Jump to instance definition and explain typeclass evidence #4392
Conversation
fendor
changed the title
fendor
force-pushed
the
fendor/wip/evidence-info
branch
2 times, most recently
from
0ccc4d3
to
5568ca3
Compare
|
Just at a high level, am I right that if I have and I call "go to implementation" on "foo" I will get the option to go to the |
|
Yes, that's exactly what will happen! |
| -> Position | ||
| -> MaybeT m [(Location, Identifier)] | ||
| gotoDefinition withHieDb getHieFile ideOpts imports srcSpans pos | ||
| = lift $ locationsAtPoint withHieDb getHieFile ideOpts imports pos srcSpans | ||
|
|
||
| -- | Locate the implementation definition of the name at a given position. | ||
| -- Finds the implementation for a overloaded function. |
There was a problem hiding this comment.
I think we can keep saying "gotoImplementation" and so on, since this corresponds to that LSP method, but let's make this comment clear about what it actually does!
| ] | ||
| prettyName (Right n, dets) | ||
| -- We don't want to print evidence variables as they are generated. | ||
| | any isEvidenceUse (identInfo dets) = pure $ maybe "" (printOutputable . renderEvidenceTree) (getEvidenceTree rf n) <> "\n" |
There was a problem hiding this comment.
the maybe "" seems supicious. What case is that for?
There was a problem hiding this comment.
The getEvidenceTree may return Nothing for a variety of reasons. For example, when the name is unexpectedly not found in the reference map. I think in practice, it will always produce something.
We could log it if we failed to build the EvidenceTree for a given name.
| @@ -307,6 +331,39 @@ atPoint IdeOptions{} (HAR _ hf _ _ (kind :: HieKind hietype)) (DKMap dm km) env | |||
| UnhelpfulLoc {} | isInternalName name || isSystemName name -> Nothing | |||
| _ -> Just $ "*Defined " <> printOutputable (pprNameDefnLoc name) <> "*" | |||
|
|
|||
| -- We want to render the root constraint even if it is a let, | |||
| -- but we don't want to render any subsequent lets | |||
There was a problem hiding this comment.
I'm assuming this makes sense to someone who knows how evidence is represented in GHC...
|
Can we include a test to see whether this now gives us some kind of useful hover information on overloaded record dot usage? |
fendor
force-pushed
the
fendor/wip/evidence-info
branch
from
5568ca3
to
dfb7002
Compare
Adds the necessary instances for handling the request type `Method_TextDocumentImplementation`. Further, wire up the appropriate handlers for the "gotoImplementation" request.
fendor
force-pushed
the
fendor/wip/evidence-info
branch
from
dfb7002
to
dacc0b7
Compare
Rebased and slightly modified version of #1983
Still requires some tests and docs.