Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Sign up
Appearance settings

Add support for unbounded look-behind expressions #1266

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Multimodcrafter wants to merge 68 commits into rust-lang:master
base: master
Choose a base branch
Loading
from epfl-systemf:captureless-lookbehinds

Conversation

@Multimodcrafter
Copy link

@Multimodcrafter Multimodcrafter commented May 15, 2025

As an example consider the regex (?<=Title:\s+)\w+ which would match the
following strings (matches underlined with ~):

Title: HelloWorld
 ~~~~~~~~~~

Title: Title: foo
 ~~~~~
 ~~~

But does not match:

  • No heading
  • title: bad case
  • Title:nospace

What

This PR implements the streaming algorithm from
Linear Matching of JavaScript Regular Expressions (Section 4.4)
for unbounded look-behinds. The same algorithm has been
implemented and merged into V8.
The addition of look-around expressions to this crate was mentioned previously
in #1153.

This PR adds support for positive and negative look-behinds with arbitrary
nesting. With the following limitations

Limitations

  • Look-behind expressions cannot contain capture groups
  • The algorithm is implemented only in the PikeVM and with prefiters off
  • Only look-behinds and no look-aheads

Capture groups outside of look-arounds are supported. With the current capture
group semantics, no linear time algorithm which would allow for capture groups
inside of look-arounds is known. However, look-behinds could be implemented in
other engines and with prefilters on. Look-aheads could also be implemented with
additional memory.

How

We implemented the streaming algorithm presented in Section 4.4 of the paper
mentioned above. The algorithm works by running the sub-automata for any
look-behind expressions in parallel to the main automaton. This is achieved by
compiling the look-behind expressions as usual but storing their start states
separately, not reachable from the main automaton.

Instead of a match state, the sub-automata for look-behinds have a
WriteLookAround state. This state causes the current position in the haystack
to be recorded in a global look-around table.

The main automaton (and the sub-automata in the case of nested look-behinds) can
then read from this table by means of a CheckLookAround instruction and
compare the stored index with the current position in the haystack. These states
work as conditional epsilon transitions, similar to the already supported "look"
assertions (e.g. ^, \b, $).

PikeVM's cache has been expanded to preserve good performance of single-match
searches (stop the look-around threads once the main automaton finishes) and of
all-matches searches (remember the look-around states when resuming a search to
prevent having to rescan the haystack from the beginning).

Testing

We have added unit tests for the new functionality in the individual test
modules to test the new parsing, translation, and compilation features. We have
further added integration tests in the form of a new toml file. All engines
apart from the PikeVM will reject look-behind expressions. Thus tests containing
look-around expressions are filtered out for engines other than the PikeVM and
Meta engine.

Future Work

We would love to get feedback on the implementation.

The next steps are to work on the current limitations. Namely, implement support
in more engines and enable prefilters. Additionally, support for look-aheads
would be implemented if the additional memory cost is acceptable.

We are open to the discussion about any of the above.

Performance

We forked rebar and added a new engine
definition (rust/regex-lookbehind) for our fork of regex. We added this new
engine definition to all benchmarks where rust/regex was already present.
Furthermore, we added some benchmark definitions to measure the performance
of the look-behind algorithm.

We ran the full suite of benchmarks twice and merged the results. They are available
in our rebar fork (results_full_combined.csv)

Results without look-behinds

The results from all benchmarks without look-behinds show that our changes do not
introduce a significant slowdown for regexes that were already supported:

$ rebar rank results_full_combined.csv -e 'rust[^2]*$' -F 'lookbehind'
Engine Version Geometric mean of speed ratios Benchmark count
------ ------- ------------------------------ ---------------
rust/regex 1.11.0 1.01 341
rust/regex-lookbehind 1.12.0 1.03 341

Note: We noticed a discrepancy across multiple runs of up to 1.51 when comparing the
current version of rust/regex:

$ rebar cmp results_full_combined.csv -e 'rust/regex(|-run2)$' -F 'lookbehind' -t 1.5
benchmark rust/regex rust/regex-run2
--------- ---------- ---------------
hyperscan/literal-russian-som 36.3 GB/s (1.51x) 54.9 GB/s (1.00x)

Due to this result, we conclude that, despite the highest speedup ratio being 1.57 when
comparing both engines across both runs, the results of all individual benchmarks
further strengthen the claim that our changes do not significantly impact performance.

Full benchmark comparison (without look-behinds) 
$ rebar cmp results_full_combined.csv -e 'rust' -F 'lookbehind'
benchmark rust/regex rust/regex-lookbehind rust/regex-lookbehind-run2 rust/regex-run2
--------- ---------- --------------------- -------------------------- ---------------
captures/contiguous-letters 9.0 MB/s (1.34x) 12.0 MB/s (1.00x) 12.0 MB/s (1.00x) 12.0 MB/s (1.00x)
curated/01-literal/sherlock-en 43.5 GB/s (1.01x) 43.8 GB/s (1.00x) 43.6 GB/s (1.01x) 43.8 GB/s (1.00x)
curated/01-literal/sherlock-casei-en 12.0 GB/s (1.00x) 11.1 GB/s (1.08x) 11.9 GB/s (1.01x) 12.0 GB/s (1.00x)
curated/01-literal/sherlock-ru 42.3 GB/s (1.01x) 30.6 GB/s (1.40x) 42.1 GB/s (1.02x) 42.8 GB/s (1.00x)
curated/01-literal/sherlock-casei-ru 9.5 GB/s (1.01x) 9.6 GB/s (1.00x) 9.6 GB/s (1.00x) 9.5 GB/s (1.01x)
curated/01-literal/sherlock-zh 50.0 GB/s (1.00x) 50.1 GB/s (1.00x) 50.0 GB/s (1.00x) 49.9 GB/s (1.00x)
