No completions for requests just before String

[rochala]

Compiler version

latest.nightly

Minimized code

object M:
  val VersionRegex = "".r
  Ver@@"1234"

Output

Nothing

Expectation

VersionRegex completion should be returned in this position.

[som-snytt]

It looks like an interpolator.

[NPCRUS]

@rochala since we wasn't able to tackle it on a spree and I think I already saw some files and places that might help me tackle this issue, would you mind if i give it a go first?

[Yummy-Yums]

@rochala this is my github

[rochala]

@rochala since we wasn't able to tackle it on a spree and I think I already saw some files and places that might help me tackle this issue, would you mind if i give it a go first?

Noo, give it a try :D let me know if you have any questions :D Good luck

[NPCRUS]

@rochala
I boiled it down to Completion.selectionCompletions, where we are trying to extract implicitConversionMembers, extensionMembers, directMembers, namedTupleMembers. I deduce that we should focus here on directMembers, which goes down to Completions.accessibleMembers and here VersionRegex is not there even before filtering.
on the other hand at this point our qualifier is this:
_root_.scala.StringContext.apply(["1234" : String]*)
I'm not sure, but I have a feeling that we are missing Ver here in the tree, so essentially for a completion it's just trying to search for available methods on StringContext.
At this point I don't know what is right or wrong, so i need some feedback on my findings.

[rochala]

Hmm, I'd recommend checking this part: https://github.com/scala/scala3/blob/main/compiler/src/dotty/tools/dotc/interactive/Completion.scala#L218-L226

If you print the current path you can see that we have a 'Select' node as head. Currently the completions are missing because we enter 'Map.empty' branch. Maybe new case that goes for 'scopeCompletions' will suffice? You will also need to adjust 2 other methods: one that calculates completion mode and probably one that calculates prefix. Just be aware of tpd vs untpd trees.

[rochala]

This is very tricky case, because we should not return only string context methods here. My idea behind this issue is that sometimes you want to call a method or anything on a string that is currently present, but right now there are no completions. That said, yes, you will probably need to handle the String contexts(I forgot about that part). They should be searched at that point. This complicates the issue though

I can't check it for sure until 1st of march, but extensionMethods and implicit conversions should return handle the search for StringContext qualifier

[NPCRUS]

We actually do not enter the third case, which returns Map.empty, we enter second case which proceeds into completer.selectionCompletions and then in the end it resolves into empty list of completions

[NPCRUS]

@rochala I looked more into it, and the more i look the more confused I become. By which mechanism should Ver@@"1234" be completed with VersionRegex? when we look into qualifier this line is desugared into StringContext("1234").Ver, obviously(also you mentioned) we should complete it through implicitConversionMembers or extensionMembers, for this to happen StringContext("1234").VersionRegex() should be valid/compilable syntax through implicit conversion or extension, but i didn't find this to be the case. On top of that interpolators like raw are also not being completed in a similar way

[rochala]

Hey, sorry for such long wait for me to answer. There is another spree coming this week, if you want I welcome you to join, and we can try handling this together, this issue turned out to be a bit harder than I estimated.

[NPCRUS]

@rochala I'd love to, but i have a flight exactly at a time of spree... Could we perhaps collaborate in any other way outside of the spree on this?

[rochala]

Ofc, lets talk about this on Discord, I'll ping you