[Diagnostics] Improve keypath diagnostic when cannot be inferred by context

[LucianoPAlmeida]

Improving the diagnostics when the key path type can't be inferred by context. e.g. let _ : AnyKeyPath = \.x

cc @xedin @hborla @varungandhi-apple :)

[varungandhi-apple]

@swift-ci please test

[xedin]

Thanks for looking into this, @LucianoPAlmeida!

Couple of thoughts:

• I think we should split this into two separate PRs, one for contextual stuff and other one for keyPath: overload.

• Regarding keyPath:

  • I think we can do better in this case and instead of saying maybe we could actually verify that keyPath: overload choice works (good that it doesn't have to be a declaration :)) and suggest adding keyPath: label.

• Regarding root/value inference failure:

  • Is it any different from a regular inability to infer generic arguments?
  • I think we should add a special diagnostic about AnyKeyPath which state that it doesn't
    provide enough contextual information to allow root type to be inferred implicitly
    (because its root type is Any). That should help resolve at least some confusion here...
  • Could we suggest a fix-it in this case to specify <#Root#> before a leading dot ?

[LucianoPAlmeida]

• I think we should split this into two separate PRs, one for contextual stuff and other one for keyPath: overload.

Sure! Let's handle contextual first here then :)

• I think we can do better in this case and instead of saying maybe we could actually verify that keyPath: overload choice works (good that it doesn't have to be a declaration :)) and
  suggest adding keyPath: label.

Noted :)

• Is it any different from a regular inability to infer generic arguments?

I didn't thought about that, perhaps there is no difference...
but I think we can tailor the diagnostics for key paths since generic arguments is more general...?

• I think we should add a special diagnostic about AnyKeyPath which state that it doesn't
  provide enough contextual information to allow root type to be inferred implicitly
  (because its root type is Any). That should help resolve at least some confusion here...

• Could we suggest a fix-it, in this case, to specify <#Root#> before a leading dot ?

I'll work on those, thanks @xedin :)

[xedin]

I think we can tailor the diagnostics for key paths since generic arguments is more general.

I agree that we can tailor diagnostic messages but I think emission of the fix/diagnostic should be based on what we do for generic parameters.

[artemcm]

I was accidentally concurrently looking into this issue. From what I gather, this PR is now resolving the more general issue with key path types not being inferred from context, but not the issue of the keyPath: label overload?

[LucianoPAlmeida]

I was accidentally concurrently looking into this issue. From what I gather, this PR is now resolving the more general issue with key path types not being inferred from context, but not the issue of the keyPath: label overload?

Initially, it was trying to handle both, but I just remove the commit since @xedin suggested split in two PRs, will do a follow up as soon as this one lands :)

[artemcm]

@LucianoPAlmeida, I can't see the removed commits, but please take a peak at #31716, I imagine it is something similar?

[varungandhi-apple]

I'm a bit confused as to what's going to happen here? Are we merging Artem's PR or this one? I can't comment on the implementation but I think the diagnostic looks good. 👍

@swift-ci please test

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 87083743c6cff0d750ef59744b6bb5d3b8862f57

[LucianoPAlmeida]

I'm a bit confused as to what's going to happen here? Are we merging Artem's PR or this one? I can't comment on the implementation but I think the diagnostic looks good. 👍

@varungandhi-apple We decided to split this PR into two parts contextual inference issues, and subscript fixes as a follow-up. So we let this one only to solve contextual issues and planning a follow up for the keyPath label... but coincidentally the second part is already handled by Artem's PR :)

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 87083743c6cff0d750ef59744b6bb5d3b8862f57

[LucianoPAlmeida]

I think Artem's covers PR the whole scope of the SR, so I removed the test cases related and will let this be only contextual stuff :)

[xedin]

@LucianoPAlmeida I think it's from context instead of by context but I might be wrong...

[xedin]

@swift-ci please smoke test