core: (irdl) ParamAttrConstraint can infer recursively

[superlopuh]

This seems to work, is there any reason not to do this?

Note: conflicts with #3456

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.29%. Comparing base (d3647b2) to head (8afe11e).
Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3477      +/-   ##
==========================================
+ Coverage   90.26%   90.29%   +0.02%     
==========================================
  Files         461      462       +1     
  Lines       57807    57881      +74     
  Branches     5565     5564       -1     
==========================================
+ Hits        52180    52261      +81     
+ Misses       4190     4187       -3     
+ Partials     1437     1433       -4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[alexarice]

I don't believe this is sound, ParamAttrConstraint only specifies the base class of the attribute and so could theoretically this could infer the wrong subclass?

[alexarice]

Note that I'm not sure if being able to capture subclasses is part of the intended semantics of BaseAttr and ParamAttrConstraint, and this might just be an artifact of using python's isinstance

[superlopuh]

Ah yes, although I'm still not sure that we should officially support the whole class hierarchy thing. The error scenario here is that the user specifies a superclass in the constraint and there's a crash at runtime if the superclass is abstract? To avoid this, we could probably check that the class has a name field?

[alexarice]

Could do an is_runtime_final check?

[alexarice]

A similar question to this is whether a BaseAttr applied to a type with no parameters should be inferable

[math-fehr]

I think I either forgot to do this, or at the time is_runtime_final didn't exist yet!
So yeah that looks good to me!

[math-fehr]

A similar question to this is whether a BaseAttr applied to a type with no parameters should be inferable

Yes it should! What we could do is write an optimization to "convert" a BaseAttr with a singleton type to an EqType, that would also work!