Fix escaped hash in regex extended modefixed escaped hash in regex extended mode by dg · Pull Request #4814 · phpstan/phpstan-src

dg · 2026-01-24T07:22:03Z

When using the x (extended) modifier, \# was incorrectly treated as the start of a comment. According to PCRE specification, \# is an escaped literal hash character and should not start a comment.

Example

preg_match('/^ ([\#.]) $/x', $s, $matches);

Before this fix, PHPStan would strip \#.]) $/x as a comment, breaking the pattern parsing.

Changes

src/Type/Regex/RegexGroupParser.php: Changed the comment-stripping regex from /(?<!\?)#.*/ to /(?<!\\)#.*/ to respect escaped hashes
Added test case testExtendedSyntaxEscapedHash in tests/PHPStan/Analyser/nsrt/preg_match_shapes.php

Note

This fix does not address # inside character classes like [#.] (which is also not a comment in PCRE). However, users can use the workaround [\#.] which now works correctly with this fix.

Copilot

Pull request overview

Fixes handling of escaped hash (\#) when stripping comments in PCRE extended (x) mode so that \# is treated as a literal # rather than the start of a comment.

Changes:

Update the freespacing comment-stripping logic in RegexGroupParser to account for escaped hashes.
Add an nsrt test case covering escaped hash behavior in extended mode.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
`src/Type/Regex/RegexGroupParser.php`	Adjusts the regex used to strip `#...` comments in `/x` mode.
`tests/PHPStan/Analyser/nsrt/preg_match_shapes.php`	Adds a regression test to validate escaped-hash handling in extended mode parsing.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-24T07:26:07Z

src/Type/Regex/RegexGroupParser.php

 				// in freespacing mode the # character starts a comment and runs until the end of the line
-				$regex = preg_replace('/(?<!\?)#.*/', '', $regex) ?? '';
+				// but \# is an escaped literal hash, not a comment
+				$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';


The new freespacing comment-stripping pattern will also treat the # in PCRE inline comment groups (?# ... ) as a line comment (because it’s not preceded by \). This is a regression from the previous (?<!\?) guard and can break valid patterns (e.g. tests/PHPStan/Analyser/nsrt/bug-11311.php contains (?# ...) inside an /x regex). Consider updating the stripping logic to exclude # that is part of (?#...) (and ideally avoid using a single lookbehind heuristic altogether).

Suggested change

$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';

// also, (?# ... ) is an inline comment group and its # must not be treated as a line comment

$regex = preg_replace('/(?<!\\\)(?<!\(\?)#.*/', '', $regex) ?? '';

Copilot · 2026-01-24T07:26:07Z

src/Type/Regex/RegexGroupParser.php

 				// in freespacing mode the # character starts a comment and runs until the end of the line
-				$regex = preg_replace('/(?<!\?)#.*/', '', $regex) ?? '';
+				// but \# is an escaped literal hash, not a comment
+				$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';


/(?<!\\\)#.*/ only checks the single character immediately before #. In extended mode, \\# (two backslashes before #) should still start a comment because the last backslash is escaped, but this regex will incorrectly preserve it. To match PCRE semantics you need to treat # as escaped only when it’s preceded by an odd-length run of backslashes (and also ensure you’re not inside a character class).

Suggested change

$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';

// and # inside a character class does not start a comment

$length = strlen($regex);

$processedRegex = '';

$inCharClass = false;

$backslashRun = 0;

for ($i = 0; $i < $length; $i++) {

$char = $regex[$i];

$escaped = ($backslashRun % 2) === 1;

// update character class state (unescaped [ and ])

if ($char === '[' && !$escaped) {

$inCharClass = true;

} elseif ($char === ']' && !$escaped) {

$inCharClass = false;

}

// start of a comment: unescaped # outside of character class

if ($char === '#' && !$escaped && !$inCharClass) {

// skip until end of line or end of pattern

while ($i + 1 < $length) {

$nextChar = $regex[$i + 1];

if ($nextChar === "\n" || $nextChar === "\r") {

break;

}

$i++;

}

$backslashRun = 0;

continue;

}

$processedRegex .= $char;

if ($char === '\\') {

$backslashRun++;

} else {

$backslashRun = 0;

}

}

$regex = $processedRegex;

Copilot · 2026-01-24T07:26:07Z

src/Type/Regex/RegexGroupParser.php

 				// in freespacing mode the # character starts a comment and runs until the end of the line
-				$regex = preg_replace('/(?<!\?)#.*/', '', $regex) ?? '';
+				// but \# is an escaped literal hash, not a comment
+				$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';


The literal /(?<!\\\)#.*/ relies on PHP single-quote escaping rules (the trailing \ is not an escape sequence), which is hard to read/maintain. Consider rewriting it in a clearer form (e.g. using \\\\ in the PHP string for a \\ regex escape) or adding a short comment explaining the escaping, to reduce the chance of accidental changes breaking the pattern.