curated/02-literal-alternate/sherlock-en 13.4 GB/s (1.00x) 13.3 GB/s (1.01x) 13.3 GB/s (1.01x) 13.4 GB/s (1.00x)
curated/02-literal-alternate/sherlock-casei-en 3.3 GB/s (1.00x) 3.2 GB/s (1.01x) 3.2 GB/s (1.02x) 3.3 GB/s (1.00x)
curated/02-literal-alternate/sherlock-ru 7.4 GB/s (1.00x) 7.3 GB/s (1.01x) 7.2 GB/s (1.02x) 7.3 GB/s (1.00x)
curated/02-literal-alternate/sherlock-casei-ru 1774.2 MB/s (1.00x) 1772.0 MB/s (1.00x) 1780.0 MB/s (1.00x) 1769.2 MB/s (1.01x)
curated/02-literal-alternate/sherlock-zh 15.2 GB/s (1.00x) 15.2 GB/s (1.00x) 15.2 GB/s (1.00x) 15.1 GB/s (1.01x)
curated/03-date/ascii 159.2 MB/s (1.01x) 160.2 MB/s (1.00x) 160.2 MB/s (1.00x) 159.2 MB/s (1.01x)
curated/03-date/unicode 159.2 MB/s (1.00x) 159.2 MB/s (1.00x) 159.2 MB/s (1.00x) 159.2 MB/s (1.00x)
curated/03-date/compile-ascii 1.25ms (1.00x) 1.26ms (1.01x) 1.29ms (1.03x) 1.26ms (1.01x)
curated/03-date/compile-unicode 4.69ms (1.04x) 4.51ms (1.00x) 4.65ms (1.03x) 4.80ms (1.06x)
curated/04-ruff-noqa/real 1695.4 MB/s (1.02x) 1728.4 MB/s (1.00x) 1711.3 MB/s (1.01x) 1677.0 MB/s (1.03x)
curated/04-ruff-noqa/tweaked 1519.3 MB/s (1.06x) 1614.2 MB/s (1.00x) 1598.4 MB/s (1.01x) 1498.0 MB/s (1.08x)
curated/04-ruff-noqa/compile-real 51.83us (1.00x) 52.13us (1.01x) 53.41us (1.03x) 52.03us (1.00x)
curated/05-lexer-veryl/single 10.0 MB/s (1.03x) 10.2 MB/s (1.00x) 10.2 MB/s (1.00x) 10.0 MB/s (1.03x)
curated/05-lexer-veryl/compile-single 241.25us (1.00x) 244.29us (1.01x) 244.16us (1.01x) 242.08us (1.00x)
curated/05-lexer-veryl/multi 74.8 MB/s (1.02x) 76.4 MB/s (1.00x) 76.0 MB/s (1.01x) 74.4 MB/s (1.03x)
curated/06-cloud-flare-redos/original 629.9 MB/s (1.00x) 629.9 MB/s (1.00x) 629.9 MB/s (1.00x) 629.9 MB/s (1.00x)
curated/06-cloud-flare-redos/simplified-short 1737.0 MB/s (1.00x) 1737.0 MB/s (1.00x) 1706.6 MB/s (1.02x) 1706.6 MB/s (1.02x)
curated/06-cloud-flare-redos/simplified-long 78.3 GB/s (1.00x) 77.6 GB/s (1.01x) 77.0 GB/s (1.02x) 77.6 GB/s (1.01x)
curated/07-unicode-character-data/parse-line 400.2 MB/s (1.00x) 378.6 MB/s (1.06x) 374.0 MB/s (1.07x) 398.5 MB/s (1.00x)
curated/07-unicode-character-data/compile 24.39us (1.00x) 25.19us (1.03x) 24.90us (1.02x) 24.49us (1.00x)
curated/08-words/all-english 123.9 MB/s (1.00x) 123.1 MB/s (1.01x) 122.3 MB/s (1.01x) 123.9 MB/s (1.00x)
curated/08-words/all-russian 23.8 MB/s (1.00x) 16.3 MB/s (1.46x) 16.3 MB/s (1.46x) 23.8 MB/s (1.00x)
curated/08-words/long-english 891.9 MB/s (1.00x) 889.0 MB/s (1.00x) 888.6 MB/s (1.00x) 892.4 MB/s (1.00x)
curated/08-words/long-russian 42.8 MB/s (1.00x) 34.1 MB/s (1.26x) 33.9 MB/s (1.27x) 43.0 MB/s (1.00x)
curated/09-aws-keys/full 1992.8 MB/s (1.01x) 2016.1 MB/s (1.00x) 1991.5 MB/s (1.01x) 1970.0 MB/s (1.02x)
curated/09-aws-keys/quick 1941.7 MB/s (1.02x) 1975.1 MB/s (1.00x) 1947.8 MB/s (1.01x) 1921.2 MB/s (1.03x)
curated/09-aws-keys/compile-full 80.69us (1.00x) 83.01us (1.03x) 82.92us (1.03x) 83.60us (1.04x)
curated/09-aws-keys/compile-quick 13.69us (1.02x) 13.96us (1.04x) 13.44us (1.00x) 13.58us (1.01x)
curated/10-bounded-repeat/letters-en 773.2 MB/s (1.05x) 796.2 MB/s (1.02x) 811.4 MB/s (1.00x) 768.7 MB/s (1.06x)
curated/10-bounded-repeat/letters-ru 710.2 MB/s (1.00x) 709.2 MB/s (1.00x) 706.0 MB/s (1.01x) 709.8 MB/s (1.00x)
curated/10-bounded-repeat/context 115.6 MB/s (1.00x) 113.1 MB/s (1.02x) 112.6 MB/s (1.03x) 115.3 MB/s (1.00x)
curated/10-bounded-repeat/capitals 917.0 MB/s (1.00x) 917.0 MB/s (1.00x) 913.4 MB/s (1.00x) 913.4 MB/s (1.00x)
curated/10-bounded-repeat/compile-context 51.87us (1.00x) 53.62us (1.03x) 53.10us (1.02x) 52.62us (1.01x)
curated/10-bounded-repeat/compile-capitals 52.49us (1.00x) 54.09us (1.03x) 53.82us (1.03x) 53.31us (1.02x)
curated/11-unstructured-to-json/extract 123.0 MB/s (1.00x) 123.4 MB/s (1.00x) 123.3 MB/s (1.00x) 123.2 MB/s (1.00x)
curated/11-unstructured-to-json/compile 17.32us (1.02x) 17.57us (1.04x) 17.70us (1.04x) 16.96us (1.00x)
curated/12-dictionary/single 784.7 MB/s (1.01x) 789.6 MB/s (1.00x) 786.7 MB/s (1.00x) 783.7 MB/s (1.01x)
curated/12-dictionary/multi 206.6 MB/s (1.00x) 199.3 MB/s (1.04x) 200.0 MB/s (1.03x) 206.4 MB/s (1.00x)
curated/12-dictionary/compile-single 6.79ms (1.00x) 6.80ms (1.00x) 6.79ms (1.00x) 6.79ms (1.00x)
curated/12-dictionary/compile-multi 13.34ms (1.09x) 12.29ms (1.01x) 12.22ms (1.00x) 13.46ms (1.10x)
curated/13-noseyparker/single 138.8 MB/s (1.00x) 136.5 MB/s (1.02x) 135.7 MB/s (1.02x) 137.0 MB/s (1.01x)
curated/13-noseyparker/multi 111.9 MB/s (1.00x) 111.3 MB/s (1.01x) 112.1 MB/s (1.00x) 111.4 MB/s (1.01x)
curated/13-noseyparker/compile-single 2.02ms (1.00x) 2.04ms (1.01x) 2.09ms (1.03x) 2.03ms (1.00x)
curated/13-noseyparker/compile-multi 2.36ms (1.00x) 2.39ms (1.01x) 2.45ms (1.04x) 2.45ms (1.04x)
curated/14-quadratic/1x 20.0 MB/s (1.00x) 19.8 MB/s (1.01x) 19.8 MB/s (1.01x) 20.0 MB/s (1.00x)
curated/14-quadratic/2x 9.6 MB/s (1.00x) 9.5 MB/s (1.01x) 9.5 MB/s (1.01x) 9.6 MB/s (1.00x)
curated/14-quadratic/10x 1902.8 KB/s (1.00x) 1892.6 KB/s (1.01x) 1884.7 KB/s (1.01x) 1902.2 KB/s (1.00x)
dictionary/compile/english 192.78ms (1.01x) 190.18ms (1.00x) 195.72ms (1.03x) 199.17ms (1.05x)
dictionary/compile/english-10 66.95ms (1.00x) 72.86ms (1.09x) 66.67ms (1.00x) 69.49ms (1.04x)
dictionary/compile/english-15 6.77ms (1.00x) 6.78ms (1.00x) 6.80ms (1.00x) 6.78ms (1.00x)
dictionary/search/english 113.0 MB/s (1.00x) 108.7 MB/s (1.04x) 109.2 MB/s (1.03x) 111.7 MB/s (1.01x)
dictionary/search/english-tiny 196.6 MB/s (1.02x) 195.4 MB/s (1.02x) 200.0 MB/s (1.00x) 198.8 MB/s (1.01x)
dictionary/search/english-10 180.5 MB/s (1.00x) 175.9 MB/s (1.03x) 175.0 MB/s (1.03x) 180.0 MB/s (1.00x)
dictionary/search/english-15 783.1 MB/s (1.00x) 785.4 MB/s (1.00x) 785.3 MB/s (1.00x) 783.2 MB/s (1.00x)
folly/awyer-inn-busted 179.0 MB/s (1.00x) 177.6 MB/s (1.01x) 177.3 MB/s (1.01x) 178.6 MB/s (1.00x)
folly/literal-never-match-rare 54.2 GB/s (1.01x) 54.7 GB/s (1.00x) 52.9 GB/s (1.03x) 52.4 GB/s (1.04x)
folly/literal-never-match-frequent 54.4 GB/s (1.00x) 52.3 GB/s (1.04x) 53.9 GB/s (1.01x) 51.8 GB/s (1.05x)
