Implement signature help #4626

jian-lin · 2025-06-08T21:49:42Z

Closes #3598

I will update progress here.

2025-06-02 - 2025-06-08

Add basic boilerplate for signature help plugin
- commit: 7a54a1d
- click for screenshot

2025-06-09 - 2025-07-13

Finish signature help plugin MVP: show function signature and highlight the current parameter
- commit: 9168b74
- click for video demo
  
  Screencast.From.2025-07-13.02-11-44.webm

2025-07-14 - 2025-07-20

Add basic tests. Some of them are not passed for now.
- commit: 62fbccf

2025-07-21 - 2025-07-27

Change expected test results considering the cursor shape (a6635ca)
Replace maybe with case for better readability (bf0b4d5)
Call extractInfoFromSmallestContainingFunctionApplicationAst once (c95d6e4)

2025-07-28 - 2025-08-03

Show more types: each type as one signature help (d603ec4)
Add tests for kind signatures and higher-order function (dca1311) (471958f)

2025-08-04 - 2025-08-10

Use structured type and type string to generate signature help (9bea0e3)
Add more tests
- middle => (fdd5acd)
- => in argument (c6ece42)
- RankNTypes(forall in middle) (fe2d618 and 081ea8f)
- LinearTypes (c3224d7)

michaelpj · 2025-06-09T07:53:42Z

hls-plugin-api/src/Ide/Types.hs

@@ -764,6 +770,10 @@ instance PluginRequestMethod Method_TextDocumentDocumentSymbol where
            si = SymbolInformation name' (ds ^. L.kind) Nothing parent (ds ^. L.deprecated) loc
        in [si] <> children'

+-- TODO(@linj) is this correct?


I think this is reasonable. Really combineResponses should have a way to return an error. There are a bunch of methods where it really only makes sense if we have one handler.

We could try to combine responses: we would combine the signatures, and so long as only one of them indicated an active signature it would be okay. But that's a bit sketchy and I doubt we'll have several anyway!

TODO: - handle more cases - add successful and (currently failed) tests - show documentation

michaelpj

I think really worth trying to start getting some tests in place!

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

michaelpj · 2025-07-13T09:08:52Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

+    if nodeHasAnnotation ("HsVar", "HsExpr") hieAst
+       then
+           case M.elems $ M.filter isUse $ getSourceNodeIds hieAst of
+               [identifierDetails] -> identType identifierDetails >>= (prettyType >>> Just)


I suggest not printing the type here and instead returning the type itself. That would mean that when you're doing mkArguments you can pattern match on the type itself, not on the string representation. Then you don't need to worry about e.g. whether it prints on multiple lines or whatever.

See #4626 (comment)

michaelpj · 2025-07-13T09:09:33Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

+            (Just $ InL argumentNumber)
+
+-- TODO(@linj) can type string be a multi-line string?
+mkArguments :: UInt -> Text -> [ParameterInformation]


I think this would be a lot simpler if you were just looking at the type itself!

Although hmm, the offsets are into the string 😱 So maybe you would need to incrementally build up the ParameterInformations and the overall label, but that would probably still be nicer?

See #4626 (comment)

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

michaelpj · 2025-07-13T09:14:09Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

+nodeHasAnnotation :: Annotation -> HieAST a -> Bool
+nodeHasAnnotation annotation = sourceNodeInfo >>> maybe False (isAnnotationInNodeInfo annotation)
+
+-- TODO(@linj): the left most node may not be the function node. example: (if True then f else g) x


write a test!

this is a great example btw. In this case I think we would have to give no signature help, I think, since it's only dynamically known which function we're getting.

We have one related test case for this now: the dynamic function test.

This test is not a unit test for this getLeftMostNode function. This test is an integration testing for the signature help plugin.

f :: Int -> Int -> Int f = _ g :: Int -> Int -> Int g = _ x = (if _ then f else g) 1 2 ^^ ^^^ ^ ^^^ ^ ^^^^^^^^

michaelpj · 2025-07-13T09:36:00Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

+    if nodeHasAnnotation ("HsVar", "HsExpr") hieAst
+        then
+            case mapMaybe extractName $ M.keys $ M.filter isUse $ getSourceNodeIds hieAst of
+                [name] -> Just name -- TODO(@linj) will there be more than one name?


