Always use typeEquivAux EraseMeasures (for integral range optimizations)

[brianrourkeboll]

Description

Followup to #16650 and #17040.

• Always use typeEquivAux EraseMeasures instead of stripMeasuresFromTy for the integral range optimizations. I should have done this in Use alternative way of ignoring measures when detecting integral types #17040 to begin with. Note that the appropriate unsigned type and checks are now correctly present in the emitted IL.

Checklist

• Test cases updated.
• Release notes entry updated.

[github-actions]

❗ Release notes required

---

✅ Found changes and release notes in following paths:

Change path	Release notes path	Description
src/Compiler	docs/release-notes/.FSharp.Compiler.Service/8.0.400.md

[psfinaki]

Is it worth adding any more tests here compared to the ones in #17040?

[brianrourkeboll]

@psfinaki

Is it worth adding any more tests here compared to the ones in #17040?

The tests added in #17040 actually already showed this problem — that's why the baselines needed to be updated in this PR. I just didn't notice that there were additional places affected by this in #17040 because I was focusing on (1) whether the code compiled at all and (2) that the generated IL used the new for/while-loops instead of calling the range operator.

I think that these tests (the ones added in #17040 and updated here) are enough to show that the class of bug caused by using stripMeasuresFromTy instead of typeEquivAux EraseMeasures is fixed.

If we wanted to be more thorough in general, though, we really should add duplicates of all the existing emitted IL tests for integral ranges but with units of measure — although we would then be faced with #17046 and fsharp/fslang-suggestions#535. We might even add additional tests that asserted that the emitted IL for the unit-less versions matched the emitted IL for the versions with units.

As a sanity check, we could also/alternatively add duplicates with units of measure of all the existing non-emitted-IL tests for ranges, like these, though again we'd need to work around fsharp/fslang-suggestions#535.

[psfinaki]

@brianrourkeboll thanks for the detailed explanation. I think this can go as is then, the tests you suggest would be helpful but definitely a separate and potentially big effort.

Thanks!