folly/literal-never-match-tricksy 7.6 GB/s (1.01x) 7.5 GB/s (1.02x) 7.7 GB/s (1.00x) 7.7 GB/s (1.00x)
grep/every-line 2.2 GB/s (1.00x) 2.2 GB/s (1.02x) 2.2 GB/s (1.02x) 2.2 GB/s (1.00x)
grep/long-words-ascii 930.3 MB/s (1.00x) 931.5 MB/s (1.00x) 931.5 MB/s (1.00x) 926.6 MB/s (1.01x)
grep/long-words-unicode 899.4 MB/s (1.00x) 898.3 MB/s (1.00x) 896.0 MB/s (1.00x) 897.1 MB/s (1.00x)
hyperscan/literal-english-nosom 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x)
hyperscan/literal-english-som 55.0 GB/s (1.00x) 54.8 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x)
hyperscan/literal-casei-english-nosom 16.4 GB/s (1.02x) 16.4 GB/s (1.02x) 16.4 GB/s (1.02x) 16.8 GB/s (1.00x)
hyperscan/literal-casei-english-som 16.8 GB/s (1.00x) 16.4 GB/s (1.02x) 16.4 GB/s (1.02x) 16.7 GB/s (1.00x)
hyperscan/literal-russian-nosom 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x)
hyperscan/literal-russian-som 36.3 GB/s (1.51x) 54.8 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x)
hyperscan/literal-casei-russian-nosom 13.1 GB/s (1.01x) 13.3 GB/s (1.00x) 13.2 GB/s (1.00x) 13.2 GB/s (1.00x)
hyperscan/literal-casei-russian-som 13.1 GB/s (1.01x) 13.3 GB/s (1.00x) 13.3 GB/s (1.00x) 13.2 GB/s (1.01x)
hyperscan/literal-suffix-nosom 20.5 GB/s (1.00x) 17.4 GB/s (1.18x) 20.4 GB/s (1.01x) 20.5 GB/s (1.00x)
hyperscan/literal-suffix-som 17.8 GB/s (1.01x) 17.7 GB/s (1.01x) 17.9 GB/s (1.00x) 17.9 GB/s (1.00x)
hyperscan/literal-inner-nosom 21.9 GB/s (1.01x) 22.0 GB/s (1.00x) 22.0 GB/s (1.00x) 22.0 GB/s (1.00x)
hyperscan/literal-inner-som 17.6 GB/s (1.18x) 20.8 GB/s (1.00x) 20.7 GB/s (1.00x) 20.5 GB/s (1.01x)
hyperscan/fixed-length-words-nosom 917.8 MB/s (1.00x) 914.6 MB/s (1.00x) 914.1 MB/s (1.00x) 917.6 MB/s (1.00x)
hyperscan/fixed-length-words-som 912.0 MB/s (1.00x) 908.9 MB/s (1.00x) 908.6 MB/s (1.00x) 912.0 MB/s (1.00x)
hyperscan/fixed-length-words-unicode-nosom 917.7 MB/s (1.00x) 917.5 MB/s (1.00x) 917.4 MB/s (1.00x) 917.5 MB/s (1.00x)
imported/leipzig/twain 45.4 GB/s (1.04x) 46.3 GB/s (1.01x) 46.9 GB/s (1.00x) 46.4 GB/s (1.01x)
imported/leipzig/twain-insensitive 16.1 GB/s (1.00x) 16.0 GB/s (1.01x) 15.8 GB/s (1.02x) 15.8 GB/s (1.02x)
imported/leipzig/shing 27.8 GB/s (1.00x) 22.1 GB/s (1.26x) 27.5 GB/s (1.01x) 27.8 GB/s (1.00x)
imported/leipzig/huck-saw 24.8 GB/s (1.00x) 24.8 GB/s (1.00x) 24.4 GB/s (1.02x) 24.5 GB/s (1.01x)
imported/leipzig/word-ending-nn 30.3 GB/s (1.01x) 30.5 GB/s (1.00x) 30.5 GB/s (1.00x) 30.5 GB/s (1.00x)
imported/leipzig/certain-long-strings-ending-x 6.3 GB/s (1.02x) 6.4 GB/s (1.00x) 6.3 GB/s (1.01x) 6.3 GB/s (1.02x)
imported/leipzig/tom-sawyer-huckle-finn 22.9 GB/s (1.00x) 22.8 GB/s (1.00x) 22.5 GB/s (1.01x) 22.7 GB/s (1.01x)
imported/leipzig/tom-sawyer-huckle-fin-insensitive 1399.8 MB/s (1.01x) 1414.1 MB/s (1.00x) 1407.6 MB/s (1.00x) 1394.7 MB/s (1.01x)
imported/leipzig/tom-sawyer-huckle-fin-prefix-short 21.3 GB/s (1.00x) 18.0 GB/s (1.18x) 21.1 GB/s (1.01x) 21.0 GB/s (1.01x)
imported/leipzig/tom-sawyer-huckle-fin-prefix-long 20.5 GB/s (1.00x) 20.4 GB/s (1.00x) 20.3 GB/s (1.01x) 20.3 GB/s (1.01x)
imported/leipzig/tom-river 18.7 GB/s (1.00x) 18.7 GB/s (1.00x) 18.6 GB/s (1.01x) 18.6 GB/s (1.00x)
imported/leipzig/ing 3.3 GB/s (1.00x) 3.3 GB/s (1.00x) 3.3 GB/s (1.01x) 3.3 GB/s (1.01x)
imported/leipzig/ing-whitespace 3.3 GB/s (1.00x) 3.2 GB/s (1.00x) 3.2 GB/s (1.01x) 3.2 GB/s (1.01x)
imported/leipzig/awyer-inn 16.3 GB/s (1.00x) 16.3 GB/s (1.00x) 16.2 GB/s (1.01x) 16.2 GB/s (1.01x)
imported/leipzig/quotes-bounded 4.8 GB/s (1.00x) 4.8 GB/s (1.00x) 4.8 GB/s (1.01x) 4.8 GB/s (1.01x)
imported/leipzig/non-ascii-alternate 25.8 GB/s (1.00x) 25.9 GB/s (1.00x) 25.8 GB/s (1.01x) 25.5 GB/s (1.02x)
imported/leipzig/math-symbols 914.5 MB/s (1.00x) 915.0 MB/s (1.00x) 909.1 MB/s (1.01x) 908.5 MB/s (1.01x)
imported/leipzig/bounded-strings-ending-z 44.8 GB/s (1.00x) 44.9 GB/s (1.00x) 44.4 GB/s (1.01x) 44.3 GB/s (1.01x)
imported/lh3lh3-reb/uri 2.5 GB/s (1.00x) 2.5 GB/s (1.00x) 2.5 GB/s (1.02x) 2.5 GB/s (1.01x)
imported/lh3lh3-reb/email 2.6 GB/s (1.01x) 2.6 GB/s (1.00x) 2.6 GB/s (1.01x) 2.5 GB/s (1.02x)
imported/lh3lh3-reb/date 2.0 GB/s (1.01x) 2.1 GB/s (1.00x) 2.0 GB/s (1.01x) 2.0 GB/s (1.03x)
imported/lh3lh3-reb/uri-or-email 867.7 MB/s (1.00x) 865.9 MB/s (1.00x) 858.1 MB/s (1.01x) 856.4 MB/s (1.01x)
imported/mariomka/email 49.6 GB/s (1.00x) 49.7 GB/s (1.00x) 49.2 GB/s (1.01x) 49.4 GB/s (1.01x)
imported/mariomka/uri 9.6 GB/s (1.00x) 9.4 GB/s (1.02x) 9.6 GB/s (1.00x) 9.5 GB/s (1.00x)
imported/mariomka/ip 3.5 GB/s (1.00x) 3.5 GB/s (1.00x) 3.5 GB/s (1.01x) 3.5 GB/s (1.01x)
imported/regex-redux/regex-redux 11.14ms (1.01x) 11.02ms (1.00x) 11.07ms (1.00x) 11.21ms (1.02x)
imported/rsc/no-exponential 567.7 MB/s (1.00x) 554.5 MB/s (1.02x) 554.5 MB/s (1.02x) 567.7 MB/s (1.00x)
imported/rsc/literal 2.3 GB/s (1.00x) 2.1 GB/s (1.10x) 2.1 GB/s (1.10x) 2.3 GB/s (1.00x)
imported/rsc/not-literal 1158.0 MB/s (1.02x) 1158.0 MB/s (1.02x) 1158.0 MB/s (1.02x) 1186.3 MB/s (1.00x)
imported/rsc/match-class 2.4 GB/s (1.00x) 2.4 GB/s (1.00x) 2.4 GB/s (1.00x) 2.4 GB/s (1.00x)
imported/rsc/match-class-in-range 3.4 GB/s (1.00x) 3.3 GB/s (1.05x) 3.3 GB/s (1.05x) 3.4 GB/s (1.00x)
imported/rsc/match-class-unicode 791.5 MB/s (1.01x) 779.4 MB/s (1.02x) 779.4 MB/s (1.02x) 795.6 MB/s (1.00x)
imported/rsc/anchored-literal-short-non-match 1239.8 MB/s (1.05x) 1239.8 MB/s (1.05x) 1180.7 MB/s (1.11x) 1305.0 MB/s (1.00x)
imported/rsc/anchored-literal-long-non-match 20.2 GB/s (1.00x) 18.2 GB/s (1.11x) 17.3 GB/s (1.17x) 19.1 GB/s (1.06x)
imported/rsc/anchored-literal-short-match 885.6 MB/s (1.00x) 855.0 MB/s (1.04x) 855.0 MB/s (1.04x) 855.0 MB/s (1.04x)
imported/rsc/anchored-literal-long-match 13.0 GB/s (1.00x) 13.0 GB/s (1.00x) 12.5 GB/s (1.04x) 13.0 GB/s (1.00x)
imported/rsc/one-pass-short 405.3 MB/s (1.03x) 415.7 MB/s (1.00x) 405.3 MB/s (1.03x) 415.7 MB/s (1.00x)
imported/rsc/one-pass-short-not 344.9 MB/s (1.00x) 330.9 MB/s (1.04x) 324.2 MB/s (1.06x) 344.9 MB/s (1.00x)
