

MacOS

On macOS systems you can use homebrew to install RE/flex with brew install re-flex. Or use MacPorts to install RE/flex with sudo port install re-flex.

NetBSD

On NetBSD systems you can use the standard NetBSD package installer (pkgsrc): http://cdn.netbsd.org/pub/pkgsrc/current/pkgsrc/devel/RE-flex/README.html

Quick install

First clone the code:

$ git clone https://github.com/Genivia/RE-flex

Then simply do a quick clean build, assuming your environment is pretty much standard:

$ ./clean.sh
$ ./build.sh

This compiles the reflex tool and installs it locally in reflex/bin. For local use of RE/flex in your project, you can add this location to your $PATH variable to enable the new reflex command:

$ export PATH=$PATH:/your_path_to_reflex/reflex/bin

Note that the libreflex.a and libreflex.so libraries are saved locally in reflex/lib. Link against the library when you use the RE/flex regex engine in your code, such as:

$ c++ <options and .o/.cpp files> -L/your_path_to_reflex/reflex/lib -lreflex

or you could statically link libreflex.a with:

$ c++ <options and .o/.cpp files> /your_path_to_reflex/reflex/lib/libreflex.a

Also note that the RE/flex header files that you will need to include in your project are locally located in include/reflex.

To install the man page, the header files in /usr/local/include/reflex, the library in /usr/local/lib and the reflex command in /usr/local/bin:

$ sudo ./allinstall.sh

Configure and make

The configure script accepts configuration and installation options. To view these options, run:

$ ./configure --help

Run configure and make:

$ ./configure && make

To build the examples also:

$ ./configure --enable-examples && make

After this successfully completes, you can optionally run make install to install the reflex command and the libreflex library:

$ sudo make install

Unfortunately, cloning from Git does not preserve timestamps which means that you may run into “WARNING: ‘aclocal-1.15’ is missing on your system.” To work around this problem, run:

$ autoreconf -fi
$ ./configure && make

The above builds the library with SSE/AVX optimizations applied. To disable AVX optimizations:

$ ./configure --disable-avx && make

To disable both SSE2 and AVX optimizations:

$ ./configure --disable-sse2 && make

Optional libraries to install

• To use PCRE2 as a regex engine with the RE/flex library and scanner generator, install PCRE2 and link your code with -lpcre2-8.

• To use Boost.Regex as a regex engine with the RE/flex library and scanner generator, install Boost and link your code with -lboost_regex or -lboost_regex-mt.

• To visualize the FSM graphs generated with reflex option --graphs-file, install Graphviz dot.

Improved Vim syntax highlighting

Copy the lex.vim file to ~/.vim/syntax/ to enjoy improved syntax highlighting for both Flex and RE/flex.

Usage

There are two ways you can use this project:

1. as a scanner generator for C++, similar to Flex;
2. as a flexible regex library API for C++.

For the first use case, use the reflex tool on the command line on a lexer specification:

$ reflex --flex --bison --graphs-file lexspec.l

This generates a scanner for Bison from the lexer specification lexspec.l and saves the finite state machine (FSM) as a Graphviz .gv file that can be visualized with the Graphviz dot tool:

$ dot -Tpdf reflex.INITIAL.gv > reflex.INITIAL.pdf
$ open reflex.INITIAL.pdf

Several examples are included to get you started. See the manual for more details.

For the second use case, use the RE/flex matcher API classes to start pattern search, matching, splitting and scanning on strings, wide strings, files, and streams.

You can select matchers that are based on different regex engines:

• RE/flex regex: #include <reflex/matcher.h> and use reflex::Matcher;
• RE/flex fuzzy regex for approximate matching: #include <reflex/fuzzymatcher.h> and use reflex::FuzzyMatcher
• PCRE2: #include <reflex/pcre2matcher.h> and use reflex::PCRE2Matcher or reflex::PCRE2UTFMatcher.
• Boost.Regex: #include <reflex/boostmatcher.h> and use reflex::BoostMatcher or reflex::BoostPosixMatcher;
• C++11 std::regex: #include <reflex/stdmatcher.h> and use reflex::StdMatcher or reflex::StdPosixMatcher.

Each matcher may differ in regex syntax features (see the full documentation), but they all share the same methods and iterators, such as:

• matches() returns nonzero if the whole input from start to end matches the specified pattern;
• find() search input and returns nonzero if a match was found, can be repeated;
• scan() scan input and returns nonzero if input at current position matches, can be repeated;
• split() returns nonzero for a split of the input at the next match, can be repeated;
• find.begin()…find.end() a filter iterator, iterates with find();
• scan.begin()…scan.end() a tokenizer iterator, iterates with scan();
• split.begin()…split.end() a splitter iterator, iterates with split().

