Report unmatched baseline ignore rules when the origin trait file was analysed

[sylfabre]

Problem

When reportUnmatchedIgnoredErrors: true and PHPStan runs in onlyFiles mode (incremental analysis), unmatched ignore rules with a specific path are silently suppressed — even when PHPStan has enough information to know the rule is genuinely unmatched.

Background

phpstan/phpstan#3099 reported false positives where unmatched path-specific ignore rules were reported even though the relevant file was never part of the analysis. This was fixed in c9df2f4 by skipping unmatched rules when the path was not in the analysed file set, and by unconditionally suppressing all path-specific unmatched rules in onlyFiles mode.

The unconditional suppression prevents a real false positive with trait errors: a trait error is attributed to the class file (Foo.php) but only surfaces when the trait file (FooTrait.php) is part of the analysis. In partial analysis without FooTrait.php, the baseline rule goes unmatched for a legitimate reason and must be suppressed.

However, this blanket suppression also means that when FooTrait.php was analysed and the rule still didn't match, the unmatched rule is silently swallowed — even though PHPStan has full information to report it correctly.

Solution

Introduce an origin field in baseline entries for errors that originate from a trait file (Error::getTraitFilePath()). This field records which file must be present in the analysis for the error to appear.

In IgnoredErrorHelperResult::process(), when an unmatched rule has a realOrigin:

• If the origin file was not analysed → suppress (no false positive, behaviour preserved from c9df2f4)
• If the origin file was analysed → report as unmatched (PHPStan has full information)

For rules without origin (user-written rules, non-trait baseline entries), the existing onlyFiles suppression is preserved unchanged.

Changes

• BaselineNeonErrorFormatter / BaselinePhpErrorFormatter: emit origin for errors with a trait file path, grouped by (message, origin) so distinct trait sources produce separate entries
• IgnoredErrorHelper: include origin in the deduplication key; normalise origin → realOrigin alongside path → realPath
• IgnoredErrorHelperResult: check realOrigin against analysedFilesKeys before deciding whether to suppress an unmatched rule in onlyFiles mode
• AnalyserTest: two new test cases — without origin (suppressed, existing behaviour preserved) and with origin where both files are analysed (reported, new behaviour)

[phpstan-bot]

You've opened the pull request against the latest branch 2.2.x. PHPStan 2.2 is not going to be released for months. If your code is relevant on 2.1.x and you want it to be released sooner, please rebase your pull request and change its target to 2.1.x.

[ondrejmirtes]

This does not seem right.

even when PHPStan has enough information to know the rule is genuinely unmatched

It doesn't have enough information. Let's say you're ignoring something in FooTrait.php. You're running "partial" analysis, onlyFiles=true. There's an unmatched ignored error in FooTrait.php. But it only happens when you're analysing Bar.php (because the class uses that trait) which you're skipping here.

My rule of thumb: do not ever analyse only specific files when you can and should analyse the whole project: https://phpstan.org/blog/why-you-should-always-analyse-whole-project