I'm not 100% sure when/if this can happen. I think it might happen if we're looking at a typeclass method, in which case we could get the generic typeclass method and the specific one perhaps? That would be a good use-case for multilple signatures, e.g. if we have

foldMap @[] f ^

then we might want to see both the generic Foldable signature and the one for the list instance.

I think my general suggestion would be: let's treat possibly-multiple node annotations as producing possibly-multiple signatures.

Implemented in d603ec4.

Related tests:

type constraint

type constraint with kind signatures

multi-line type with type constraint

I'm not 100% sure when/if this can happen.

I do not have a principled conclusion. But from my experiments, the summary is that

There are multiple types iif there are type variables.

There seems to be only one Name.

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

@fendor

This is discussed with @fendor.

This improves performance. It also improves correctness because functionName, functionType and argumentNumber are extracted from the same AST.

Vscode and Emacs(eglot) seems to treat newline as a normal char.

See comment for a comparison with alternative methods.

jian-lin · 2025-08-07T09:26:21Z

All tests on Linux and darwin passed! Not sure what happens on windows.

Added a few more tests and now 3 out of 87 tests failed.

Test 5 does not pass.

jian-lin · 2025-08-07T18:19:22Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

a bit offtopic: Is 2-space indentation instead of 4 allowed for new code?

The .editorconfig says 4.

However, some newly added plugins use 2:

hls-cabal-gild-plugin by @fendor

hls-stan-plugin

Follow the .editorconfig :)
hls-cabal-gild-plugin was copy-pasted from the hls-cabal-fmt-plugin

jian-lin · 2025-08-07T18:34:02Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

+findArgumentRanges :: Type -> [(UInt, UInt)]
+findArgumentRanges functionType =
+    let functionTypeString = printOutputableOneLine functionType
+        functionTypeStringLength = fromIntegral $ T.length functionTypeString
+        splitFunctionTypes = filter notTypeConstraint $ splitFunTysIgnoringForAll functionType
+        splitFunctionTypeStrings = printOutputableOneLine . fst <$> splitFunctionTypes
+        -- reverse to avoid matching "a" of "forall a" in "forall a. a -> a"
+        reversedRanges =
+            drop 1 $ -- do not need the range of the result (last) type
+                findArgumentStringRanges
+                    0
+                    (T.reverse functionTypeString)
+                    (T.reverse <$> reverse splitFunctionTypeStrings)
+     in reverse $ modifyRange functionTypeStringLength <$> reversedRanges
+    where
+        modifyRange functionTypeStringLength (start, end) =
+            (functionTypeStringLength - end, functionTypeStringLength - start)
+
+{-
+The implemented method uses both structured type and unstructured type string.
+It provides good enough results and is easier to implement than alternative
+method 1 or 2.
+
+Alternative method 1: use only structured type
+This method is hard to implement because we need to duplicate some logic of 'ppr' for 'Type'.
+Some tricky cases are as follows:
+- 'Eq a => Num b -> c' is shown as '(Eq a, Numb) => c'
+- 'forall' can appear anywhere in a type when RankNTypes is enabled
+  f :: forall a. Maybe a -> forall b. (a, b) -> b
+- '=>' can appear anywhere in a type
+  g :: forall a b. Eq a => a -> Num b => b -> b
+- ppr the first argument type of '(a -> b) -> a -> b' is 'a -> b' (no parentheses)
+- 'forall' is not always shown
+
+Alternative method 2: use only unstructured type string
+This method is hard to implement because we need to parse the type string.
+Some tricky cases are as follows:
+- h :: forall a (m :: Type -> Type). Monad m => a -> m a
+-}
+findArgumentStringRanges :: UInt -> Text -> [Text] -> [(UInt, UInt)]


Context:

Implement signature help #4626 (comment)

Implement signature help #4626 (comment)

Implement signature help #4626 (comment)

@michaelpj

I implemented a mixed way which uses both structured type and type string. See the comment for the reason of doing it this way instead of using only structured type or only type string.

It seems to work pretty well. "3 out of 87 tests failed".

