Treat mb_str_pad() as side-effect-free like other mbstring string functions

[zonuexe]

Among the mbstring string-processing functions in the issue's reproducer, phpstorm-stubs marks mb_str_pad() as #[Pure(true)], while comparable functions such as mb_strlen(), mb_strcut(), and mb_trim() are marked #[Pure]. The metadata generator therefore emits hasSideEffects => true for mb_str_pad only, so @phpstan-pure functions cannot call it.

This change treats mb_str_pad as hasSideEffects => false so it matches those siblings.

This is a pragmatic override, not a claim that the function is theoretically pure. The result can change if the analysed program changes internal encoding (mb_internal_encoding() or default_charset). PHPStan does not track encoding-state changes made by the analysed program. Some existing return-type extensions (mb_strlen() and mb_str_split()) use the analyser process's mb_internal_encoding() when the encoding argument is omitted; mb_str_pad() itself has no dedicated return-type extension. Encoding-dependent purity (explicit encoding = pure, omitted = impure) remains out of scope.

This is not the same rationale as the ob_get_level() exception in #6439: PHPStan models output-buffer level changes, but it does not model the encoding state that affects mb_str_pad(). This override is an explicit consistency trade-off, not a scope-managed-state optimization.

functionMetadata_original.php records this project-specific policy, while the generator special case prevents phpstorm-stubs' current #[Pure(true)] annotation from rejecting or reversing it during regeneration.

Applying functionMetadata on BetterReflection's fallback path is not done here: name-keyed built-in metadata can leak to user polyfills and unrelated built-ins. This PR does not change the pre-existing cross-version discovery of PHP 8.4-only mbstring names under phpVersion: 8.3.

Addresses Issue 1 in phpstan/phpstan#15224. Issue 2 (encoding-sensitive purity) remains open and is not addressed by this PR.

Made with Cursor

[phpstan-bot]

This pull request has been marked as ready for review.

[staabm]

I sent a comment into phpstorm stubs, as the inconsistency for mb_str_pad vs. all other functions was created on that end, and it does not really make sense IMO.

see JetBrains/phpstorm-stubs#1836