The input matched and searched may be a string, a wide string, a file, or a stream. Searching is incremental, meaning that the input is not buffered as a whole in memory, but rather buffered in parts in a sliding window of a few KB. The window size may grow to fit a pattern match. UTF-16/32 file input with a UTF BOM is automatically normalized and matched as UTF-8.

For example, using Boost.Regex (alternatively use PCRE2 reflex::PCRE2Matcher or reflex::PCRE2UTFMatcher to match Unicode UTF-8 input):

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
// use a BoostMatcher to check if the birthdate string is a valid date
if (reflex::BoostMatcher("\\d{4}-\\d{2}-\\d{2}", birthdate).matches() != 0)
  std::cout << "Valid date!" << std::endl;

With a group capture to fetch the year:

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
// use a BoostMatcher to check if the birthdate string is a valid date
reflex::BoostMatcher matcher("(\\d{4})-\\d{2}-\\d{2}", birthdate);
if (matcher.matches() != 0)
  std::cout << std::string(matcher[1].first, matcher[1].second) << " was a good year!" << std::endl;

A pattern match made by any of the regex engines to match input with matches(), or search with find(), or tokenize with scan(), or split with split() includes detailed information that can be retrieved with the following methods:

• accept() returns group capture index (or zero if not captured/matched)
• text() returns const char* to 0-terminated match (ends in \0)
• strview()returns std::string_view text match (preserves \0s) (C++17)
• str() returns std::string text match (preserves \0s)
• wstr() returns std::wstring wide text match (converted from UTF-8)
• chr() returns first 8-bit char of the text match (str()[0] as int)
• wchr() returns first wide char of the text match (wstr()[0] as int)
• pair() returns std::pair<size_t,std::string>(accept(),str())
• wpair() returns std::pair<size_t,std::wstring>(accept(),wstr())
• size() returns the length of the text match in bytes
• wsize() returns the length of the match in number of wide characters
• lines() returns the number of lines in the text match (>=1)
• columns() returns the number of columns of the text match (>=0)
• begin() returns const char* to non-0-terminated text match begin
• end() returns const char* to non-0-terminated text match end
• rest() returns const char* to 0-terminated rest of input
• span() returns const char* to 0-terminated match enlarged to span the line
• line() returns std::string line with the matched text as a substring
• wline() returns std::wstring line with the matched text as a substring
• more() tells the matcher to append the next match (when using scan())
• less(n) cuts text() to n bytes and repositions the matcher
• lineno() returns line number of the match, starting at line 1
• columno() returns column number of the match in characters, starting at 0
• lineno_end() returns ending line number of the match, starting at line 1
• columno_end() returns ending column number of the match, starting at 0
• bol() returns const char* to begin of matching line (not 0-terminated)
• border()returns the byte offset from the start of the line of the match
• first() returns input position of the first character of the match
• last() returns input position + 1 of the last character of the match
• at_bol() true if matcher reached the begin of a new line \n
• at_bob() true if matcher is at the begin of input and no input consumed
• at_end() true if matcher is at the end of input
• [0] operator returns std::pair<const char*,size_t>(begin(),size())
• [n] operator returns n’th capture std::pair<const char*,size_t>

Note: POSIX matchers do not generally support group capturing, e.g. BoostPosixMatcher and StdPosixMatcher do not. RE/flex is an efficient backtrack-free DFA-based POSIX engine that supports a limited form of capturing, limited to outermost gouping, such as (abc)|(def) which has two groups. This may be extended in a future release to full capturing.

To search a string for words \w+ to display with the column number of each word found:

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
// use a BoostMatcher to search for words in a sentence
reflex::BoostMatcher matcher("\\w+", "How now brown cow.");
while (matcher.find() != 0)
  std::cout << "Found " << matcher.text() << " at column " << matcher.columno() << std::endl;

The split method is roughly the inverse of the find method and returns text located between matches. For example using non-word matching \W+:

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
// use a BoostMatcher to search for words in a sentence
reflex::BoostMatcher matcher("\\W+", "How now brown cow.");
while (matcher.split())
  std::cout << "Found " << matcher.text() << std::endl;

To pattern match the content of a file, where the file may use UTF-8, 16, or 32 encodings that are automatically converted when a UTF BOM is present:

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
// use a BoostMatcher to search and display words from a FILE
FILE *fd = fopen("somefile.txt", "r");
if (fd == NULL)
  exit(EXIT_FAILURE);
reflex::BoostMatcher matcher("\\w+", fd);
while (matcher.find())
  std::cout << "Found " << matcher.text() << std::endl;
fclose(fd);

Same again, but this time with a C++ input stream:

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
// use a BoostMatcher to search and display words from a stream
std::ifstream file("somefile.txt", std::ifstream::in);
reflex::BoostMatcher matcher("\\w+", file);
while (matcher.find())
  std::cout << "Found " << matcher.text() << std::endl;
file.close();

Stuffing the search results into a container using RE/flex iterators:

#include <reflex/boostmatcher.h> // reflex::BoostMatcher, reflex::Input, boost::regex
#include <vector>         // std::vector
// use a BoostMatcher to convert words of a sentence into a string vector
reflex::BoostMatcher matcher("\\w+", "How now brown cow.");
std::vector<std::string> words(matcher.find.begin(), matcher.find.end());

Use C++11 range-based loops with RE/flex iterators:

#include <reflex/pcre2matcher.h> // reflex::PCRE2TFMatcher, reflex::Input, std::regex
// use a PCRE2UTFMatcher to search for words in a sentence
reflex::PCRE2UTFMatcher matcher("\\w+", "How now brown cow.");
for (auto& match : matcher.find)
  std::cout << "Found " << match.text() << std::endl;

Note that we cannot generally simplify this loop to the following, because the temporary matcher object is destroyed (some compilers handle this in C++23):

for (auto& match : reflex::PCRE2UTFMatcher matcher("\\w+", "How now brown cow.").find);
  std::cout << "Found " << match.text() << std::endl;

RE/flex also allows you to convert expressive regex syntax forms such as \p Unicode classes, character class set operations such as [a-z--[aeiou]], escapes such as \X, and (?x) mode modifiers, to a regex string that the underlying regex library understands and will be able to use:

• std::string reflex::Matcher::convert(const std::string& regex, reflex::convert_flag_type flags)
• std::string reflex::PCRE2Matcher::convert(const std::string& regex, reflex::convert_flag_type flags)
• std::string reflex::PCRE2UTFMatcher::convert(const std::string& regex, reflex::convert_flag_type flags)
• std::string reflex::BoostMatcher::convert(const std::string& regex, reflex::convert_flag_type flags)
• std::string reflex::StdMatcher::convert(const std::string& regex, reflex::convert_flag_type flags)

For example:

#include <reflex/matcher.h> // reflex::Matcher, reflex::Input, reflex::Pattern
// use a Matcher to check if sentence is in Greek:
static const reflex::Pattern pattern(reflex::Matcher::convert("[\\p{Greek}\\p{Zs}\\pP]+", reflex::convert_flag::unicode));
if (reflex::Matcher(pattern, sentence).matches() != 0)
  std::cout << "This is Greek" << std::endl;

We use convert with optional flag reflex::convert_flag::unicode to make . (dot), \w, \s and so on match Unicode and to convert \p Unicode character classes.

Conversion is fast (it runs in linear time in the size of the regex), but it is not without some overhead. Making converted regex patterns static as shown above saves the cost of conversion to just once to support many matchings.

How to contribute?

Please see CONTRIBUTING.

Where do I find the documentation?

Read more about RE/flex in the manual.

License and copyright

RE/flex by Robert van Engelen, Genivia Inc. Copyright © 2016-2025, All rights reserved.

RE/flex is distributed under the BSD-3 license LICENSE.txt. Use, modification, and distribution are subject to the BSD-3 license.