imported/rsc/one-pass-long-prefix 506.0 MB/s (1.00x) 467.8 MB/s (1.08x) 467.8 MB/s (1.08x) 495.9 MB/s (1.02x)
imported/rsc/one-pass-long-prefix-not 506.0 MB/s (1.00x) 467.8 MB/s (1.08x) 467.8 MB/s (1.08x) 506.0 MB/s (1.00x)
imported/rsc/long-needle1 56.1 GB/s (1.00x) 53.8 GB/s (1.04x) 53.8 GB/s (1.04x) 56.1 GB/s (1.00x)
imported/rsc/long-needle2 372.9 MB/s (1.02x) 377.5 MB/s (1.01x) 378.0 MB/s (1.01x) 380.7 MB/s (1.00x)
imported/rsc/easy0-32 2.3 GB/s (1.00x) 2.2 GB/s (1.09x) 2.2 GB/s (1.09x) 2.3 GB/s (1.04x)
imported/rsc/easy0-1k 23.9 GB/s (1.00x) 23.3 GB/s (1.02x) 21.7 GB/s (1.10x) 23.3 GB/s (1.02x)
imported/rsc/easy0-32k 54.5 GB/s (1.00x) 53.7 GB/s (1.02x) 53.7 GB/s (1.02x) 54.5 GB/s (1.00x)
imported/rsc/easy0-1mb 55.0 GB/s (1.00x) 55.0 GB/s (1.00x) 55.0 GB/s (1.00x) 55.0 GB/s (1.00x)
imported/rsc/easy1-32 1034.8 MB/s (1.02x) 992.6 MB/s (1.07x) 992.6 MB/s (1.07x) 1057.3 MB/s (1.00x)
imported/rsc/easy1-1k 20.7 GB/s (1.00x) 19.4 GB/s (1.06x) 19.4 GB/s (1.06x) 20.7 GB/s (1.00x)
imported/rsc/easy1-32k 649.7 GB/s (1.00x) 623.2 GB/s (1.04x) 623.2 GB/s (1.04x) 623.2 GB/s (1.04x)
imported/rsc/easy1-1mb 21230.0 GB/s (1.00x) 19930.2 GB/s (1.07x) 19531.6 GB/s (1.09x) 21230.0 GB/s (1.00x)
imported/rsc/medium-32 1042.0 MB/s (1.00x) 987.1 MB/s (1.06x) 987.1 MB/s (1.06x) 1023.0 MB/s (1.02x)
imported/rsc/medium-1k 17.8 GB/s (1.02x) 17.2 GB/s (1.06x) 17.2 GB/s (1.06x) 18.1 GB/s (1.00x)
imported/rsc/medium-32k 565.6 GB/s (1.00x) 535.8 GB/s (1.06x) 535.8 GB/s (1.06x) 555.3 GB/s (1.02x)
imported/rsc/medium-1mb 18085.0 GB/s (1.00x) 17133.1 GB/s (1.06x) 17133.1 GB/s (1.06x) 17756.1 GB/s (1.02x)
imported/rsc/hard-32 582.2 MB/s (1.00x) 564.4 MB/s (1.03x) 564.4 MB/s (1.03x) 582.2 MB/s (1.00x)
imported/rsc/hard-1k 17.8 GB/s (1.00x) 16.9 GB/s (1.05x) 16.9 GB/s (1.05x) 17.5 GB/s (1.02x)
imported/rsc/hard-32k 555.3 GB/s (1.00x) 526.6 GB/s (1.05x) 526.6 GB/s (1.05x) 555.3 GB/s (1.00x)
imported/rsc/hard-1mb 10279.9 GB/s (1.01x) 10067.9 GB/s (1.03x) 10067.9 GB/s (1.03x) 10389.2 GB/s (1.00x)
imported/rsc/reallyhard0-32 397.9 MB/s (1.00x) 386.8 MB/s (1.03x) 386.8 MB/s (1.03x) 392.3 MB/s (1.01x)
imported/rsc/reallyhard0-1k 11.5 GB/s (1.00x) 11.2 GB/s (1.02x) 10.7 GB/s (1.07x) 11.1 GB/s (1.04x)
imported/rsc/reallyhard0-32k 49.9 GB/s (1.00x) 47.2 GB/s (1.06x) 47.2 GB/s (1.06x) 49.9 GB/s (1.00x)
imported/rsc/reallyhard0-1mb 52.5 GB/s (1.04x) 36.0 GB/s (1.51x) 54.5 GB/s (1.00x) 53.4 GB/s (1.02x)
imported/rsc/reallyreallyhard0-32 317.9 MB/s (1.00x) 313.4 MB/s (1.01x) 315.7 MB/s (1.01x) 317.9 MB/s (1.00x)
imported/rsc/reallyreallyhard0-1k 10.6 GB/s (1.01x) 10.8 GB/s (1.00x) 10.8 GB/s (1.00x) 10.6 GB/s (1.01x)
imported/rsc/reallyreallyhard0-32k 48.7 GB/s (1.00x) 48.7 GB/s (1.00x) 48.8 GB/s (1.00x) 48.8 GB/s (1.00x)
imported/rsc/reallyreallyhard0-1mb 56.8 GB/s (1.00x) 56.8 GB/s (1.00x) 56.8 GB/s (1.00x) 56.8 GB/s (1.00x)
imported/rsc/reallyreallyreallyhard0-32 330.9 MB/s (1.01x) 330.9 MB/s (1.01x) 330.9 MB/s (1.01x) 333.1 MB/s (1.00x)
imported/rsc/reallyreallyreallyhard0-1k 10.2 GB/s (1.08x) 11.0 GB/s (1.00x) 10.9 GB/s (1.01x) 10.9 GB/s (1.01x)
imported/rsc/reallyreallyreallyhard0-32k 47.1 GB/s (1.07x) 50.3 GB/s (1.00x) 50.1 GB/s (1.00x) 47.1 GB/s (1.07x)
imported/rsc/reallyreallyreallyhard0-1mb 56.8 GB/s (1.00x) 56.8 GB/s (1.00x) 56.4 GB/s (1.01x) 56.8 GB/s (1.00x)
imported/sherlock/name-sherlock 48.3 GB/s (1.00x) 48.3 GB/s (1.00x) 48.2 GB/s (1.00x) 48.3 GB/s (1.00x)
imported/sherlock/name-holmes 36.7 GB/s (1.00x) 27.5 GB/s (1.34x) 36.8 GB/s (1.00x) 36.5 GB/s (1.01x)
imported/sherlock/name-sherlock-holmes 48.5 GB/s (1.00x) 48.5 GB/s (1.00x) 48.5 GB/s (1.00x) 48.6 GB/s (1.00x)
imported/sherlock/name-sherlock-casei 14.2 GB/s (1.01x) 14.1 GB/s (1.02x) 14.3 GB/s (1.00x) 14.2 GB/s (1.01x)
imported/sherlock/name-holmes-casei 9.7 GB/s (1.00x) 9.7 GB/s (1.00x) 9.8 GB/s (1.00x) 9.8 GB/s (1.00x)
imported/sherlock/name-sherlock-holmes-casei 13.9 GB/s (1.01x) 14.0 GB/s (1.00x) 14.0 GB/s (1.00x) 13.8 GB/s (1.01x)
imported/sherlock/name-whitespace 28.0 GB/s (1.33x) 37.1 GB/s (1.00x) 37.3 GB/s (1.00x) 37.2 GB/s (1.00x)
imported/sherlock/name-alt1 43.4 GB/s (1.00x) 43.3 GB/s (1.00x) 43.3 GB/s (1.00x) 43.3 GB/s (1.00x)
imported/sherlock/name-alt2 12.0 GB/s (1.02x) 12.0 GB/s (1.02x) 12.2 GB/s (1.00x) 12.1 GB/s (1.01x)
imported/sherlock/name-alt3 10.9 GB/s (1.01x) 10.9 GB/s (1.00x) 10.9 GB/s (1.00x) 10.8 GB/s (1.01x)
imported/sherlock/name-alt3-casei 1986.0 MB/s (1.00x) 1977.5 MB/s (1.00x) 1966.6 MB/s (1.01x) 1980.1 MB/s (1.00x)
imported/sherlock/name-alt4 10.9 GB/s (1.01x) 11.0 GB/s (1.00x) 11.0 GB/s (1.00x) 11.0 GB/s (1.00x)
imported/sherlock/name-alt4-casei 5.9 GB/s (1.00x) 5.8 GB/s (1.02x) 5.8 GB/s (1.02x) 5.9 GB/s (1.00x)
imported/sherlock/name-alt5 11.5 GB/s (1.02x) 11.7 GB/s (1.00x) 11.7 GB/s (1.00x) 11.6 GB/s (1.01x)
imported/sherlock/name-alt5-casei 4.1 GB/s (1.02x) 4.2 GB/s (1.00x) 4.2 GB/s (1.00x) 4.1 GB/s (1.01x)
imported/sherlock/no-match-uncommon 38.0 GB/s (1.00x) 38.0 GB/s (1.00x) 38.0 GB/s (1.00x) 37.9 GB/s (1.00x)
imported/sherlock/no-match-common 37.9 GB/s (1.00x) 38.0 GB/s (1.00x) 37.9 GB/s (1.00x) 38.0 GB/s (1.00x)
imported/sherlock/no-match-really-common 44.8 GB/s (1.00x) 44.7 GB/s (1.00x) 44.6 GB/s (1.00x) 44.6 GB/s (1.00x)
imported/sherlock/the-lower 3.7 GB/s (1.03x) 3.7 GB/s (1.02x) 3.8 GB/s (1.01x) 3.8 GB/s (1.00x)
imported/sherlock/the-upper 25.7 GB/s (1.02x) 26.2 GB/s (1.00x) 26.1 GB/s (1.00x) 25.3 GB/s (1.04x)
imported/sherlock/the-casei 1941.9 MB/s (1.00x) 1937.0 MB/s (1.00x) 1938.7 MB/s (1.00x) 1940.0 MB/s (1.00x)
imported/sherlock/everything-greedy 294.0 MB/s (1.00x) 279.5 MB/s (1.05x) 279.5 MB/s (1.05x) 294.0 MB/s (1.00x)
imported/sherlock/everything-greedy-nl 394.0 MB/s (1.00x) 380.8 MB/s (1.03x) 380.8 MB/s (1.03x) 391.3 MB/s (1.01x)
imported/sherlock/letters 59.8 MB/s (1.00x) 59.7 MB/s (1.01x) 60.1 MB/s (1.00x) 59.5 MB/s (1.01x)
imported/sherlock/letters-upper 579.0 MB/s (1.00x) 579.0 MB/s (1.00x) 579.0 MB/s (1.00x) 579.0 MB/s (1.00x)
imported/sherlock/letters-lower 61.2 MB/s (1.00x) 60.9 MB/s (1.01x) 61.3 MB/s (1.00x) 60.7 MB/s (1.01x)
