Point at type ascription before macro invocation on expansion parse error

[estebank]

Fix#47666. Follow up to #62791.

r? @petrochenkov

[estebank]

The only relevant change is the last commit, which adds diagnostic information to Mac_ to keep the state of last_type_ascription at the invocation place location, to be able to ferry it when expanding the macro through ExpansionData.

[petrochenkov]

I need to think whether a less hacky solution exists, I wouldn't want to merge this as is.

[estebank]

Given the way this works it seems to me like the only available option is to collect the state of the parser at the invocation place. I understand the hesitance of merging the PR as is, but I think it is as good a starting point as we're gonna find without a larger refactor.

[edmilsonefs]

Hey! This is a ping from triage, we would like to know if you @petrochenkov could give us a few minutes to share your thoughts on it.

[petrochenkov]

Low priority, deferred until the weekend (unless I deal with other reviews earlier).

(@estebank, could you rebase this perhaps in the meantime? #62791 has landed.)

[petrochenkov]

So, the general idea is to create a parser snapshot for every macro invocation, so when the macro is expanded we can parse its output using a parser from that snapshot rather than a fresh parser.

Except that the only part of the parser this PR snapshots is the prior_type_ascription field.
And the snapshots are kept directly in the AST and not in some side table because we don't have node IDs assigned this early.

Since we create a fresh parser when expanding any macro anyway (which is not an entirely cheap operation), I wonder, how much more expensive it would be to clone the existing one instead.
Let's land this PR in the meantime, the diff is not large, it won't be hard to rework or revert if necessary.

[petrochenkov]

@bors r+

[bors]

📌 Commit c82e1f2 has been approved by petrochenkov

[bors]

⌛ Testing commit c82e1f2 with merge f01b9f8...

[bors]

☀️ Test successful - checks-azure
Approved by: petrochenkov
Pushing f01b9f8 to master...