Reporting bugs

Visit GitHub to report bugs: https://github.com/Genivia/RE-flex

Changelog

• Nov 14, 2016: 0.9.0 beta released
• Nov 15, 2016: 0.9.1 improved portability
• Nov 17, 2016: 0.9.2 improvements and fixes for minor issues
• Nov 19, 2016: 0.9.3 replaces %import with %include, adds freespace option -x, fixes minor issues
• Nov 20, 2016: 0.9.4 fixes minor issues, added new examples/json.l
• Nov 25, 2016: 0.9.5 bug fixes and improvements
• Dec 1, 2016: 0.9.6 portability improvements
• Dec 6, 2016: 0.9.7 bug fixes, added option --regexp-file, Python tokenizer
• Dec 9, 2016: 0.9.8 fixes minor issues, improved reflex tool options --full and --fast, generates scanner with FSM table or a fast scanner with FSM code, respectively
• Jan 8, 2017: 0.9.9 bug fixes and improved Flex compatibility
• Jan 15, 2017: 0.9.10 improved compatibility with Flex options, fixed critical issue with range unions
• Jan 25, 2017: 0.9.11 added C++11 std::regex matching engine support, moved .h files to include/reflex, requires #include <reflex/xyz.h> from now on, fixed errno_t portability issue
• Mar 3, 2017: 0.9.12 refactored and improved, includes new regex converters for regex engines that lack regex features such as Unicode character classes
• Mar 4, 2017: 0.9.13 improved warning and error messages
• Mar 6, 2017: 0.9.14 reflex option -v shows stats with execution timings, bug fixes
• Mar 8, 2017: 0.9.15 added wtext(), wpair(), winput() methods, other improvements
• Mar 22, 2017: 0.9.16 bug fixes, speed improvements, improved option --unicode regex conversion, also with (?u:), changed wtext() to wstr() and added a str() method
• Mar 24, 2017: 0.9.17 improvements
• Mar 26, 2017: 0.9.18 added reflex option -p (--perf-report) for performance debugging, added doc/man/reflex.1 man page, added interactive readline example
• Mar 31, 2017: 0.9.19 fixed reflex option -m, lexer.in(i) now resets the lexer, fixed reassigning the same input to the lexer that caused UTF BOM to be read twice
• Apr 5, 2017: 0.9.20 EBCDIC file translation, other improvements
• Apr 10, 2017: 0.9.21 fixed option -P to support multiple lexer classes in one application, added configure installation script, optional quick install with allinstall.sh (renamed from install.sh)
• Apr 12, 2017: 0.9.22 improved explanations of matches(), find(), scan(), split() that return nonzero for a match, other minor improvements
• May 24, 2017: 0.9.23 improved portability, added file encoding conversions for CP-1250 to CP-1258, CP 437, and CP 850/858
• Jun 24, 2017: 0.9.24 added an option for users to define their own custom code pages to translate input, fixed # in free space mode
• Jun 28, 2017: 0.9.25 fixed --fast FSM not always halting on EOF after a mismatch; fixed buffer realloc, added new examples/csv.l
• Jul 5, 2017: 0.9.26 fixed wstr() always returning UTF-16 strings (should be UTF-16 only when std::wstring requires it)
• Sep 26, 2017: 0.9.27 the Flex-compatible yy_scan_string(), yy_scan_bytes() and yy_scan_buffer() functions now create a new buffer as in Flex, delete this buffer with yy_delete_buffer(); fixed examples to work with newer Bison versions (Bison 3.0.4)
• Dec 12, 2017: 0.9.28 added yy_scan_wstring and yy_scan_wbuffer for wide string scanning with Flex-like functions
• Jan 28, 2018: 1.0.0 removed dynamic exception specifications to comply with C++17, upgraded to stable release 1.0
• Feb 24, 2018: 1.0.1 added Unicode IsBlockName categories
• Mar 6, 2018: 1.0.2 added namespace nesting with %option namespace=NAME1.NAME2.NAME3 ...
• Mar 7, 2018: 1.0.3 fixed --namespace and %option namespace
• Apr 22, 2018: 1.0.4 updated to Unicode 10, cleaned up code to remove tool warnings
• Jun 29, 2018: 1.0.5 updated --namespace for options --fast and --full to support the generation of multiple optimized lexers placed in namespaces.
• Jul 9, 2018: 1.0.6 added --bison-cc option to generate scanners for Bison 3.0 %skeleton "lalr1.cc" C++ parsers, included two examples flexexample9xx and reflexexample9xx to demo this feature.
• Jul 12, 2018: 1.0.7 added --bison-cc-namespace and --bison-cc-parser options to customize Bison 3.0 %skeleton "lalr1.cc" C++ parsers.
• Jul 30, 2018: 1.0.8 updated to Unicode 11.
• Aug 21, 2018: 1.0.9 fixed reflex regex library matching with range quantifiers by correcting coding typo.
• Dec 8, 2018: 1.0.10 fixed columno() to take tab spacing into account.