imported/sherlock/words 115.6 MB/s (1.04x) 120.5 MB/s (1.00x) 120.2 MB/s (1.00x) 119.2 MB/s (1.01x)
imported/sherlock/before-holmes 17.9 GB/s (1.00x) 17.9 GB/s (1.00x) 17.8 GB/s (1.01x) 17.9 GB/s (1.00x)
imported/sherlock/before-after-holmes 21.0 GB/s (1.00x) 20.8 GB/s (1.01x) 20.8 GB/s (1.01x) 17.6 GB/s (1.19x)
imported/sherlock/holmes-cochar-watson 10.5 GB/s (1.01x) 10.5 GB/s (1.00x) 10.6 GB/s (1.00x) 10.5 GB/s (1.01x)
imported/sherlock/holmes-coword-watson 1866.2 MB/s (1.00x) 1870.7 MB/s (1.00x) 1866.0 MB/s (1.00x) 1869.2 MB/s (1.00x)
imported/sherlock/quotes 2.9 GB/s (1.00x) 2.9 GB/s (1.00x) 2.9 GB/s (1.00x) 2.9 GB/s (1.01x)
imported/sherlock/line-boundary-sherlock-holmes 41.7 GB/s (1.00x) 41.6 GB/s (1.00x) 41.8 GB/s (1.00x) 41.7 GB/s (1.00x)
imported/sherlock/word-ending-n 899.8 MB/s (1.00x) 886.7 MB/s (1.01x) 890.2 MB/s (1.01x) 897.1 MB/s (1.00x)
imported/sherlock/repeated-class-negation 23.0 GB/s (1.03x) 23.5 GB/s (1.01x) 23.7 GB/s (1.00x) 23.7 GB/s (1.00x)
imported/sherlock/ing-suffix 3.3 GB/s (1.00x) 3.2 GB/s (1.01x) 3.2 GB/s (1.02x) 3.3 GB/s (1.00x)
imported/sherlock/ing-suffix-limited-space 3.5 GB/s (1.01x) 3.6 GB/s (1.00x) 3.6 GB/s (1.00x) 3.5 GB/s (1.00x)
opt/accelerate/whole-line 1422.7 MB/s (1.00x) 1414.1 MB/s (1.01x) 1411.3 MB/s (1.01x) 1425.6 MB/s (1.00x)
opt/accelerate/non-dna 1990.3 MB/s (1.01x) 2004.2 MB/s (1.00x) 1998.8 MB/s (1.00x) 1996.0 MB/s (1.00x)
opt/backtrack/words-english 232.1 MB/s (1.00x) 225.0 MB/s (1.03x) 225.0 MB/s (1.03x) 232.1 MB/s (1.00x)
opt/backtrack/words-russian 55.6 MB/s (1.02x) 56.5 MB/s (1.00x) 56.3 MB/s (1.00x) 55.9 MB/s (1.01x)
opt/fixed-length/too-small-ascii 357.6 MB/s (1.00x) 301.2 MB/s (1.19x) 301.2 MB/s (1.19x) 336.6 MB/s (1.06x)
opt/fixed-length/too-small-unicode 2.1 GB/s (1.00x) 1807.0 MB/s (1.19x) 1807.0 MB/s (1.19x) 2.1 GB/s (1.00x)
opt/fixed-length/too-big-ascii 3.0 GB/s (1.00x) 2.5 GB/s (1.19x) 2.4 GB/s (1.25x) 3.0 GB/s (1.00x)
opt/fixed-length/too-big-unicode 2.6 GB/s (1.00x) 2.2 GB/s (1.19x) 2.0 GB/s (1.25x) 2.4 GB/s (1.06x)
opt/fixed-length/go33484-1 582.1 GB/s (1.00x) 490.2 GB/s (1.19x) 465.7 GB/s (1.25x) 547.8 GB/s (1.06x)
opt/fixed-length/go33484-2 582.1 GB/s (1.00x) 490.2 GB/s (1.19x) 490.2 GB/s (1.19x) 582.1 GB/s (1.00x)
opt/fixed-length/go33484-3 58.2 GB/s (1.00x) 49.0 GB/s (1.19x) 49.0 GB/s (1.19x) 54.8 GB/s (1.06x)
opt/literal-alt/one-pattern 787.3 MB/s (1.00x) 776.8 MB/s (1.01x) 776.8 MB/s (1.01x) 787.3 MB/s (1.00x)
opt/literal-alt/pattern-per-word 454.7 MB/s (1.00x) 451.2 MB/s (1.01x) 456.4 MB/s (1.00x) 436.1 MB/s (1.05x)
opt/nfa-sparse/small-repeated-class-bytes 1496.2 KB/s (1.00x) 1402.5 KB/s (1.07x) 1392.0 KB/s (1.07x) 1341.1 KB/s (1.12x)
opt/nfa-sparse/small-repeated-class-unicode 1491.5 KB/s (1.00x) 1399.9 KB/s (1.07x) 1394.3 KB/s (1.07x) 1481.1 KB/s (1.01x)
opt/onepass/fn-predicate 1078.5 MB/s (1.05x) 1135.9 MB/s (1.00x) 1134.0 MB/s (1.00x) 1073.5 MB/s (1.06x)
opt/onepass/first-three-words-english 557.1 MB/s (1.02x) 567.9 MB/s (1.00x) 567.9 MB/s (1.00x) 557.1 MB/s (1.02x)
opt/onepass/first-three-words-russian 726.0 MB/s (1.01x) 736.8 MB/s (1.00x) 735.8 MB/s (1.00x) 723.7 MB/s (1.02x)
opt/onepass/word-boundary-english 901.1 MB/s (1.04x) 934.4 MB/s (1.00x) 936.3 MB/s (1.00x) 895.1 MB/s (1.05x)
opt/onepass/word-boundary-russian 1160.4 MB/s (1.04x) 1207.8 MB/s (1.00x) 1205.7 MB/s (1.00x) 1157.6 MB/s (1.04x)
opt/prefilter/literal-english 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x)
opt/prefilter/literal-casei-english 16.8 GB/s (1.00x) 16.4 GB/s (1.02x) 16.4 GB/s (1.02x) 16.8 GB/s (1.00x)
opt/prefilter/literal-russian 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x) 54.9 GB/s (1.00x)
opt/prefilter/literal-casei-russian 13.2 GB/s (1.01x) 13.3 GB/s (1.00x) 13.3 GB/s (1.00x) 13.3 GB/s (1.00x)
opt/prefilter/rust-functions 15.7 GB/s (1.00x) 15.7 GB/s (1.00x) 15.7 GB/s (1.00x) 15.7 GB/s (1.00x)
opt/reverse-anchored/word-end 33261.6 GB/s (1.00x) 31044.1 GB/s (1.07x) 30042.7 GB/s (1.11x) 32114.6 GB/s (1.04x)
opt/reverse-inner/holmes 20.9 GB/s (1.01x) 21.0 GB/s (1.00x) 20.9 GB/s (1.01x) 21.0 GB/s (1.00x)
opt/reverse-inner/email 81.9 GB/s (1.01x) 82.8 GB/s (1.00x) 83.0 GB/s (1.00x) 82.4 GB/s (1.01x)
opt/reverse-inner/factored-prefix 9.7 GB/s (1.01x) 9.7 GB/s (1.01x) 9.7 GB/s (1.01x) 9.9 GB/s (1.00x)
opt/reverse-inner/no-quadratic-backward 918.1 MB/s (1.00x) 913.7 MB/s (1.00x) 914.8 MB/s (1.00x) 918.1 MB/s (1.00x)
opt/reverse-inner/no-quadratic-forward 459.9 MB/s (1.00x) 459.0 MB/s (1.00x) 459.0 MB/s (1.00x) 459.9 MB/s (1.00x)
opt/reverse-suffix/holmes 17.8 GB/s (1.01x) 17.9 GB/s (1.00x) 17.9 GB/s (1.01x) 17.9 GB/s (1.00x)
opt/reverse-suffix/no-quadratic 915.9 MB/s (1.00x) 912.6 MB/s (1.00x) 912.6 MB/s (1.00x) 915.9 MB/s (1.00x)
reported/i1095-word-repetition/unicode-compile 28.57ms (1.06x) 27.00ms (1.00x) 27.53ms (1.02x) 29.18ms (1.08x)
reported/i1095-word-repetition/unicode-search 868.3 KB/s (1.13x) 779.1 KB/s (1.26x) 882.5 KB/s (1.11x) 981.5 KB/s (1.00x)
reported/i1095-word-repetition/ascii-compile 34.88us (1.00x) 35.19us (1.01x) 35.33us (1.02x) 34.71us (1.00x)
reported/i1095-word-repetition/ascii-search 448.0 MB/s (1.00x) 446.3 MB/s (1.01x) 445.5 MB/s (1.01x) 448.8 MB/s (1.00x)
reported/i13-subset-regex/original-ascii 6.3 GB/s (1.02x) 6.4 GB/s (1.00x) 6.4 GB/s (1.00x) 6.3 GB/s (1.01x)
reported/i13-subset-regex/original-unicode 377.4 MB/s (1.00x) 365.0 MB/s (1.03x) 365.1 MB/s (1.03x) 376.1 MB/s (1.00x)
reported/i13-subset-regex/big-ascii 99.3 MB/s (1.00x) 94.5 MB/s (1.05x) 92.9 MB/s (1.07x) 98.7 MB/s (1.01x)
reported/i13-subset-regex/big-unicode 35.3 MB/s (1.04x) 33.3 MB/s (1.10x) 36.7 MB/s (1.00x) 35.3 MB/s (1.04x)
reported/i13-subset-regex/huge-ascii 15.1 MB/s (1.10x) 15.9 MB/s (1.04x) 15.7 MB/s (1.06x) 16.6 MB/s (1.00x)
reported/i13-subset-regex/huge-unicode 16.8 MB/s (1.00x) 15.7 MB/s (1.07x) 15.7 MB/s (1.07x) 16.7 MB/s (1.01x)
reported/i13-subset-regex/huge-ascii-nosuffixlit 16.8 MB/s (1.00x) 15.9 MB/s (1.06x) 15.7 MB/s (1.07x) 15.9 MB/s (1.06x)
