max()/min() should expect non-empty-array

[staabm]

fix was easier then expected ;)

closes phpstan/phpstan#7239

[staabm]

the error in

https://github.com/sebastianbergmann/phpunit/blob/93d4bf4c37aec6384bb9e5d390d9049a463a7256/src/TextUI/DefaultResultPrinter.php#LL541C30-L541C30

seems valid

---

the error in

https://github.com/slevomat/coding-standard/blob/8c49f8e6e4e3661ee6e568a06676146475f73430/SlevomatCodingStandard/Sniffs/ControlStructures/AssignmentInConditionSniff.php#L77

seems valid

---

the error in

https://github.com/doctrine/orm/blob/35c44a56677adb3ce796138b6e4934ce93ec6811/lib/Doctrine/ORM/Tools/EntityGenerator.php#L790

seems valid

these error cases should check for non-empty-ness or should otherwise make sure the typen given to max/min is not empty.
other errors look unrelated

[phpstan-bot]

This pull request has been marked as ready for review.

[ondrejmirtes]

These failures are the reason why I don't like narrowing the parameter types of built-in functions...

I don't see it being justified enough to force user to write non-empty-array PHPDoc in their own code after patch PHPStan version upgrade...

[staabm]

I agree with this sentiment in general.

in this case though the error we protect the user from is a fatal error (since php8+):
https://3v4l.org/OE4CD
https://3v4l.org/vhUXS

[ondrejmirtes]

Yeah but in existing code, it's very likely these arrays are always empty, although not annotated like that...

[staabm]

not sure I get your point: do you mean there are a lot of situations in which the array will be empty but we don't know about it?

[hemberger]

Yeah but in existing code, it's very likely these arrays are always empty, although not annotated like that...

I think maybe "are always non-empty" is what was meant?

In the codebases that I've been analyzing recently, there are definitely edge cases where min/max will be passed an empty array unintentionally. No warning is emitted by PHPStan currently (as of 1.9.13), though I was actually more aware of these cases prior to the fix to the PHP 8+ min/max return types in #2161, since I would get an (erroneous) warning about using its result, e.g. int|false when int was expected.

It always seemed to me that PHPStan was designed to not let errors like these raise naturally, and so enforcing non-empty-array seems technically correct (and genuinely useful in some cases). But it is likely to be a pedantic warning in many cases, and it is an arduous task to assess the "emptiness" of every array. Does it perhaps warrant a stricter analysis parameter such as checkEmptyArrayPassedToMinMax?

Either way, I fully trust you two to settle on the solution that is best for the PHP community. :)

[staabm]

I think one way could be to enable the feature only in bleeding edge, or alternatively only in phpstan-strict-rules?

[VincentLanglet]

the error in

sebastianbergmann/phpunit@93d4bf4/src/TextUI/DefaultResultPrinter.php#LL541C30-L541C30

seems valid

I might be wrong but itt's not really valid for

$lines = preg_split('/\r\n|\r|\n/', $buffer);
$padding = max(array_map('\strlen', $lines));

I would expect preg_split to returns a non-empty-array (like the explode), then array_map doesn't touch the number of arguments.

Currently it's considered as an array (https://phpstan.org/r/284ce5f5-f222-4518-8144-a90ead09b60c)
Do you think preg_split return type can be changed to non-empty-array|false ?

[VincentLanglet]

I think one way could be to enable the feature only in bleeding edge, or alternatively only in phpstan-strict-rules?

To me, bleeding edge can be considered as "I'm sure we will release this feature in 2.0, but we want to avoid BC break".
Here, it's not sure we want this change because it could add to many noise in some codebase.

What about an experimental flag ?
experimental: false allow to disable some debatable features, and it still allow to release this to get opinion on such change ?

[ondrejmirtes]

Did bleedingEdge support here: 06b746d

Merged this PR as: 35b4cac

Thank you.

[staabm]

thanks