• Jan 18, 2019: 1.0.11 fixed GCC 8.2 warnings, additional enhancements.
• Jan 21, 2019: 1.0.12 the reflex tool now reads files using reflex::Input.
• Feb 20, 2019: 1.1.0 code quality updates.
• Mar 6, 2019: 1.1.1 fixed configure and make install header files, updated --bison-locations option.
• Mar 7, 2019: 1.1.2 fixed reflex tool handling of backslashes in file paths.
• Mar 11, 2019: 1.1.3 updated to Unicode 12, examples can now be built with ./configure --enable-examples.
• Mar 27, 2019: 1.1.4 fixed reflex tool common top code block insertion for all inclusive states.
• Apr 6, 2019: 1.1.5 improved reflex tool command-line option handling, updated documentation.
• Apr 29, 2019: 1.2.0 added --bison-complete option, new ugrep utility example, updated manual, fixes minor issues.
• May 15, 2019: 1.2.1 added reflex::convert_flag::basic to convert BRE syntax to ERE syntax, used by ugrep.
• May 28, 2019: 1.2.2 fixed MinGW builds, fixed reflex::Input::get() to return positive character code, matcher option "N" for scan and find matches empty input (^$).
• Jun 21, 2019: 1.2.3 fixed reflex long regex string generation, namespaces with option --full, updated documentation and other improvements.
• Jun 24, 2019: 1.2.4 fixed an issue with reflex::StdMatcher (std::regex) causing failures to match input with split.
• Jul 2, 2019: 1.2.5 added reflex::Input::in(const char *memptr, size_t memlen) to read a memory segment (for scanning etc.), added reflex::Input::streambuf class to use a reflex::Input object as a std::streambuf, improved yy_scan_buffer and yy_scan_bytes.
• Jul 3, 2019: 1.2.6 fixed a problem with lazy quantifiers used within negative patterns.
• Jul 4, 2019: 1.2.7 added reflex::Input::dos_streambuf to convert DOS CRLF to LF, other improvements.
• Jul 8, 2019: 1.2.8 added support for inverted mode modifiers (?-imsux) to reflex::convert and reflex::Pattern.
• Jul 11, 2019: 1.2.9 portability improvements.
• Jul 14, 2019: 1.2.10 added AbstractMatcher::set_bob(), moved AbstractMatcher::peek() to public, minor improvements.
• Jul 21, 2019: 1.3.0 added subtractive start condition scoping with <^...>, added undent \k anchor to undo indenting changes (“undenting”) with an example in examples/indent2.l, improved indent \i and dedent \j anchors and other improvements.
• Jul 24, 2019: 1.3.1 added matcher().tabs(n) to set tab size, used by columno() and indent \i and dedent \j anchors, new reflex::Pattern methods, other improvements.
• Jul 27, 2019: 1.3.2 added defined name expansion in bracket lists for the union ||, intersection &&, and subtraction -- operations, e.g. [||{letter}||{digit}] expands into [a-zA-Z0-9] when letter is defined as [a-zA-Z] and digit is defined as [0-9], see Character Classes in the documentation.
• Aug 5, 2019: 1.3.3 fixed reflex::BoostMatcher (and derived reflex::BoostPosixMatcher, reflex::BoostPerlMatcher) regression bug that crept into the 1.2.4 update.
• Aug 7, 2019: 1.3.4 speed improvements for non-fast options.
• Aug 8, 2019: 1.3.5 further speed improvements for both fast and non-fast options (15% to 30% faster).
• Aug 12, 2019: 1.3.6 added lexer and matcher buffer(base, size) methods and improved Flex-compatible yy_scan_buffer(base, size), these functions scan memory fast with zero copy overhead; added mmap.l example to scan an mmap-ed file fast with mmap(2) and buffer(base, size); other improvements.
• Aug 16, 2019: 1.3.7 added reflex::BufferedInput::dos_streambuf to improve dos_streambuf speed by buffering (reflex::Input::dos_streambuf is unbuffered), fixed %option token-type to apply without restrictions.
• Aug 17, 2019: 1.3.8 added caching of lineno() and columno() to increase speed, which is essential for large buffers such as large mmap-ed files scanned with buffer(base, size); other improvements.
• Sep 4, 2019: 1.4.0 added reflex option -S (--find) for efficient searching instead of scanning input (i.e. efficiently ignoring unmatched input) demonstrated with new findfast and findsearch examples; changed --nodefault to throw an exception when option --flex is not used and when the default rule is triggered.
• Sep 10, 2019: 1.4.1 fixed lineno() caching issue (1.3.8 bug); faster find.
• Sep 15, 2019: 1.4.2 faster find for patterns beginning with optional repetitions such as .*.
• Sep 29, 2019: 1.4.3 added reflex::AbstractMatcher::clone() to clone a referenced concrete matcher object.
