core: Add default valued attributes

[alexarice]

Adds the equivalents for the tablegen DefaultValuedAttribute and DefaultValuedOptionalAttribute.

This is a blocker for updating mlir.

---

@zero9178, would you be able to check if the test cases agree with what you think mlir should be doing, as you seemed to have some idea about what was going on here when I last asked about this.

Lastly is there a nice python way to specify that an opt_prop_def should return an Attribute instead of Attribute | None when it is specified with a default value?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.03%. Comparing base (84a215d) to head (91d67cd).
Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3305      +/-   ##
==========================================
+ Coverage   90.01%   90.03%   +0.01%     
==========================================
  Files         445      445              
  Lines       55965    56166     +201     
  Branches     5364     5390      +26     
==========================================
+ Hits        50379    50571     +192     
- Misses       4177     4179       +2     
- Partials     1409     1416       +7     

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

[zero9178]

Looks great to me and the implementation seems to capture the nuances of the differences between optional and non-optional default values attributes/properties as far as I can tell. Just one test-case is missing for full branch coverage I believe.

[zero9178]

Can we also add a test here where we parse in the operation as a generic but don't specify in the dictionary an optioanl attribute/property?

In MLIR the auto-generated getter (say getDefOptProp) should then still return True, while getAttr("def_opt_propt") would return null/missing

[alexarice]

I feel this test particularly but also the one above it should probably not live here, but am not sure exactly where they should go

[alexarice]

Moving it to test_operation_definition.py

[alexarice]

Also added some overloads to opt_attr_def and opt_prop_def so that pyright knows that the accessor always returns an attribute when a default value is specified

[alexarice]

I hit a bug with renamed properties when continuing to update mlir. Should hopefully be fixed in latest commit

[math-fehr]

Very cool!
I find ODS semantics quite weird, I didn't know you could have optional attributes with default values, I think it completely breaks the roundtripping.
Thanks for looking at it and implementing it!