[yoeunes]

This is a research branch. It runs PHPStan's regex shape inference on PHPRegex, a PCRE2 pattern parser that runs on PHP 7.4, to see how the library holds up against PHPStan's tests.

What the branch does

• RegexGroupParser reads the PHPRegex AST, and the type of each group comes from the library's capture shape analysis.
• IgnoredRegexValidator reads ignoreErrors patterns with the same parser.
• Patterns are read for the analysed PHP version and the PCRE2 version it bundles. A pcreVersion parameter, or config.platform.lib-pcre in composer.json, sets that version.
• With bleeding edge, RegularExpressionPatternRule also reports a pattern the running PHP compiles but the analysed PHP does not.
• The preg_replace_callback() callback is checked against the type its flags give, each preg_replace_callback_array() callback gets the shape of its own pattern, and preg_replace() reads the group references of its replacement.

How it is checked

RegexArrayShapeMatcherEngineTest takes every constant pattern of the preg_* fixtures, matches it with the engine on generated subjects and with every flag combination, and checks that the inferred type accepts what the engine writes. All 327 patterns pass.

The full test suite passes, and make phpstan reports no errors.

poc/ measures the branch and upstream 2.3.x on four pattern sets (the PCRE2 and php-src test suites, patterns from open-source projects, generated edge cases), with a witness for each contradicted pattern and the scripts to replay them: see poc/README.md.

Issues this branch answers

• Support non-auto-capturing modifier in preg_match() inference phpstan/phpstan#11356
• Possibly empty regex matches are infered as non-empty/non-falsy phpstan/phpstan#12840
• Generated baseline can't be used if regex can't be compilated to UTF-8 chars phpstan/phpstan#12629
• A few unparsed regex usecases phpstan/phpstan#11746
• Incorrect lower bound -1 for offset in mandatory preg_match() groups when 0 (non-negative-int) is expected phpstan/phpstan#14959
• Misinterpreting Whether Regular Expression Piece Exists phpstan/phpstan#11602
• preg_replace_callback doesn't account for flags in callback signature phpstan/phpstan#10396
• *ParameterClosureTypeExtension for functions which accept a array of callables phpstan/phpstan#11438
• wrong preg_replace() type inference phpstan/phpstan#12997
• Support tagged union in preg_match phpstan/phpstan#11443 (partly)
• Successful preg_match() with pattern /^[a-z]+$/ fails to narrow string type to non-empty-lowercase-string phpstan/phpstan#15018
• preg_match: infer digit capture groups (\d+) as decimal-int-string instead of numeric-string, for + int arithmetic phpstan/phpstan#14909 (where the pattern rules out a leading zero)

[staabm]

ondrej did something in phpstan#6851

[yoeunes]

ondrej did something in phpstan#6851

Hello @staabm, thanks for the heads-up. Glad to see Hoa gone from PHPStan. I'll probably rebase this branch on the new parser and keep exploring 🙏🏻.

[ondrejmirtes]

The fixes on the PR were actually inspired by test cases in PCRE and PHP.