• Oct 23, 2019: 1.4.4 improved option --flex for Flex compatibility; fixed option --token-type with option --flex, now properly defines YY_NULL and yyterminate; fixed AbstractMatcher::buffer(n) for large n; faster find.
• Nov 5, 2019: 1.5.0 added border(), span(), line(), wline(), and skip(c) methods; added new section on error reporting and recovery to the documentation; fixed yy_scan_string() and yy_scan_buffer() when called before calling yylex() for the first time; improved performance.
• Nov 7, 2019: 1.5.1 improvements; added dos.l demo example of reflex::InputBuffer::dos_streambuf.
• Nov 12, 2019: 1.5.2 fixed an internal buffer allocation issue that may cause a crash when input lines are longer than 16KB (regression bug that crept into in 1.5.0).
• Nov 21, 2019: 1.5.3 added lineno_end() and columno_end() methods, updated columns() with clarifications in the updated documentation; expanded the documentation with additional error reporting and handling techniques with RE/flex and Bison bridge and complete configurations; FSM code generation improvements.
• Nov 22, 2019: 1.5.4 added flexexample11xx example with Flex specification and Bison complete parser; minor improvements.
• Dec 23, 2019: 1.5.5 expanded the skip(c) methods with a wchar_t wide character parameter and a UTF-8 string parameter to skip input; added new option --token-eof.
• Dec 28, 2019: 1.5.6 added new option --noindent to speed up pattern matching and lexical analysis by disabling indentation tracking in the input (also disables anchors \i, \j, and \k); speed improvements.
• Jan 19, 2020: 1.5.7 expanded file encoding formats to include ISO-8859-2 to 16, MacRoman, KOI8; fixed a bug in line() and span().
• Feb 3, 2020: 1.5.8 added wunput() method; added lex.vim improved Flex and RE/flex Vim syntax highlighting; added yaml.l example; fixed --freespace with --unicode when bracket lists contain a #; character class operators {+}, {-}, {&} now accept defined names as first operands and inverted character classes; indent anchor \k now matches only when indent level is changed as documented.
• Mar 2, 2020: 1.6.0 added PCRE2 regex matcher classes and updated reflex option --matcher=pcre2-perl; optimized RE/flex matcher find() with AVX/SSE2/NEON/AArch64; updated and improved regex converters.
• Mar 3, 2020: 1.6.1 fixed missing PCRE2 regex type std::string in generated scanners.
• Mar 10, 2020: 1.6.2 improved reflex::PCRE2Matcher; fixed MSVC++ x86 32-bit build error when HAVE_AVX512BW is enabled (requires AVX512BW).
• Mar 19, 2020: 1.6.3 updated to Unicode 13; fixed start condition scope specifications for patterns that start with a {; updated lex.vim.
• Mar 22, 2020: 1.6.4 added option --yy to enable --flex and --bison, but also defines the global FILE* variables yyin and yyout for enhanced Lex/Flex compatibility (yyin is otherwise a pointer to the reflex::Input object to read files, streams, and strings).
• Mar 23, 2020: 1.6.5 updated to permit } as closing marker for %top{, %class{, and %init{ code blocks, i.e. %} or } may be used as closing markers.
• Mar 31, 2020: 1.6.6 fixed an issue where a trailing backslash in a pattern in a lexer specification causes a reflex-generated C++ comment to extend to the next line, which results in a compilation warning and possibly a line of code being skipped.
• Apr 30, 2020: 1.6.7 minor improvements to parse and convert regex patterns to FSMs.
• May 14, 2020: 2.0.0 faster FSM construction; new FSM VM opcodes; relaxed limits of pattern length and complexity (max 16,711,679 FSM opcode words, from 65,536 words previously) for high-performance pattern matching with very long and complex regex patterns.
• May 25, 2020: 2.0.1 resolved 64-bit build warnings; fixed reflex CRLF output in C++ source code on Windows.
• Jun 29, 2020: 2.1.0 added Bison complete locations filename member access; added example Mini C compiler minic using RE/flex scanner with Bison 3.2 C++ complete locations, compiles C-like source code to Java bytecode (class files); added fast fuzzy (approximate) regex matcher reflex::FuzzyMatcher derived from reflex::Matcher.
• Jul 8, 2020: 2.1.1 added %option params to extend lex()/yylex() parameters; updated AVX2 detection for SIMD optimizations.
• Jul 9, 2020: 2.1.2 minor update to fix MSVC++ compiler error.
• Aug 3, 2020: 2.1.3 improved --bison-bridge option; updated examples.
• Aug 19, 2020: 2.1.4 changed IN_HEADER to yyIN_HEADER when --flex is used with --header-file; added reflex::Input::Handler event handler for custom handling of FILE* errors and non-blocking FILE* streams.
