Return nil from passthrough-try-completion when table is nil

[wyuenho]

Since #4602 the passthrough-try-completion function returns a cons cell even the list of completions is empty, so completion-in-region-functions that rely on the old behavior to call the exit function with an empty string candidate and nil candidates, which will make the exit func throw. This PR fixes this issue.

[kiennq]

I wonder in which case candidate can be nil though. Have exit function called on nil seems to be bug of completion framework to me honestly.

[wyuenho]

candidate won't be nil when given to the the exit func as a parameter, but it can be an empty string, such as the case of point sitting at the end of a line.

candidate in the let form can be nil here when the CAPF returns nil as the list of candidates. In which case, everything that follows blows up.

[kiennq]

candidate in the let form can be nil here when the CAPF returns nil as the list of candidates. In which case, everything that follows blows up.

When CAPF return nil as candidates, I don't think the exit-fn can be triggered in that case unless that's the new thing from CAPF (and I'm curious of why that even a thing to consider). Normally exit-fn is invoked when a candidate is selected, if there's no candidate, normally nothing can be selected and nothing should be run.
I purposely fallback to a candidate item in the candidates list if the exit-fn is invoked with a new string instead of the string that contains the necessary properties. The blow up can happen if the candidate and candidates list are out-of-sync. In that case, it would be better to allow the blow up to happen instead of silent fail so we can try to fix that, as the out-of-sync shouldn't happen at all.
Do you have a repro of when this happens so we can take a closer look to figure out why the out-of-sync happens?

[wyuenho]

Ahhh, it appears I've opened a whole new can of worms...

Corfu's completion-in-region-function calls completion-try-completion, which will call lsp-completion-passthrough-try-completion, which just returns ("" . nil), then corfu proceeds to here, which is when the exit func is called with an empty string and an empty candidates list.

I guess the correct fix should be that lsp-completion-passthrough-try-completion should return nil when table is nil.

[wyuenho]

This appears to be a regression from #4602

[kiennq]

I guess the correct fix should be that lsp-completion-passthrough-try-completion should return nil when table is nil.

I agree, this should be a correct fix for the passthough function.

[wyuenho]

Should be good to go now

[kiennq]

Thank you