Go to definition doesn't work for record syntax

[newhoggy]

Doesn't work for constructor

Doesn't work for fields

[cocreature]

Thanks for the bug report, that definitely sounds like something that we should fix. If you want to take a look yourself, my guess would be that getSrcSpanInfos is the culprit.

[jacg]

It seems to work fine here:

... both for constructors and fields (though for fields it still refers to the position of the constructor).

Can you provide a minimal failing example?

[newhoggy]

I defined person0 and person1.

The only time I get the popup is when I hover over the Person constructor in person1.

And Go To Definition doesn't work anywhere in the person0 case.

[ndmitchell]

I get definitions for the Person constructor in both person0 (due to https://github.com/digital-asset/ghcide/blob/master/src/Development/IDE/Spans/Calculate.hs#L122) and person1 (due to https://github.com/digital-asset/ghcide/blob/master/src/Development/IDE/Spans/Calculate.hs#L121). I don't get information about name or age because it doesn't have any cases that examine the record fields. This whole piece of code was copied wholesale from Intero, upgraded through a few GHC versions, lightly hacked for "things that didn't work", and could do with someone giving it some love. However, it shouldn't be that hard to make it a lot more powerful. If using SYB was OK then finding all names and grabbing the nearest enclosing Loc would probably be perfect.

[jacg]

If using SYB was OK

Has SYB been proscribed?

[ndmitchell]

SYB seems to be responsible for a lot of problems, e.g. haskell/haskell-ide-engine#1371. It's not fast and it is memory hungry. That said, it is concise and if it can be formulated properly, it's probably correct by construction.

[jacg]

I got tired of trying to verify manually what works and what does not, so I wrote some automated tests, which you can find here.

The one entitled "plain data constructor" passes while the one entitled "record data constructor" fails, which seems to agree with what @newhoggy observes above, as well as my own observations in both VSCode and Emacs; but it disagrees with what @ndmitchell observes above.

I get definitions for the Person constructor in [...] person0 (due to https://github.com/digital-asset/ghcide/blob/master/src/Development/IDE/Spans/Calculate.hs#L122)

Removing line 122, has no observable effect on any of the tests I have written. (I have confirmed that removal of lines 120, 121, or 147 does break some of my tests).

[jacg]

This whole piece of code [...] could do with someone giving it some love.

I'd like to have a go, but would benefit immensely from some pointers on how best to get stuck in.

[ndmitchell]

@mpickering ported the hie code over as part of https://github.com/digital-asset/ghcide/issues/101#issuecomment-534498465. If we can use that implementation rather than our own that makes most sense - less code = better.

[mpickering]

I have wanted to experiment with geniplate, in
order to generate efficient traversal functions. That might be worth looking into.

[jneira]

I've just test it in hls and works for me, i guess it should work in ghcide too.
@newhoggy feel free to reopen if it is no the case