core: remove generic and generic check from TypedAttribute

[alexarice]

As discussed in #3492, this check feels unnecessary and is unwanted for making DenseIntOrFPElementsAttr a TypedAttribute

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.34%. Comparing base (020cae7) to head (dffa305).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3504      +/-   ##
==========================================
- Coverage   90.34%   90.34%   -0.01%     
==========================================
  Files         464      464              
  Lines       58052    58037      -15     
  Branches     5550     5549       -1     
==========================================
- Hits        52448    52434      -14     
  Misses       4177     4177              
+ Partials     1427     1426       -1     

☔ 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

[superlopuh]

Do you know why it was added in the first place? I'm wondering why it wasn't just using OO, is it for precise placement of the type variable in the list of parameters? It feels like that should be possible by redeclaring the type property at the desired index. Do you feel like that would be worth a go? It would also decouple the dense stuff, I have a feeling.

[alexarice]

I'm not really sure why it was there, and @math-fehr wasn't sure either. I kind of feel that enforcing that the type appear as a parameter at all rather than just as an interface method on TypedAttribute is overkill, but I guess it was there for irdl things.

I'm actually really struggling to get all the variances correct on dense PR atm so any suggestions on sorting this stuff would be good.

[superlopuh]

Would you like me to have a go at my own suggestion?

[alexarice]

Go for it. I'm starting to wonder if the generic parameter on TypedAttribute can be removed completely

[superlopuh]

My attempt was a failure, let's go back to this idea. Is there any way to keep the unit test deleted here? So we keep the fact that the type is indeed the same as the TypedAttribute parameter, but check it differently?

[alexarice]

Is there any way to keep the unit test deleted here?

What do you mean?

[superlopuh]

Like to make the check work differently, instead of looking for the parameter of the Generic, look for the parameter of `TypedAttribute1?

[alexarice]

I would prefer just to remove the generic parameter on TypedAttribute

[math-fehr]

I just had a comment but otherwise nice!

[alexarice]

@superlopuh you happy with this (and the next PR)?

[superlopuh]

Oh this is a case where I think I'm only happy with this together with the next PR, as after this PR we can declare something as typed with a certain type, but remove the code that verifies this at runtime. If TypedAttribute is no longer generic then this PR is dead code elimination.

[alexarice]

Should I merge the 2 PRs?

[superlopuh]

I would recommend mergint them in one go, yep

[alexarice]

Great, I'm just merging main into the second one to update FloatAttr

[alexarice]

That did not work...