Suggested change

$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';

// (?<!\\\\) is a negative lookbehind for a backslash; '\\\\' in PHP becomes '\\' in the regex

$regex = preg_replace('/(?<!\\\\)#.*/', '', $regex) ?? '';

\# was incorrectly treated as comment start in /x mode

ondrejmirtes · 2026-01-24T08:01:26Z

/cc @staabm Please review 😊

staabm

I was not aware of such a regex feature, but implementation wise it looks good to me.

thank you

ondrejmirtes · 2026-01-25T14:58:00Z

Thank you!

Copilot AI review requested due to automatic review settings January 24, 2026 07:22

Copilot started reviewing on behalf of dg January 24, 2026 07:22 View session

Copilot AI reviewed Jan 24, 2026

View reviewed changes

fixed escaped hash in regex extended mode

1252f55

\# was incorrectly treated as comment start in /x mode

dg force-pushed the pull-regex-ex branch from 696bc81 to 1252f55 Compare January 24, 2026 08:01

staabm approved these changes Jan 24, 2026

View reviewed changes

ondrejmirtes merged commit f1298b5 into phpstan:2.1.x Jan 25, 2026
631 of 639 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix escaped hash in regex extended modefixed escaped hash in regex extended mode#4814

Fix escaped hash in regex extended modefixed escaped hash in regex extended mode#4814
ondrejmirtes merged 1 commit intophpstan:2.1.xfrom
dg:pull-regex-ex

dg commented Jan 24, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 24, 2026

Uh oh!

Copilot AI Jan 24, 2026

Uh oh!

Copilot AI Jan 24, 2026

Uh oh!

ondrejmirtes commented Jan 24, 2026

Uh oh!

staabm left a comment

Uh oh!

Uh oh!

ondrejmirtes commented Jan 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

	$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';
	// also, (?# ... ) is an inline comment group and its # must not be treated as a line comment
	$regex = preg_replace('/(?<!\\\)(?<!\(\?)#.*/', '', $regex) ?? '';

-				$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';
+				// and # inside a character class does not start a comment
+				$length = strlen($regex);
+				$processedRegex = '';
+				$inCharClass = false;
+				$backslashRun = 0;
+				for ($i = 0; $i < $length; $i++) {
+					$char = $regex[$i];
+					$escaped = ($backslashRun % 2) === 1;
+					// update character class state (unescaped [ and ])
+					if ($char === '[' && !$escaped) {
+						$inCharClass = true;
+					} elseif ($char === ']' && !$escaped) {
+						$inCharClass = false;
+					}
+					// start of a comment: unescaped # outside of character class
+					if ($char === '#' && !$escaped && !$inCharClass) {
+						// skip until end of line or end of pattern
+						while ($i + 1 < $length) {
+							$nextChar = $regex[$i + 1];
+							if ($nextChar === "\n" || $nextChar === "\r") {
+								break;
+							}
+							$i++;
+						}
+						$backslashRun = 0;
+						continue;
+					}
+					$processedRegex .= $char;
+					if ($char === '\\') {
+						$backslashRun++;
+					} else {
+						$backslashRun = 0;
+					}
+				}
+				$regex = $processedRegex;

	$regex = preg_replace('/(?<!\\\)#.*/', '', $regex) ?? '';
	// (?<!\\\\) is a negative lookbehind for a backslash; '\\\\' in PHP becomes '\\' in the regex
	$regex = preg_replace('/(?<!\\\\)#.*/', '', $regex) ?? '';

Conversation

dg commented Jan 24, 2026

Example

Changes

Note

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 24, 2026

Choose a reason for hiding this comment

Uh oh!

ondrejmirtes commented Jan 24, 2026

Uh oh!

staabm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ondrejmirtes commented Jan 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants