False positive: Offset ... on array... on left side of ?? always exists and is not nullable.

[schlndh]

Bug report

Phpstan gets confused when an associative array is created in a loop and subsequently the array is foreach-ed over. It thinks that all the offsets that may have been assigned are assigned and it complains about ?? null. It's one of the new errors that I saw after upgrading to 1.0, so it's probably a regression (unless the message changed and wasn't caught by the baseline anymore).

Code snippet that reproduces the problem

https://phpstan.org/r/1420808b-7d48-41c2-936e-8e17b0bdae94

Expected output

It shouldn't complain about ?? null.

[chaos0815]

Not sure if same usecase, but same error:
see https://phpstan.org/r/be6a2d6e-bf05-41be-a803-cc92d9e34c05

[ondrejmirtes]

@schlndh PHPStan is smart and I think that it knows what's going on in your example: https://phpstan.org/r/1b680ac5-e330-47e7-bc0e-132f83f31588 - the 'a' index always exists.

@chaos0815 unrelated, please open a separate issue.

[schlndh]

@ondrejmirtes Thanks, but I don't understand your explanation. Clearly, it can happen that $arr[$i]['a'] is not set even if $arr[$i] is set (which I now realize is the difference between the first and second ?? null). Therefore, it is valid to write $val['a'] ?? null to acknowledge that fact (and mute the undefined index warning from PHP). So PHPStan shouldn't complain about it.

Of course, I understand that an array structure can be arbitrarily complex and PHPStan can never understand everything. In this case, I'd say that non-empty-array<int<0, 99>, array{a?: int<0, 99>, b?: int<0, 99>}> or non-empty-array<int<0, 99>, array{a: int<0, 99>}|array{b: int<0, 99>}> would be better (I assume that either of them would prevent PHPStan from complaining about the ?? null, but I haven't checked). But I'm not sure whether it can be inferred from the code in a reasonable way.

Ultimately, it's probably best not to write code like this in the first place and use proper types instead. 😅

[ondrejmirtes]

I'm sorry, I misread the example, I thought that in both if/else 'a' is set and not 'b'. So yeah, this is a bug :)

[phpstan-bot]

@schlndh After the latest commit in dev-master, PHPStan now reports different result with your code snippet:

@@ @@
-18: Offset 'a' on array{a: int<min, 99>, b: int<min, 99>} on left side of ?? always exists and is not nullable.
+18: Offset 'a' on array{a: int<0, 99>, b: int<0, 99>} on left side of ?? always exists and is not nullable.

Full report

Line	Error
18	Offset 'a' on array{a: int<0, 99>, b: int<0, 99>} on left side of ?? always exists and is not nullable.

[phpstan-bot]

@schlndh After the latest commit in 1.6.x, PHPStan now reports different result with your code snippet:

@@ @@
-18: Offset 'a' on array{a: int<min, 99>, b: int<min, 99>} on left side of ?? always exists and is not nullable.
+No errors

[phpstan-bot]

@ondrejmirtes After the latest commit in 1.6.x, PHPStan now reports different result with your code snippet:

@@ @@
-17: Dumped type: non-empty-array<int<0, 99>, array{a: int<0, 99>, b: int<0, 99>}>
-20: Offset 'a' on array{a: int<0, 99>, b: int<0, 99>} on left side of ?? always exists and is not nullable.
+17: Dumped type: non-empty-array<int<0, 99>, array{a?: int<0, 99>, b?: int<0, 99>}&non-empty-array>

Full report

Line	Error
17	Dumped type: non-empty-array<int<0, 99>, array{a?: int<0, 99>, b?: int<0, 99>}&non-empty-array>

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.