#6840 Report useless array filter

[leongersen]

New branch based on 1.5.x, I could not reopen #1076.

Implements phpstan/phpstan#6840.

[ondrejmirtes]

I'm looking at this - it's really nice, Some observations that would make this even better:

1. Calling array_filter without callback on always-truthy values is similarly useless as on always-falsy values. Please detect that too.
2. We should also detect situations with array_filter($array, null, ARRAY_FILTER_USE_KEY) and array_filter($array, null, ARRAY_FILTER_USE_BOTH) - not sure how the latter one behaves, needs to be tested.

[leongersen]

1. Calling array_filter without callback on always-truthy values

That's what this PR checks for. Do you mean the opposite? Calling array_filter on only falsy values will always result in [] and is likely an error, but not without effect.

2. Sounds good, I'll add support for that.

[ondrejmirtes]

Yes, that's what I meant, the rule should report both always-truthy and always-falsy arrays :)

[herndlm]

USE_BOTH without a callback only uses the first arg (= value) AFAIK, see also https://github.com/phpstan/phpstan-src/blob/1.4.10/tests/PHPStan/Analyser/data/array-filter.php#L36

But better double-check :) if the referenced test is wrong, I need to fix smth in the ArrayFilter extension

[leongersen]

I checked, passing null explicitly as the callback (with or without the mode parameter) results in the following warning:

Warning: array_filter() expects parameter 2 to be a valid callback, no array or string given

[herndlm]

Ah maybe that explains why the key mode test above my rererence also looks weird 😅 would be interesting if and how it filters

[leongersen]

It doesn't filter at all:

var_dump(@array_filter([0,1], null)); => NULL

I've added support for reporting arrays with only falsey values.

[leongersen]

The failing test is unrelated.

[ondrejmirtes]

The failure is definitely related as the file contains this: array_filter([]);

Feel free to modify the file to contains something that's valid, like [0, 1, 2] :)

[leongersen]

Turns out file-without-errors is not without errors 🙈

[ondrejmirtes]

I pushed some extra commits to adjust this to my taste.

[ondrejmirtes]

Thank you.

[ondrejmirtes]

Also to keep the BC promise (https://phpstan.org/user-guide/backward-compatibility-promise) I had to do this: 7627b52

The library of performed rules on analysed code, and the meaning of rule levels does not change in minor/patch versions. This means that PHPStan will not look for new categories of errors in a minor/patch version. New rules are added only in major versions, and also as part of bleeding edge which is an opt-in preview of the next major version.

[herndlm]

@leongersen fyi I think null callables would work fine, they only work since PHP 8 and weirdly only ever filter by value (= ignore any given mode)

See https://3v4l.org/Cj17A