reported/i13-subset-regex/huge-unicode-nosuffixlit 16.9 MB/s (1.00x) 15.9 MB/s (1.06x) 15.9 MB/s (1.06x) 16.7 MB/s (1.01x)
reported/i787-keywords/compile 348.12us (1.00x) 350.08us (1.01x) 353.12us (1.01x) 348.11us (1.00x)
reported/i787-keywords/ascii 209.5 MB/s (1.00x) 208.0 MB/s (1.01x) 207.5 MB/s (1.01x) 209.4 MB/s (1.00x)
reported/i787-keywords/unicode 209.4 MB/s (1.00x) 207.4 MB/s (1.01x) 207.1 MB/s (1.01x) 208.4 MB/s (1.00x)
reported/i787-keywords/opt-ascii 417.4 MB/s (1.00x) 415.6 MB/s (1.00x) 415.4 MB/s (1.00x) 416.0 MB/s (1.00x)
reported/i787-keywords/opt-unicode 415.5 MB/s (1.00x) 413.9 MB/s (1.01x) 413.7 MB/s (1.01x) 416.6 MB/s (1.00x)
reported/i988-cloudflare-compile/javascript-obfuscation 14.38us (1.00x) 14.92us (1.04x) 14.87us (1.03x) 14.45us (1.00x)
reported/i988-cloudflare-compile/sql-injection 142.19us (1.00x) 144.06us (1.01x) 145.88us (1.03x) 142.43us (1.00x)
slow/quadratic-regex-1x 9.9 MB/s (1.00x) 9.9 MB/s (1.00x) 9.9 MB/s (1.00x) 9.9 MB/s (1.00x)
slow/quadratic-regex-2x 4.8 MB/s (1.00x) 4.8 MB/s (1.00x) 4.8 MB/s (1.00x) 4.8 MB/s (1.00x)
slow/quadratic-haystack-1x 20.0 MB/s (1.00x) 19.8 MB/s (1.01x) 19.8 MB/s (1.01x) 20.0 MB/s (1.00x)
slow/quadratic-haystack-2x 9.6 MB/s (1.00x) 9.5 MB/s (1.01x) 9.5 MB/s (1.01x) 9.6 MB/s (1.00x)
test/dot/default-new-line 45.4 MB/s (1.00x) 41.5 MB/s (1.10x) 39.7 MB/s (1.14x) 45.4 MB/s (1.00x)
test/dot/default-carriage-return 41.5 MB/s (1.00x) 39.7 MB/s (1.04x) 38.1 MB/s (1.09x) 39.7 MB/s (1.04x)
test/dot/dotall-new-line 43.3 MB/s (1.00x) 38.1 MB/s (1.14x) 38.1 MB/s (1.14x) 41.5 MB/s (1.05x)
test/dot/dotall-carriage-return 41.5 MB/s (1.00x) 39.7 MB/s (1.04x) 39.7 MB/s (1.04x) 41.5 MB/s (1.00x)
test/dot/multiline-new-line 45.4 MB/s (1.00x) 39.7 MB/s (1.14x) 39.7 MB/s (1.14x) 43.3 MB/s (1.05x)
test/dot/multiline-carriage-return 43.3 MB/s (1.00x) 39.7 MB/s (1.09x) 38.1 MB/s (1.14x) 41.5 MB/s (1.05x)
test/dot/dotall-multiline-new-line 41.5 MB/s (1.00x) 38.1 MB/s (1.09x) 36.7 MB/s (1.13x) 41.5 MB/s (1.00x)
test/dot/dotall-multiline-carriage-return 39.7 MB/s (1.04x) 38.1 MB/s (1.09x) 38.1 MB/s (1.09x) 41.5 MB/s (1.00x)
test/func/leftmost-first 303.4 MB/s (1.00x) 256.8 MB/s (1.18x) 256.8 MB/s (1.18x) 303.4 MB/s (1.00x)
test/func/dollar-only-matches-end 56.1 MB/s (1.00x) 53.0 MB/s (1.06x) 51.5 MB/s (1.09x) 56.1 MB/s (1.00x)
test/func/non-greedy 77.3 MB/s (1.00x) 75.3 MB/s (1.03x) 71.5 MB/s (1.08x) 73.4 MB/s (1.05x)
test/model/count 630.1 MB/s (1.00x) 598.1 MB/s (1.05x) 598.1 MB/s (1.05x) 630.1 MB/s (1.00x)
test/model/count-spans 476.8 MB/s (1.00x) 470.5 MB/s (1.01x) 470.5 MB/s (1.01x) 476.8 MB/s (1.00x)
test/model/count-captures 420.1 MB/s (1.00x) 415.1 MB/s (1.01x) 415.1 MB/s (1.01x) 420.1 MB/s (1.00x)
test/model/grep 375.7 MB/s (1.00x) 375.7 MB/s (1.00x) 364.6 MB/s (1.03x) 375.7 MB/s (1.00x)
test/model/grep-captures 106.0 MB/s (1.00x) 105.4 MB/s (1.01x) 105.4 MB/s (1.01x) 106.0 MB/s (1.00x)
test/model/compile 11.06us (1.01x) 10.97us (1.01x) 10.91us (1.00x) 11.13us (1.02x)
test/unicode/case/ascii-only 100.4 MB/s (1.00x) 90.8 MB/s (1.11x) 86.7 MB/s (1.16x) 100.4 MB/s (1.00x)
test/unicode/case/ascii-with-unicode 65.8 MB/s (1.00x) 61.5 MB/s (1.07x) 61.5 MB/s (1.07x) 65.8 MB/s (1.00x)
test/unicode/case/unicode 65.8 MB/s (1.00x) 61.5 MB/s (1.07x) 59.6 MB/s (1.10x) 65.8 MB/s (1.00x)
test/unicode/decimal/ascii-only 136.2 MB/s (1.00x) 124.4 MB/s (1.10x) 124.4 MB/s (1.10x) 130.0 MB/s (1.05x)
test/unicode/decimal/unicode 114.4 MB/s (1.00x) 102.2 MB/s (1.12x) 102.2 MB/s (1.12x) 114.4 MB/s (1.00x)
test/unicode/invalid-utf8/dot-matches-xFF 41.5 MB/s (1.00x) 39.7 MB/s (1.04x) 38.1 MB/s (1.09x) 41.5 MB/s (1.00x)
test/unicode/invalid-utf8/dot-no-matches-xFF 45.4 MB/s (1.00x) 39.7 MB/s (1.14x) 39.7 MB/s (1.14x) 45.4 MB/s (1.00x)
test/unicode/invalid-utf8/dot-matches-codepoint-prefix 77.3 MB/s (1.00x) 73.4 MB/s (1.05x) 71.5 MB/s (1.08x) 77.3 MB/s (1.00x)
test/unicode/invalid-utf8/dot-no-matches-codepoint-prefix 124.4 MB/s (1.00x) 110.0 MB/s (1.13x) 106.0 MB/s (1.17x) 114.4 MB/s (1.09x)
test/unicode/invalid-utf8/xFF-matches-xFF 50.2 MB/s (1.00x) 43.3 MB/s (1.16x) 43.3 MB/s (1.16x) 50.2 MB/s (1.00x)
test/unicode/letter/pL-matches-bmp-delta 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 70.6 MB/s (1.08x) 76.3 MB/s (1.00x)
test/unicode/letter/pLbraced-matches-bmp-delta 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 68.1 MB/s (1.12x) 76.3 MB/s (1.00x)
test/unicode/letter/pLbraced-matches-nonbmp-delta 146.7 MB/s (1.00x) 136.2 MB/s (1.08x) 127.2 MB/s (1.15x) 141.3 MB/s (1.04x)
test/unicode/letter/pLetter-matches-bmp-delta 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 68.1 MB/s (1.12x) 76.3 MB/s (1.00x)
test/unicode/letter/pLetter-casei-matches-bmp-delta 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 68.1 MB/s (1.12x) 73.4 MB/s (1.04x)
test/unicode/letter/pLetter-gc-equals-matches-bmp-delta 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 70.6 MB/s (1.08x) 76.3 MB/s (1.00x)
test/unicode/letter/pLetter-gc-colon-matches-bmp-delta 79.5 MB/s (1.00x) 70.6 MB/s (1.12x) 68.1 MB/s (1.17x) 76.3 MB/s (1.04x)
test/unicode/utf8/dot-matches-byte 84.8 MB/s (1.00x) 81.2 MB/s (1.04x) 82.9 MB/s (1.02x) 82.9 MB/s (1.02x)
test/unicode/utf8/dot-matches-codepoint 146.7 MB/s (1.00x) 136.2 MB/s (1.08x) 131.5 MB/s (1.12x) 146.7 MB/s (1.00x)
test/unicode/whitespace/ascii-only 136.2 MB/s (1.00x) 124.4 MB/s (1.10x) 119.2 MB/s (1.14x) 136.2 MB/s (1.00x)
test/unicode/whitespace/unicode 114.4 MB/s (1.00x) 106.0 MB/s (1.08x) 106.0 MB/s (1.08x) 114.4 MB/s (1.00x)
test/unicode/word-boundary/ascii-only 86.7 MB/s (1.00x) 79.5 MB/s (1.09x) 76.3 MB/s (1.14x) 86.7 MB/s (1.00x)
test/unicode/word-boundary/unicode-alphabetic 11.2 MB/s (1.00x) 11.2 MB/s (1.01x) 11.2 MB/s (1.00x) 11.1 MB/s (1.01x)
test/unicode/word-boundary/unicode-join-control 15.9 MB/s (1.01x) 15.9 MB/s (1.01x) 16.0 MB/s (1.00x) 15.8 MB/s (1.01x)
test/unicode/word-boundary/unicode-mark 11.2 MB/s (1.00x) 11.2 MB/s (1.01x) 11.2 MB/s (1.00x) 11.2 MB/s (1.00x)
