[CountOnNullRector] Should understand array/countable variable in trait method

[gnutix]

This PR adds two fixtures relating to trait handling by CountOnNullRector. Currently :

• PropertyWithinTraitMethod : passes 👍
• VariableWithinTraitMethod : fails 👎

If I change the Fixture from a trait to a class they both pass, so this is related to how a variable within a trait is handled.

I spent some time digging and tried to use \Rector\NodeTypeResolver\PerNodeTypeResolver\VariableTypeResolver::resolve() that seems to be written to handle resolution of variables (though it's used nowhere in Rector's code, only in its own Test file 🤔 ) by calling :

• resolveTypesFromScope($node), which calls...
• resolveNodeScope($node), which calls on \Rector\NodeTypeResolver\PHPStan\Collector\TraitNodeScopeCollector...
• getScopeForTraitAndNode($traitName, $node))

But it returned null, which was resolved to MixedType in resolveTypesFromScope and bubbled up back to \Rector\Php71\Rector\FuncCall\CountOnNullRector::refactor which passes the isNullableType check and thus refactors the code.

@TomasVotruba Any idea how to deal with this one ?

[gnutix]

Could this be related to phpstan/phpstan#1601 (fixed in the new release by PHPStan 0.12.5) ?

[TomasVotruba]

Could this be related to phpstan/phpstan#1601 (fixed in the new release by PHPStan 0.12.5) ?

We might guess, but testing would answer us.

[TomasVotruba]

master now has phpstan/phpstan 0.12.5.

I tried to re-run checks, but I think you'll have to rebase

[gnutix]

Last push force is the rebase on master. We'll see.

[gnutix]

@TomasVotruba Not related to PHPStan 0.12.5. Any idea ?

[TomasVotruba]

@gnutix What do you mean? Link might help

[gnutix]

I mean that I've rebased this PR on master, which includes PHPStan 0.12.15, and it didn't fix the broken test. So the issue remains, and I have no lead / clue on why this is broken nor how to fix this. So I'm asking for your help / guidance. :)

[TomasVotruba]

This might be more tricky. Traits sucks hard.

Are you ok with merging as it is?
I'll have to look at it locally with trial error.

[gnutix]

Sure!

[TomasVotruba]

Both passes after merge 👍 probably fixed by type refactoring

[TomasVotruba]

Or not... false positive: paths have changed from packages to rules

[TomasVotruba]

After 30-40 minutes of digging, traits are impossible to analyse without class by PHPStan. It provides many false positives on standalone analysis, so we better skip them.

If you find a stable way of using PHPStan for traits here, I'm open to the PR.