Fix/cache fail silent

[ygottschalk]

see discussion in #9925

[ygottschalk]

#9925 (comment) will not be catched by this. For fixing that, we would need to check the unserialized array to not contain __PHP_Incomplete_Class for unserialize (and whatever the equivalent for igbinary_unserialize might be)

Should that be done in this PR? 👍 👎

[kkmuffme]

Good:

1. Errors when failing to write to cache should be ignored too though?

2. use grep 'throw .*cache' on psalm's codebase to find additional places that you missed

[ygottschalk]

1. Errors when failing to write to cache should be ignored too though?

Throw if a serious error happened (eg psalm is explicitly configured to use cache but cache dir is not writeable)

I don't think so. Either it's disposable or not; otherwise we will end up in a discussion on what is "serious" and what is not? e.g. is a E_WARNING serious? (when we (un)serialize/compress)

...?!

2. use grep 'throw .*cache' on psalm's codebase to find additional places that you missed

only thing I could identify beeing relevant is \Psalm\Internal\Provider\ClassLikeStorageProvider::exhume which considers things to be in cache but I could not confirm that there is a check confirming that....

[dkarlovi]

IMO we can add a few more cases if they're obvious, but shouldn't block this PR until we reach 100% completeness, Psalm's cache handling was a total mess, repeated in 10 different places before #9889 and nobody cared about it then, now that we've significantly moved the needle, we've all become code gourmands, nothing but the very best is even worth considering. :)

Let's be pragmatic and see this merged, the further inconsistencies (if any) can easily be fixed as discovered, and the found inconsistencies will actually be places I've missed in #9889 so they'll work exactly like the previous release, not improving, but not breaking either.

[ygottschalk]

This should get fixed (and in here is a possible solution: https://wiki.php.net/rfc/improve_unserialize_error_handling), but that can happen in a different PR.

I would really like to have some form of output for the write error or keep throwing....

[ro0NL]

#9925 (comment)

note it's no blocker for us, just the first example i encountered, and it's probably an edge case

i agree with @dkarlovi, if we cant solve this generically, then baby steps it is

[orklah]

Let's be pragmatic and see this merged

agreed :)

Thanks!