test/unicode/word-boundary/unicode-decimal-number 15.9 MB/s (1.01x) 16.0 MB/s (1.00x) 16.0 MB/s (1.00x) 15.6 MB/s (1.02x)
test/unicode/word-boundary/unicode-connector-punctuation 15.9 MB/s (1.01x) 16.0 MB/s (1.00x) 15.9 MB/s (1.01x) 15.9 MB/s (1.01x)
test/unicode/word/ascii-only 86.7 MB/s (1.00x) 79.5 MB/s (1.09x) 73.4 MB/s (1.18x) 86.7 MB/s (1.00x)
test/unicode/word/unicode-alphabetic 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 70.6 MB/s (1.08x) 76.3 MB/s (1.00x)
test/unicode/word/unicode-join-control 114.4 MB/s (1.00x) 102.2 MB/s (1.12x) 102.2 MB/s (1.12x) 114.4 MB/s (1.00x)
test/unicode/word/unicode-mark 76.3 MB/s (1.00x) 70.6 MB/s (1.08x) 70.6 MB/s (1.08x) 76.3 MB/s (1.00x)
test/unicode/word/unicode-decimal-number 114.4 MB/s (1.00x) 102.2 MB/s (1.12x) 102.2 MB/s (1.12x) 110.0 MB/s (1.04x)
test/unicode/word/unicode-connector-punctuation 114.4 MB/s (1.00x) 102.2 MB/s (1.12x) 102.2 MB/s (1.12x) 114.4 MB/s (1.00x)
unicode/codepoints/any-one 458.3 MB/s (1.00x) 401.1 MB/s (1.14x) 409.8 MB/s (1.12x) 448.9 MB/s (1.02x)
unicode/codepoints/any-all 445.1 MB/s (1.00x) 436.7 MB/s (1.02x) 435.4 MB/s (1.02x) 443.7 MB/s (1.00x)
unicode/codepoints/letters-one 14.4 MB/s (1.00x) 12.3 MB/s (1.18x) 12.3 MB/s (1.18x) 14.5 MB/s (1.00x)
unicode/codepoints/letters-alt 14.7 MB/s (1.00x) 12.3 MB/s (1.19x) 12.5 MB/s (1.18x) 14.6 MB/s (1.00x)
unicode/codepoints/letters-lower-or-upper 910.6 MB/s (1.00x) 910.6 MB/s (1.00x) 908.6 MB/s (1.00x) 908.6 MB/s (1.00x)
unicode/codepoints/contiguous-greek 914.6 MB/s (1.00x) 914.6 MB/s (1.00x) 912.6 MB/s (1.00x) 912.6 MB/s (1.00x)
unicode/compile/one-letter 118.19us (1.07x) 110.24us (1.00x) 110.04us (1.00x) 118.28us (1.07x)
unicode/compile/fifty-letters 5.57ms (1.08x) 5.16ms (1.00x) 5.23ms (1.01x) 5.66ms (1.10x)
unicode/compile/fifty-letters-ascii 9.32us (1.00x) 9.36us (1.00x) 9.32us (1.00x) 9.34us (1.00x)
unicode/compile/match-every-line 42.54us (1.00x) 43.15us (1.01x) 43.09us (1.01x) 42.72us (1.00x)
unicode/compile/match-every-line-ascii 10.23us (1.01x) 10.20us (1.00x) 10.17us (1.00x) 10.31us (1.01x)
unicode/compile/negated-class-matches-codepoint 32.76us (1.00x) 33.20us (1.01x) 33.01us (1.01x) 32.72us (1.00x)
unicode/overlapping-words/ascii 79.7 MB/s (1.00x) 73.3 MB/s (1.09x) 77.1 MB/s (1.04x) 80.0 MB/s (1.00x)
unicode/overlapping-words/english 7.1 MB/s (1.00x) 7.1 MB/s (1.00x) 7.1 MB/s (1.01x) 7.1 MB/s (1.00x)
unicode/overlapping-words/russian 6.4 MB/s (1.01x) 6.5 MB/s (1.00x) 6.5 MB/s (1.01x) 6.5 MB/s (1.01x)
unicode/word/boundary-any-english 123.4 MB/s (1.00x) 122.9 MB/s (1.01x) 122.4 MB/s (1.01x) 123.9 MB/s (1.00x)
unicode/word/boundary-any-russian 26.6 MB/s (1.00x) 17.0 MB/s (1.56x) 16.9 MB/s (1.57x) 26.5 MB/s (1.00x)
unicode/word/boundary-long-english 905.3 MB/s (1.00x) 901.5 MB/s (1.00x) 901.2 MB/s (1.00x) 905.1 MB/s (1.00x)
unicode/word/boundary-long-russian 43.3 MB/s (1.00x) 35.1 MB/s (1.23x) 34.9 MB/s (1.24x) 43.1 MB/s (1.00x)
unicode/word/around-holmes-english 55.8 GB/s (1.00x) 55.7 GB/s (1.00x) 55.7 GB/s (1.00x) 55.7 GB/s (1.00x)
unicode/word/around-holmes-russian 36.4 MB/s (1.01x) 34.8 MB/s (1.06x) 34.7 MB/s (1.06x) 36.7 MB/s (1.00x)
wild/bibleref/compile 289.02us (1.04x) 279.52us (1.01x) 277.77us (1.00x) 289.70us (1.04x)
wild/bibleref/long 883.4 MB/s (1.00x) 883.2 MB/s (1.00x) 882.8 MB/s (1.00x) 883.0 MB/s (1.00x)
wild/bibleref/short 62.5 MB/s (1.00x) 62.8 MB/s (1.00x) 61.9 MB/s (1.02x) 62.6 MB/s (1.00x)
wild/bibleref/line 732.4 MB/s (1.01x) 739.2 MB/s (1.00x) 737.9 MB/s (1.00x) 732.5 MB/s (1.01x)
wild/caddy/caddy 449.5 MB/s (1.00x) 441.7 MB/s (1.02x) 441.7 MB/s (1.02x) 451.5 MB/s (1.00x)
wild/dot-star-capture/rust-src-tools 536.8 MB/s (1.00x) 524.0 MB/s (1.02x) 522.0 MB/s (1.03x) 527.1 MB/s (1.02x)
wild/grapheme/compile 304.73us (1.07x) 285.59us (1.00x) 285.06us (1.00x) 305.07us (1.07x)
wild/grapheme/source-code 122.9 MB/s (1.00x) 120.4 MB/s (1.02x) 117.4 MB/s (1.05x) 122.5 MB/s (1.00x)
wild/grapheme/codepoints 451.4 MB/s (1.00x) 437.2 MB/s (1.03x) 420.5 MB/s (1.07x) 444.2 MB/s (1.02x)
wild/parol-veryl/ascii 10.0 MB/s (1.02x) 10.2 MB/s (1.00x) 10.2 MB/s (1.00x) 10.0 MB/s (1.02x)
wild/parol-veryl/unicode 8.5 MB/s (1.04x) 8.9 MB/s (1.00x) 8.9 MB/s (1.00x) 8.5 MB/s (1.04x)
wild/parol-veryl/multi-patternid-ascii 74.4 MB/s (1.03x) 76.4 MB/s (1.00x) 76.0 MB/s (1.01x) 74.4 MB/s (1.03x)
wild/parol-veryl/multi-captures-ascii 33.9 MB/s (1.05x) 35.6 MB/s (1.00x) 35.5 MB/s (1.00x) 33.9 MB/s (1.05x)
wild/ruff/whitespace-around-keywords 376.6 MB/s (1.01x) 381.9 MB/s (1.00x) 379.7 MB/s (1.01x) 373.6 MB/s (1.02x)
wild/ruff/noqa 1699.1 MB/s (1.02x) 1728.4 MB/s (1.00x) 1714.1 MB/s (1.01x) 1683.4 MB/s (1.03x)
wild/ruff/unnecessary-coding-comment 1520.8 MB/s (1.00x) 1508.9 MB/s (1.01x) 1497.3 MB/s (1.02x) 1500.2 MB/s (1.01x)
wild/ruff/string-quote-prefix 2.1 GB/s (1.03x) 2.2 GB/s (1.00x) 2.2 GB/s (1.01x) 2.1 GB/s (1.03x)
wild/ruff/space-around-operator 497.7 MB/s (1.01x) 503.6 MB/s (1.00x) 497.2 MB/s (1.01x) 493.4 MB/s (1.02x)
wild/ruff/shebang 956.2 MB/s (1.12x) 1070.7 MB/s (1.00x) 1061.2 MB/s (1.01x) 950.6 MB/s (1.13x)
wild/rustsec-cargo-audit/original-unix 29.5 GB/s (1.03x) 30.1 GB/s (1.01x) 29.8 GB/s (1.02x) 30.3 GB/s (1.00x)
wild/rustsec-cargo-audit/original-windows 28.5 GB/s (1.01x) 28.9 GB/s (1.00x) 28.1 GB/s (1.03x) 27.8 GB/s (1.04x)
wild/rustsec-cargo-audit/both-slashes 26.3 GB/s (1.19x) 30.9 GB/s (1.01x) 30.6 GB/s (1.02x) 31.3 GB/s (1.00x)
wild/rustsec-cargo-audit/both-alternate 30.6 GB/s (1.01x) 30.6 GB/s (1.01x) 30.3 GB/s (1.02x) 30.9 GB/s (1.00x)
wild/url/compile 2.80ms (1.00x) 2.83ms (1.01x) 2.85ms (1.02x) 2.80ms (1.00x)
wild/url/search 110.8 MB/s (1.00x) 109.3 MB/s (1.01x) 109.0 MB/s (1.02x) 110.6 MB/s (1.00x)

Results with look-behinds

To get an estimate for performance of "real-world regexes" using look-behinds,
we extracted all regexes that contain look-behind expressions from the snort
ruleset. We chose this as a source of regexes because it has been used as a
benchmark for look-arounds before in Efficient Matching of Regular Expressions with Lookaround Assertions.

