Autocompletion no longer counts qualified imports

[Akhristenko]

Steps to reproduce

I have 2 modules

module MyModule (myid) where

myid :: a -> a
myid a = a

module MyModuleB (myid') where

import qualified MyModule

myid' :: a -> a
myid' = MyModule.myid

I delete myid in second module and invoke autocompletion when cursor placed after MyModule.

Expected behaviour

In autocompletion list i see only functions from MyModule

Actual behaviour

In autocompletion list i see all function from all modules

[pepeiborra]

Thanks for reporting, this is likely a regression after #2040

[pepeiborra]

I'm not so sure this is a regression anymore. It's actually a new feature! Previously autocompletion only suggested imported identifiers, now it suggests also unimported identifiers and automatically inserts a new import if necessary. Qualifiers are no longer used only for filtering, but also to inform the format of the inserted import. However qualifiers are still used for filtering, and list of completions should put the functions from current imports first. Can you confirm whether this is the case for you?

[Akhristenko]

However qualifiers are still used for filtering, and list of completions should put the functions from current imports first.

It's not true, at least on version 1.3.0 it show all names in alphabetical order. But if hls will showing names imported in qualifiers ordered alphabetically before all other names it will solve this issue.

[Akhristenko]

ideally i would like this behavior:

• if we invoke autocomplete on qualifier, which was not imported, we see all functions and autoimport perform after we select one of them.
• if we invoke autocomplete on qualifier, which was imported, we see only functions which was imported. But if we invoke autocomplete second time we see all functions and can autoimport one of them.

[alexnaspo]

I see a decent path forward on this one, is it worth picking up? @pepeiborra

[pepeiborra]

I would start with a set of tests demonstrating the problem, and then we can discuss what needs to be done to enforce the behaviour that I outlined above.

Implementing the behaviour that @Akhristenko suggested is not something I would recommend. "If we invoke autocomplete a second time" is not a condition that the LSP server can detect.