• Sep 20, 2020: 2.1.5 added matcher method lineno(n) to set or change the line number to n; added yyset_lineno(n,s) to flexlexer.h; updated Mini C compiler example.
• Oct 12, 2020: 3.0.0 fixed a regression bug since v2.x; redesigned internals to increase IO efficiency and regex pattern search speed.
• Oct 24, 2020: 3.0.1 improved handling of UTF-16/32 file encodings on Windows to prevent ^Z eof when files are opened in text mode.
• Mar 16, 2021: 3.0.2 minor improvements and additions; fixed MinGW builds.
• Apr 29, 2021: 3.0.3 fixed reflex::Input copy constructor; minor improvements.
• May 27, 2021: 3.0.4 improvements and fixes for minor issues.
• Jun 1, 2021: 3.0.5 minor usability improvements; changed \d to match Unicode when option unicode is enabled.
• Jun 4, 2021: 3.0.6 minor improvements.
• Jun 6, 2021: 3.0.7 updated examples to fix build problems.
• Jul 15, 2021: 3.0.8 minor improvements.
• Jul 22, 2021: 3.0.9 SIMD code split into simd_avx2.cpp and simd_avx512bw.cpp to support runtime CPU ID checking when the library is built with ./configure; make, disable AVX with ./configure --disable-avx, disable SSE2 with ./configure --disable-sse2; UTF-16LE BOM detection correction.
• Aug 14, 2021: 3.0.10 fixed missing simd.h after installation, added REFLEX_BUFFER_SIZE to customize the initial size and growth of the input buffer.
• Sep 14, 2021: 3.0.11 minor change to apply --prefix to the generated REFLEX_code_[PREFIX]STATE code.
• Oct 11, 2021: 3.0.12 updated to Unicode 14; fixed a compilation issue with --params when used with --flex.
• Dec 10, 2021: 3.1.0 improved buffering mechanism with reduced buffer sizes; faster columno() for long lines; fix CP-1251 table typo.
• Feb 20, 2022: 3.2.0 new %option ctorinit; faster compilation of regular expressions to tables and direct code DFAs; refactored SIMD source code to enable AVX2 and AVX512BW optimizations in multi-version matcher code; updated Windows binary file opening.
• Feb 23, 2022: 3.2.1 regression bug in 3.2.0 fix.
• Mar 13, 2022: 3.2.2 minor improvements.
• Apr 2, 2022: 3.2.3 fixed C++17 compliance and dso build.
• Apr 29, 2022: 3.2.4 new %begin directive; new --batch=SIZE option argument.
• Apr 30, 2022: 3.2.5 minor correction.
• May 6, 2022: 3.2.6 improved fuzzy matcher.
• May 10, 2022: 3.2.7 improved fuzzy matcher, see also https://github.com/Genivia/FuzzyMatcher.
• Jul 6, 2022: 3.2.8 minor update to correct a lexer file parsing issue in the reflex code generator tool.
• Aug 12, 2022: 3.2.9 add cmake rules for Windows builds.
• Aug 21, 2022: 3.2.10 bug fix in regular expression converter.
• Aug 29, 2022: 3.2.11 new lua2lisp transpiler example with Bison complete symbols and locations.
• Jan 6, 2023: 3.2.12 regression bug fix in trailing context pattern matching, e.g. xy and x/x patterns collided when they should not; updated yaml parser example.
• Jan 24, 2023: 3.3.0 overall update; corrected a problem with Unicode characters in regex patterns adjacent to curly braces; permit cxx extension in FSM filename output.
• Mar 6, 2023: 3.3.1 for consistency include NAME in default header file name, tables file name, and graphs file name when option --prefix is specified.
• Mar 17, 2023: 3.3.2 fix a performance issue with case-insensitive pattern construction.
• May 28, 2023: 3.3.3 fix yyrestart dropping the first character; faster Matcher::find().
• May 31, 2023: 3.3.4 fix Matcher::find() initialization issue in 3.3.3.
• Jun 12, 2023: 3.3.5 improve source code output of lexer class definitions.
• Jul 11, 2023: 3.3.6 faster Matcher::find(); improved --stdout to include tables.
• Jul 17, 2023: 3.3.7 faster Matcher::find().
• Aug 4, 2023: 3.3.8 minor update to sync up the code base with the ugrep project.
• Aug 16, 2023: 3.3.9 fix avx512bw compilation error; new LineMatcher matching engine.
• Sep 16, 2023: 3.4.0 fix FuzzyMatcher::DEL flag when this is the only flag selected; fix FuzzyMatcher::matches() bug that incorrectly matched an extra character before the end of the input; optimize find(); updated saving the FSM pred[] hashes to a file, which has changed; increase default buffer size REFLEX_BUFSZ to 128K for best throughput performance.
• Sep 25, 2023: 3.4.1 make word boundaries \b, \B, \< and \> applicable anywhere in a pattern.
