Regular expressions are commonly used in pattern search algorithms. This tool is based on the work of John Maddock (See his own regex comparison here) and the sljit project (See their regex comparison here).
| dependency | version |
|---|---|
| Cmake | >=3.0 |
| Ragel | 6.9 |
| Python | >=3.0 |
| Boost (*2) | >=1.57 |
| Pcap | >=0.8 |
| Autoconf | 2.69 (*) |
| Automake | 1.15 (*) |
| Autopoint | 0.19.7 (*) |
| Gettext | 0.19.7 (*) |
| Libtool | 2.4.6 (*) |
| Git | 2.11.0 (*) |
(*) Tested with named version only. Older versions may work too. (*2) Needs boost-regex to be installed as a component
The following regex engines are supported and covered by the tool:
- CTRE
- Hyperscan
- Oniguruma
- RE2
- Tre
- PCRE2
- Rust regex crate
- Regress regex crate
- Boost regex
- C++ Standard regex
- YARA
The engines are built from their sources. In the case an installed engine should be used,
the corresponding cmake variable INCLUDE_<name> has to be set to system. The configuration script
tries to locate the library and uses the library for linking the benchmark.
The same variable can be set to disabled to exclude an engine.
The configuration script distinguishes between nightly and other Rust toolchains to enable the SIMD-feature which is currently available in the nightly built only. The SIMD-feature improves the throughput of the regex crate for defined expressions.
The different engines have different requirements which are not described here. Please see the related project documentations.
On Ubuntu 20.04 these were necessary installs to get the build done from a stock AWS box
$ apt install build-essential cmake rustc cargo automake autoconf autopoint autogen \
libtool libprotobuf-dev libprotobuf-c-dev protobuf-compiler ninja-build \
ragel libpcap pcaputils pkg-config libboost-dev flex bisonIn the case all depencies are fulfilled, just configure and build the cmake based project:
mkdir build && cd build
cmake ..
makeThe make command will build all engines and the test tool regex_perf.
To build the test tool or a library only, call make with corresponding target, i.e.:
make regex_perfThe test tool calls each engine with a defined set of different regular expression on a given file. The repository contains a ~16Mbyte large text file (3200.txt) which can be used for measuring.
./src/regex_perf -f ./3200.txtBy default, the tool repeats each test 5 times and prints the best time of each test. The overall time to process each regular expression is measured and accounted. The scoring algorithhm distributes the fastest engine 5 points, the second fastest 4 points and so on. The score points help to limit the impact of a slow regular expression eninge test in comparision to the absolut time value.
You can specify a file to write the test results per expression and engine:
./src/regex_perf -f ./3200.txt -o ./results.csvThe test tool writes the results in a csv-compatible format.
We included a spreadsheet generator for easy visualization of the results.
Once you have ran the results and obtained the results.csv file, you can create a spreadsheet
with (assuming you are still in the build directory)
python3 ../genspreadsheet.py results.csvIt will save an Excel spreadsheet with the name regex-results-YYYYMMDD-HHMMSS.xlsx in the current
directory.
Unfortunately it is not possible to run both standard C++ from GCC/stdlibc++ and clang+libc++ at the same time, it is just the way that cmake selects a single compiler.
To run with clang+libc++ use the following recipe:
mkdir build && cd build
cmake \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_EXE_LINKER_FLAGS="-lc++abi -lc++" \
-DCMAKE_CXX_COMPILER=/usr/local/bin/clang++ \
-DCMAKE_C_COMPILER=/usr/local/bin/clang \
-DCMAKE_CXX_FLAGS_INIT="-std=c++20 -stdlib=libc++ -march=native -mtune=native" \
-G Ninja ..These results were obtained in an AMD Threadripper 3960X (Zen2) at 3.8 GHz running Ubuntu 20.04.5 LTS.
IceLake Xeon Platinum 8375C @ 2.90GHz (AWS C6i instance) - no mitigations
