Type narrowed to non-empty-string is not retained by template

[bendavies]

Bug report

Hi there,

In the following snippet, I was expecting the dumped type to be non-empty-string, but that is lost somewhere and broadened to string

Thanks!

Code snippet that reproduces the problem

https://phpstan.org/r/15313fea-5813-4d90-89f9-38f27cba1e32

Expected output

Dumped type: non-empty-string

Did PHPStan help you today? Did it make you happy in any way?

Of course. how did we ever code php without it?

[stof]

My guess is that the native type you have on the function disables type inference from the array_map call for that function.

[dkarlovi]

It works for class-string so it's specific to non-empty-string.

[EmielM]

This also applies to union of string literals, eg 'a'|'b' that - in template inference - becomes a string.

Simple example: https://phpstan.org/r/529665e1-5525-4b4f-81bf-f1848e852ae8

Our use case where this is often an issue is a generic "tryParse" function: https://phpstan.org/r/12dc41cd-2f6d-4485-ab2c-fdbfb749e067

[stof]

Well, in this case, this is quite a good thing though. Otherwise, it would be impossible to make anything else than an identity function for any generic method allowing scalar types for its bound.

[ondrejmirtes]

Last night I came up with the idea that we mostly shouldn't generalize the generic type variables, except when they're in object generics, like Foo<int>.

Here's the resulting PR: phpstan/phpstan-src#2818

We can't do this for objects, because I want new Collection([1, 2, 3]) to become Collection<int>. This could be improved in the future thanks to this suggestion: #6732 (comment) (but it's pretty complex to implement so for now it has to wait).

The new behaviour now only applies to bleeding edge (https://phpstan.org/blog/what-is-bleeding-edge) so definitely enable it to get the taste of the future 👍

[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.