Use exact print for suggest missing constraint code actions #1221
Conversation
The Rewrite abstraction is similar to D.IDE.GHC.ExactPrint.Graft but it does fewer things more efficiently: - It doesn't annotate things for you (so it doesn't destroy user format) - It doesn't provide a Monoid instance (for now) - It doesn't need a fully parsed source - It doesn't use SYB to perform the replacement - It doesn't diff to compute the result The use case is code actions where you don't have the SrcSpan that you need to edit at hand, and instead you need to traverse the AST manually to locate the declaration to edit
Tweaking the suggest constraints tests to reflect the increased precision in whitespace preservation
pepeiborra
force-pushed
the
implicit-params-exactprint-cleanup
branch
from
c7aaa05 to
bf12e63
Compare
| @@ -739,7 +739,7 @@ suggestConstraint df parsedModule diag@Diagnostic {..} | |||
| where | |||
| findMissingConstraint :: T.Text -> Maybe T.Text | |||
| findMissingConstraint t = | |||
| let regex = "(No instance for|Could not deduce) \\((.+)\\) arising from a use of" | |||
| let regex = "(No instance for|Could not deduce) \\((.+)\\) arising from" -- a use of / a do statement | |||
There was a problem hiding this comment.
why not push this into the regex? feels weird to have it as a comment next to the regex.
There was a problem hiding this comment.
I did not push it into the regex because I don't know if there are other suffixes to this sentence that could result in more opportunities for a quick fix.
Why is it weird to have a comment next to a regex? We do it in most regexes in this module
There was a problem hiding this comment.
It feels weird because you are saying "after this regex comes one of these two suffixes". But what you really mean is these two values have been observed in the wild, there might be more, who knows.
|
Not related directly to this PR, but I am currently moving ghc-exactprint into the GHC AST, and open to suggestions about what the API should be. Will be opening a discussion about it soon. |
This is a big PR with several changes, split in well defined commits for ease of review.
Main changes
1. Cache the annotated AST
If we are going to use ghc-exactprint in code actions, It makes sense to cache the annotated AST in the Shake graph
2. Introduce a Rewrite abstraction
Code actions that intend to preserve user formatting cannot use the existing
Development.IDE.GHC.ExactPrint.Graftabstraction, which throws away all format. So instead we introduce a weakerRewriteabstraction that does fewer things:The use case is code actions where you don't have the SrcSpan that you need to edit at hand, and instead you need to traverse the AST manually to locate the declaration to edit. The ergonomics are not as good as
Graftthough, there's certainly room to improve.3. Refactor the existing suggest missing constraint code actions to use exact print
This means replacing all the stringy code for editing instance/signature heads with a ghc-exactprint implementation which respects whitespace and other user formatting. As a bonus, I made it handle missing
Monadconstraints too. /cc @jhrcek who I believe was the original author.4. Add suggestions for missing implicit parameters
This was very easy to do and I have a sweet spot for implicit parameters :)