Fix !isset() with ArrayDimFetch

[staabm]

the current 1.10.x handles the !isset() context for variables only.
it works only correct for !isset($var) and its implications on the variable.

with this PR we add support for !isset($var['abc']) which means

• we handle implications on $var's array-shape based on the $var['abc']-offset type (based on nullability and optional/non-optional offset state)

with this PR we know when

• offsets can only be null or non-null in true/falsey-context
• offsets may or will not exist in falsey-context
• on array-shapes with just a single offset we know about the certainty implications on the var in falsey-context

the newly added tests in tests/PHPStan/Analyser/data/falsy-isset.php best reflect the new type inference implications.

closes phpstan/phpstan#9908
closes phpstan/phpstan#9426
closes phpstan/phpstan#8724
closes phpstan/phpstan#5128

[staabm]

the remaining error is strange

Offset 'input' does not exist on array{input: Bug8724b\Input}|array{label: string}.

I have not yet an idea about it... but maybe we should even expect this error?

type-wise the offset label always exists, which means the null-coalescing is unnecessary and the code right to ?? is actually dead code..?

/** * @param array{label:string}|array{input:Input} $data */ function test(array $data): void {	$label = $data['label'] ?? $data['input']->getLabel(); } 

[phpstan-bot]

This pull request has been marked as ready for review.

[clxmstaab]

ohh I get it now.. I think the PR fixes the initial reported snippet but not the one posted within a followup comment

[ondrejmirtes]

Can you please explain in your own words what are the faults of the current handling (as of 1.10.x) and what approach this PR does to fix that?

[staabm]

I added a description to the PR

[ondrejmirtes]

Alright, what about nested expressions like $foo['abc']['def']? Does it work correctly?

[mvorisek]

I would expect the type to be simplified to array{bar: 1|null}