• Oct 7, 2023: 3.5.0 updated to Unicode 15.1; clarify . (dot) with %unicode enabled, which is a catch-all pattern; update \X to match only valid Unicode characters.
• Nov 5, 2023: 3.5.1 minor improvements.
• Feb 17, 2024: 4.0.0 faster Matcher::find() with a new DFA cut algorithm to optimize match prediction speed and accuracy, see also ugrep 5.0; apply Unicode pattern canonicalization with reflex::convert(..., reflex::convert_flag::unicode).
• Feb 23, 2024: 4.0.1 new rawk example to demonstrate awk-like fast search in C++; enable <<EOF>> rules for option find to generate a fast search engine.
• Mar 5, 2024: 4.1.0 improved lazy quantifiers for POSIX regex lazy matching in linear time using an advanced DFA transformation algorithm introduced in RE/flex in 2016.
• Mar 11, 2024: 4.1.1 minor update to correct a DFA construction problem for POSIX regex lazy quantifiers matching too much in some cases.
• Mar 17, 2024: 4.1.2 updated configure scripts; cast negative ctype function arguments (problem detected on NetBSD 10).
• Mar 27, 2024: 4.2.0 support pkg-config with reflex.pc (and reflexmin.pc minimized library) to use the reflex library -lreflex.
• Apr 10, 2024: 4.2.1 minor update to adjust the current input pointer by one when not matching anything.
• May 12, 2024: 4.3.0 faster Matcher::find() with refactored SIMD (SSE2/AVX2/AVX512BW/NEON/AArch64) code; larger default 256KB buffer (from 128KB).
• Jun 6, 2024: 4.4.0 upgraded reflex::Matcher and reflex::FuzzyMatcher to respect Unicode word boundaries instead of only ASCII \<, \>, \b, \B; upgraded regex Unicode converters to Unicode [::] character classes instead of only ASCII [[:alpha:]] etc.; improved FSM code generation without local c0.
• Aug 1, 2024: 4.5.0 minor speed improvements.
• Oct 29, 2024: 5.0.0 faster SIMD regex search methods; improved anchor and word boundary matching speed and support; new std::string_view strview() matcher method.
• Nov 29, 2024: 5.1.0 appease -Woverload-virtual and -Wshadow warnings; fix a bug in case-insensitive Unicode negated character class matching too much.
• Jan 9, 2025: 5.1.1 fix a minor issue with case-insensitive matching when regex patterns are specified in certain ways in combination with “string-like” patterns such that one ore more alternating sub-patterns overlap from the start.
• Jan 24, 2025: 5.2.0 new file encoding type null_data to read NUL as LF and vice versa; supports reading xargs -0 output for example.
• Jan 27, 2025: 5.2.1 fix a regression issue with non-SIMD-optimized search with Matcher::find().
• Feb 3, 2025: 5.2.2 Matcher::find() speed improvements for certain regex patterns that do not match the input.
• Feb 28, 2025: 5.3.0 improve case-insensitive matching to support the Unicode upper/lower sets of letters.
• Apr 29, 2025: 5.4.0 natively support non-blocking FILE* input, i.e. for fcntl O_NONBLOCK the regex matchers will wait for input to become available again instead of giving up with an error; changed reflex::Input::Handler see documentation; remove compiler warnings.
• May 16, 2025: 5.5.0 new reflex option -D to immediately debug a scanner’s lexer patterns against a specified input file; fix an issue with Matcher::find() for certain short patterns with ^$ anchors; mark likely and unlikely branches in hot paths for Matcher::find() performance.
• Jun 19, 2025: 6.0.0 faster Matcher::find() with new and expanded predict-match PM3+PM5 methods to replace PM4; updated FSM code generation to support the expanded prediction tables now stored in compressed form as a hex string when reflex option find is used with fast or full to generate a pre-compiled search engine.
• Mar 6, 2026: 6.1.0 minor improvements to align the code base with recent ugrep updates.
• May 11, 2026: 6.2.0 pattern matcher’s AbstractMatcher::avail() method accepts an optional max length parameter; permit very long lines in the input (256MB long lines) when using AbstractMatcher::line() and AbstractMatcher::eol() without a begin-of-line loss (BOLSZ threshold can be changed.)
• May 17, 2026: 6.3.0 add new AbstractMatcher methods chr_last(), wchr_last(), chr_next(), wchr_next(), empty(); add new example words.l.
• Aug 26, 2026: 6.4.0 improve DFA growth control measures.
• Sep 15, 2026: 6.5.0 fix an unlikely edge case when converting a file input encoding to UTF-8, inserts extra byte in the converted output; fix reflex tool option -i (lexer option case-insensitive) causing a “syntax error” issue with some inverted character classes.