Some problems are not reported when `showTypeOnHover` is enabled

[lkrms]

I have a problem reported by PHPStan only when I run it on the command line--or when I disable showTypeOnHover.

It seems the extension's TreeFetcher somehow changes PHPStan's type inferences in some circumstances--perhaps only when bleeding edge is enabled? (I'm seeing the behaviour after moving to PHPStan 1.12.10 and enabling bleeding edge; it's not clear which was the trigger.)

The difference I'm observing is that PHPStan thinks a string is non-empty normally, but with TreeFetcher in the mix, the non-empty flag is removed. Ironically, the latter behaviour is correct, but that's beside the point ;)

[SanderRonde]

Well that's really strange. I can't say that I have a clue what's causing this, maybe a PHPStan-internal bug? Will keep this issue open for a while to see if anybody else is experiencing the same issue but chances are I won't figure it out. Do you maybe have an example repo for it happening? That would help a bunch :)

[lkrms]

If you check out this tag: https://github.com/salient-labs/toolkit/releases/tag/v0.99.62

And run composer install, you should get No error with identifier identical.alwaysFalse is reported on line 209 for src/Toolkit/Http/FormData.php, which you won't get when running phpstan directly.

Hopefully it doesn't make a liar of me 😜

Edit: just confirmed it happens on both platforms I use (macOS and Linux). Phew.

[SanderRonde]

First of all thanks for the repo, I never would have been able to find this without it. I've debugged it a bunch and it is indeed a PHPStan bug. I've opened an issue: https://github.com/phpstan/phpstan/issues/12065

I'll close this issue for now since it's not directly related to the extension but will keep you up to date on the status of the other issue.

[SanderRonde]

Bit of an update: they found the bug and a fix is underway in a new version of PHPStan (so it'll probably be a while before you see it).