Consider triggering completion automatically at snippet tabstops

[muirdm]

It would be cool if lsp-mode could automatically trigger completion at each snippet tabstop. For example, when expanding foo(${1:a}, ${2:b}), lsp-mode would trigger completion (with an empty prefix) at $1. If the user chooses a candidate (or starts typing to filter candidates), the "a" placeholder would be overwritten. After finishing completion at $1, the user presses TAB and advances to $2, automatically triggering a new completion.

I considered opening an LSP proposal to add a "COMPLETE" snippet variable so the server explicitly triggers this behavior, but it seems like you almost always want completion for snippet tabstops/placeholders, so I'm not sure we need that. Are there any common cases where a user wouldn't want completion triggered at tabstops?

[yyoncho]

Are there any common cases where a user wouldn't want completion triggered at tabstops?

1. If there is only one candidate it will be automatically inserted. It may cause an endless loop as well.
2. If there are a lot of candidates on top level and the user has configured the autocompletion to be displayed manually and/or with bigger prefix it will lead to unwanted behaviour.

I tested it with changing the behaviour of TAB command and I saw 1.

I did some quick research and I did not saw enough hooks in yasnippet to implement that. IMO this could go as an optional feature(e. g. FAQ entry) and it should go in lsp-mode core after yasnippet exposes the hooks.

[muirdm]

Thanks for the help!

this could go as an optional feature

Yeah, definitely should be optional. Perhaps certain known-to-be-performant servers could enable it by default if it proves to be a popular option.

If there is only one candidate it will be automatically inserted.

I was imagining the candidate wouldn't be inserted - it would behave like you triggered completion and there was only one candidate, i.e. the candidate shows up "pending" at your cursor, but you still need to press TAB to insert it. In other words, you press TAB once to select the function call foo() which gives you foo(someCandidate<a>, *b*), then you press TAB again to insert someCandidate yielding foo(someCandidate, *b*), then you press TAB again to advance to next tabstop which triggers completion at b: foo(someCandidate, otherCandidate<b>), etc.

not saw enough hooks in yasnippet to implement

I figured it might be hard to implement. I know yasnippet allows embedding elisp code in the snippet - maybe we could jam something in there? Otherwise, do you have an idea of what we need in particular from yasnippet?

[yyoncho]

I was imagining the candidate wouldn't be inserted

I guess this can be avoided but this is the behaviour of M-x company-complete when there is only one candidate on my setup.

Otherwise, do you have an idea of what we need in particular from yasnippet?

We need a hook on field expanded hook(or whatever the yasnippet term is). Then (add-hook 'yasnippet-on-field-expanded-hook 'company-complete). The feature isn't even lsp-mode related, it is integration between yasnippet and company-mode.

[muirdm]

I tried hacking this in but I couldn't get it to work for both the initial snippet field and later snippet fields. I think it had to do with the initial completion not actually being done when we enter the first snippet field, so triggering a completion there didn't work. Were you able to get it working for both the first snippet field and later fields? If so, can you share your approach?

[yyoncho]

This is what I was testing.

(advice-add 'yas-next-field-or-maybe-expand :after (lambda (&rest _args)
                                                     (company-complete)))

[muirdm]

Thanks. That works for the snippet fields after the first, but I can't figure out how to make it start completion for the first snippet field.

When I try something like:

;; XXX assumes the snippet has a tabstop
(advice-add 'yas-expand-snippet :after (lambda (&rest _args) (company-complete-common)))

to trigger completion for the first field, I get:

Company: An error occurred in post-command
Company: frontend company-pseudo-tooltip-unless-just-one-frontend error "Company: backend nil error "Symbol’s function definition is void: nil" with args (annotation int())" on command post-command

I think it is related to the initial completion not being finished when the second completion is triggered (?).

[yyoncho]

You may try to do it with (run-on-timer 0 ...) instead of calling directly company.

[muirdm]

You may try to do it with (run-on-timer 0 ...)

I tried that before. It avoids the error but the company menu doesn't actually show up until I do something to trigger redisplay (like scrolling). I tried calling force-window-update and redisplay but I didn't know how to get the company UI to display automatically.

[muirdm]

I was able to get this functionality working with the below advice. It works for the first snippet field, subsequent snippet fields, and doesn't trigger completion at the final $0 snippet field or after finishing nested snippets. It uses the private yas--field-modified-p, but without that it is hard to tell if it is the first time in a snippet field or not leading to unwanted triggering of completion.

(defun complete-if-yas-field (&rest _)
  (let ((field (yas-current-field)))
    (when (and field
               (not (yas--field-modified-p field)))
      (company-manual-begin))))

(advice-add 'company-complete-selection :after 'complete-if-yas-field)
(advice-add 'yas-next-field :after 'complete-if-yas-field)

[leungbk]

Apparently something like this is specified in the protocol:

Placeholders can have choices as values. The syntax is a comma separated enumeration of values, enclosed with the pipe-character, for example ${1|one,two,three|}. When the snippet is inserted and the placeholder selected, choices will prompt the user to pick one of the values.

I'm not sure if this should be an upstream patch for yasnippet (eglot would benefit from this as well) or if we should simply implement it here in lsp-mode.

[yyoncho]

It is already supported by yas but the syntax is different, here it is java method template:

# -*- mode: snippet; require-final-newline: nil -*-
# name: method
# key: m
# --
${1:$$(yas-choose-value '("public" "private" "protected" ""))} ${2:void} ${3:name}(${4:args}) {
            $0
}

[muirdm]

The "choice" snippet placeholder is limited and I don't think it is feasible to have language servers pre-calculate completions to stick in the snippets in general, not to mention nested completions/snippets. It is much simpler and more general to have the client trigger a normal completion at snippet tabstops since that will work with any snippet from any server without changes.

I've been using my advice above and it does exactly what I want with a few minor issues.