Check types of missing definition in add import quick suggestions

[ilyakooo0]

Current behavior:

Even though the type of toLower is known from the context (Char -> Char), "quick fix" purposes to import functions that, despite having the same name, are known not to be correct type-wise (Data.Text.Lazy (toLower) for example).

Proposed behavior:

When the type of the undefined function is fully known, filter quick fixes by type.

[jneira]

This would be a really great improvement and i dare to say it should no be too difficult. Also, i think i read that suggestion in another issue but i cant find it out right now.

[pepeiborra]

There are two quick fixes involved here: suggest new import and suggest extend import. The code for these quick fixes is at:

haskell-language-server/ghcide/src/Development/IDE/Plugin/CodeAction.hs

Lines 796 to 805 in 0b3bb10

	suggestExtendImport :: ExportsMap -> ParsedSource -> Diagnostic -> [(T.Text, CodeActionKind, Rewrite)]
	suggestExtendImport exportsMap (L _ HsModule {hsmodImports}) Diagnostic{_range=_range,..}
	| Just [binding, mod, srcspan] <-
	matchRegexUnifySpaces _message
	"Perhaps you want to add ‘([^’]*)’ to the import list in the import of ‘([^’]*)’ *\\((.*)\\).$"
	= suggestions hsmodImports binding mod srcspan
	| Just (binding, mod_srcspan) <-
	matchRegExMultipleImports _message
	= mod_srcspan >>= uncurry (suggestions hsmodImports binding)
	| otherwise = []

and at

haskell-language-server/ghcide/src/Development/IDE/Plugin/CodeAction.hs

Lines 1241 to 1250 in 0b3bb10

	constructNewImportSuggestions
	:: ExportsMap -> (Maybe T.Text, NotInScope) -> Maybe [T.Text] -> [(CodeActionKind, NewImport)]
	constructNewImportSuggestions exportsMap (qual, thingMissing) notTheseModules = nubOrdOn snd
	[ suggestion
	| Just name <- [T.stripPrefix (maybe "" (<> ".") qual) $ notInScope thingMissing]
	, identInfo <- maybe [] Set.toList $ Map.lookup name (getExportsMap exportsMap)
	, canUseIdent thingMissing identInfo
	, moduleNameText identInfo `notElem` fromMaybe [] notTheseModules
	, suggestion <- renderNewImport identInfo
	]

In order to implement this, you will need to:

1. retrieve the type of the identifier I that is out of scope,
2. the type of the import candidate C
3. check if they are compatible

I'm not really sure what's the best way to retrieve I, possibly parsing the error message?

Retrieving C should be easy - just look through the tcg_type_env in the TcGblEnv.

To check if they are compatible, I would do this with tcUnifyTy.

[Javran]

The error message seems like a stable one to match against (data constructor has a slightly different message):
https://github.com/ghc/ghc/blob/4c87a3d1d14f9e28c8aa0f6062e9c4201f469ad7/compiler/GHC/Tc/Errors.hs#L1210-L1211
plus many GHC testsuite relies on this exact string.

I'll try with approach outlined in #754 (comment) and see how it goes.

[July541]

@pepeiborra Sorry to disturb you, I took a look at this and found the type of identifier can get from TcGblEnv by TcModuleResult, but candidates seem not. I have trouble connecting ExportsMap to TcModuleResult, do you have any suggestions?

[pepeiborra]

@July541 have you seen my recent comment in #2081 about this? It describes an easier approach, albeit too slow.

To answer your question: To retrieve the type of the candidate you would need to use the GHC api, similar to what ghci :info does. I'm not on my PC but you can easily find the code for :info by grepping for infoCmd in the ghc repository.