One failed test case is f :: Integer -> Num Integer => Integer -> Integer. We matched Num Integer as the first argument. This probably can be fixed by using regex.

Another similar failed test case is f :: forall l. l -> forall a. a -> a. When we should highlight the argument l, we highlight the l for the second forall.

Here is another failed test case.

f :: a -> forall a. a -> a f = _

The printed type string for f is f :: forall a. a -> forall a1. a1 -> a1. This seems tricky to fix in the current implementation because it needs us to duplicate the renaming logic (the second forall a becomes forall a1) of ppr.

This case only happens when RankNTypes is used so I do not think it is very common.

jian-lin · 2025-08-07T18:37:03Z

plugins/hls-signature-help-plugin/src/Ide/Plugin/SignatureHelp.hs

+nodeHasAnnotation :: Annotation -> HieAST a -> Bool
+nodeHasAnnotation annotation = sourceNodeInfo >>> maybe False (isAnnotationInNodeInfo annotation)
+
+-- TODO(@linj): the left most node may not be the function node. example: (if True then f else g) x


We have one related test case for this now: the dynamic function test.

This test is not a unit test for this getLeftMostNode function. This test is an integration testing for the signature help plugin.

f :: Int -> Int -> Int f = _ g :: Int -> Int -> Int g = _ x = (if _ then f else g) 1 2 ^^ ^^^ ^ ^^^ ^ ^^^^^^^^

WIP: add basic boilerplate for signature help plugin

7a54a1d

michaelpj reviewed Jun 9, 2025

View reviewed changes

WIP: finish signature help plugin MVP

9168b74

TODO: - handle more cases - add successful and (currently failed) tests - show documentation

jian-lin force-pushed the pr/signature-help branch from 7a02359 to 9168b74 Compare July 12, 2025 18:33

michaelpj reviewed Jul 13, 2025

View reviewed changes

jian-lin added 3 commits July 16, 2025 13:51

WIP: remove unused dependencies

4d7fa57

WIP: fix func-tests

622c8ec

WIP: add basic tests

62fbccf

jian-lin force-pushed the pr/signature-help branch from 67f5cdb to 62fbccf Compare July 16, 2025 06:52

jian-lin added 9 commits July 25, 2025 15:45

Change expected test results considering the cursor shape

a6635ca

This is discussed with @fendor.

Replace maybe with case for better readability

bf0b4d5

Call extractInfoFromSmallestContainingFunctionApplicationAst once

c95d6e4

This improves performance. It also improves correctness because functionName, functionType and argumentNumber are extracted from the same AST.

Add missing ParameterInformation for multi-line tests

3dc1ec8

Vscode and Emacs(eglot) seems to treat newline as a normal char.

Add a signature help test for type constraint with kind signatures

dca1311

Add a signature help test for higher-order function

471958f

Show more types: each type as one signature help

d603ec4

Adjust import for the removal of hie-compat (haskell#4613)

faa4e48

Use structured type and type string to generate signature help

9bea0e3

See comment for a comparison with alternative methods.

jian-lin added 5 commits August 8, 2025 02:17

Add signature help test: 2 type constraints

db5e59e

Add signature help test: middle =>

fdd5acd

Test 5 does not pass.

Add signature help test: => in argument

c6ece42

Add 2 signature help tests: RankNTypes(forall in middle)

fe2d618

Add signature help test: LinearTypes

c3224d7

jian-lin commented Aug 7, 2025

View reviewed changes

Add another signature help tests: RankNTypes(forall in middle)

081ea8f

Uh oh!

Implement signature help #4626

Are you sure you want to change the base?

Implement signature help #4626

Uh oh!

Conversation

jian-lin commented Jun 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

2025-06-02 - 2025-06-08

2025-06-09 - 2025-07-13

2025-07-14 - 2025-07-20

2025-07-21 - 2025-07-27

2025-07-28 - 2025-08-03

2025-08-04 - 2025-08-10

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelpj left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jian-lin commented Aug 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jian-lin Aug 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jian-lin commented Jun 8, 2025 •

edited

Loading

jian-lin commented Aug 7, 2025 •

edited

Loading

jian-lin Aug 7, 2025 •

edited

Loading