Unfortunately, this ruleset is licensed in a way that prohibits us from
distributing it. See the reproduction section below to learn where to get the
ruleset from and how to extract the regexes.

Furthermore, we wrote a couple of very simple benchmarks to demonstrate that
our implementation respects linearity.

We chose to compare our implementation to python/re, as this engine is readily
available, hence easy to benchmark, and used ubiquitously. Note, however, that
python/re only supports bounded length look-behinds, while our implementation
supports unbounded ones as well.

Look-behind benchmark comparison 
$ rebar cmp results_full_combined.csv -f 'lookbehind' -e '[^2]$'
benchmark python/re rust/regex-lookbehind
--------- --------- ---------------------
lookbehind/snort/snort-0 4.1 GB/s (1.00x) 62.2 MB/s (67.07x)
lookbehind/snort/snort-1 356.9 MB/s (1.00x) 43.8 MB/s (8.16x)
lookbehind/snort/snort-2 170.4 MB/s (1.00x) 74.3 MB/s (2.29x)
lookbehind/snort/snort-3 130.3 MB/s (1.02x) 132.6 MB/s (1.00x)
lookbehind/snort/snort-4 3.8 GB/s (1.00x) 64.4 MB/s (60.92x)
lookbehind/snort/linear-haystack-1000 234.8 MB/s (1.00x) 40.4 MB/s (5.82x)
lookbehind/snort/linear-haystack-10000 253.8 MB/s (1.00x) 40.4 MB/s (6.28x)
lookbehind/snort/linear-haystack-100000 257.2 MB/s (1.00x) 40.8 MB/s (6.31x)
lookbehind/snort/linear-haystack-many-1000 33.9 MB/s (1.00x) 15.4 MB/s (2.20x)
lookbehind/snort/linear-haystack-many-10000 32.5 MB/s (1.00x) 15.4 MB/s (2.11x)
lookbehind/snort/linear-haystack-many-100000 32.7 MB/s (1.00x) 15.4 MB/s (2.12x)

A few things to note:

  1. The regexes in snort-0 and snort-4 are the only ones where there is an
    opportunity for prefiltering based on a prefix literal, which we haven't implemented currently.
    This explains the huge difference in speedup compared to all other regexes.
  2. For regexes containing no look-behinds, there are a few benchmarks where the
    speedup ratio between pyhton/re and rust/* is similar to the values seen
    here (e.g. imported/sherlock/everything-greedy-nl, curated/08-words/long-russian).
    We therefore conclude that the baseline performance for regexes with
    look-behinds is reasonable.
  3. The constant throughput in the linear-haystack benchmarks shows that our
    algorithm indeed runs in linear time.

How to reproduce

Please follow these instructions to reproduce our results:

  1. Visit snort.org and create a free account.
  2. Go to Downloads > Rules and download
    snapshot 3200 of the rules in the "Registered" column.
  3. Clone our rebar fork
  4. Extract the contents of the downloaded archive to a new directory called
    snortrules-snapshot-3200 in the root of the cloned repo.
  5. Check the script benchmark_lookbehind.sh for the prerequisites. If you are
    on a debian/ubuntu system, you can install them easily by running
    ./benchmark_lookbehind.sh --install (requires root privileges).
  6. Execute ./benchmark_lookbehind.sh to run the benchmark.
  7. Find the results in the files results_full.csv and
    results_lookbehind.csv, which are placed in the directory containing the
    rebar fork.

Acknowledgements

This was a joint effort by @shilangyu and @Multimodcrafter, supervised by Aurèle Barrière and Clément Pit-Claudel at EPFL's SYSTEMF.

Hywan reacted with hooray emoji septatrix, null-dev, lawrenceadams, lnicola, Aurele-Barriere, boazy, Mylloon, RGBCube, CertainLach, bluss, and 17 more reacted with heart emoji
Copy link

An optimization to the bounded-length look-behinds yields large speed-ups on benchmarks used for evaluation (up to 150 times faster):

benchmark python/re rust/regex-lookbehind rust/regex-lookbehind-new
--------- --------- --------------------- -------------------------
lookbehind/snort/snort-0 2.2 GB/s (1.00x) 45.0 MB/s (50.40x) 1034.7 MB/s (2.19x)
lookbehind/snort/snort-1 204.0 MB/s (1.00x) 34.3 MB/s (5.94x) 34.1 MB/s (5.99x)
lookbehind/snort/snort-2 107.1 MB/s (71.24x) 53.0 MB/s (143.94x) 7.5 GB/s (1.00x)
lookbehind/snort/snort-3 100.7 MB/s (80.25x) 102.2 MB/s (79.08x) 7.9 GB/s (1.00x)
lookbehind/snort/snort-4 2041.9 MB/s (1.00x) 45.9 MB/s (44.52x) 967.3 MB/s (2.11x)

rust/regex-lookbehind-new is this PR together with the applied optimization which has been implemented here: epfl-systemf#11. We would be happy to discuss and merge that too, once the original PR gains interest.
Aurele-Barriere reacted with thumbs up emoji

Copy link
Author

We have now published a write-up of our development process here. This might help in understanding some of our design choices better.

Furthermore, it also describes some additional work on the bounded backtracker to support the same features there and the challenges that go along with it. These changes are available on the backtracking branch of our fork.
BurntSushi, CertainLach, slanterns, signalwerk, messense, joseph-gio, dlight, Aurele-Barriere, radekmie, NoahTheDuke, and 11 more reacted with heart emoji

Copy link

hugopendlebury commented Jul 29, 2025 •
edited
Loading

Sorry maybe I'm being naive... But this PR sounds like a very useful addition, is there anything that could be done to encourage this to be accepted ?

Copy link

Sorry maybe I'm being naive... But this PR sounds like a very useful addition, is there anything that could be done to encourage this to be accepted ?

#1273 (comment)

This comment was marked as duplicate.

Multimodcrafter and others added 23 commits August 25, 2025 10:33 
This is the first step to supporting captureless lookbehind assertions
The lack of recursing into the inner expression of a lookaround is correct under the current assumption that lookarounds cannot have capture groups. But once the restriction is lifted, this wrong implementation can be very subtle to find. Instead, we can already do the filtering and accept it being a no-op for now.
This makes it consistent with parser's ErrorKind::UnsupportedLookAround.
shilangyu and others added 25 commits August 25, 2025 10:36
Copy link
Member

Thanks so much for submitting this!

I think it would be a good idea to experiment with this on master before making it available in regex proper. (Where once it's available, we're committed to it indefinitely.) But if I'm being a honest, a PR this big is probably just never going to get merged. I would likely need days of focused effort to dig into this. It's probably just not going to happen any time soon. 68 commits in a single PR is not really acceptable.

Instead, I think a more plausible route here is the following:

  • A change that adds opt-in support for parsing look-behind in regex-syntax. (This may have "mundane" changes to regex-automata to reject HIR that have look-behind in them.)
  • A change that adds support for matching look-behinds in regex-automata.
  • A change that makes regex opt into look-behind support. (If the previous two steps are done right, this part should be trivial and can probably just be dropped.)

Ideally it would be broken up into more PRs than the above, but I don't know where the best logical breakpoints are besides what I listed above. And ideally, the commits wouldn't be "here is my development history," but rather, "here is a logical series of patches that can each be reviewed in relative isolation."

I also want to be clear that I have not yet decided on whether this should be available in regex proper yet. But I think these are the first steps toward making that decision.
Chaoses-Ib and keith-hall reacted with thumbs up emoji

Copy link

Thank you for your feedback and for being open to experiment before commitment. We agree such a large PR is hard to review and reason about. We will come up with a more detailed plan for smaller, more focused PRs. Once the plan is done, we will be happy to receive feedback on it before we get to work.
Aurele-Barriere, BurntSushi, Multimodcrafter, andylizi, NoahTheDuke, Nytelife26, and brancengregory reacted with thumbs up emoji

Copy link

Hi @BurntSushi! Here is what we think is a feasible sequence of self-contained PRs:

  1. Parsing, AST, and HIR support of look-behinds hidden behind a cargo feature in regex-syntax.

    Notably, this feature flag would not be enabled in any other crate, removing the need for stub unimplemented everywhere.

  2. Compilation of regexes with look-behinds into NFA hidden behind an analogous cargo feature in regex-automata.

    All engines would reject regexes with look-behinds.

  3. Look-behind support in the PikeVM.

    Starting with this PR, look-behinds can be used to match strings, allowing users to start experimenting with look-behinds.

  4. Optimize match-all to be linear in the presence of look-behinds.

    This is an important optimization (caching of state related to look-behinds) but with a non-trivial implementation and therefore warrants a separate PR.

  5. Look-behind support in the meta engine.

  6. Enable look-behind support in regex.

    This can be either done by enabling the feature flag in regex-syntax and regex-automata, or by removing the feature flags altogether.

  7. (bonus) Bounded look-behind optimization.

    Mentioned here.

  8. (bonus) Look-behind support in the bounded backtracker.

    Mentioned here.

Each PR preserves both correctness and the guarantees of every crate.

Please let us know what you think about this split. If it sounds good, we will be happy to start working on them right away!
null-dev, lmmx, brancengregory, vqhdev, and aapoalas reacted with heart emoji

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@bartekpacia bartekpacia bartekpacia approved these changes

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

AltStyle によって変換されたページ (->オリジナル) /