Trying to use dabbrev-capf within capf with company

[garyo]

I'm using company (with vertico etc.) for completion. The default backend is company-capf:

  (setq company-backends '(company-capf company-dabbrev-code
                                        company-dabbrev company-etags
                                        company-keywords))

I'm trying to set things up so by default I get dabbrev completion. I'm trying to set that up in capf like this:

;; Use ~setq-default~ because we need to set the global version, not buffer-local.
;; Otherwise we'd normally use ~add-to-list~.
(setq-default completion-at-point-functions (cons 'dabbrev-capf completion-at-point-functions))

I can see the capf list is OK, but when I go to a markdown mode buffer (which uses the global value of capf) and try to complete a word, I get this error:

Company: An error occurred in auto-begin
Company: backend company-capf error "Symbol’s function definition is void: dabbrev-capf" with args (prefix)

The full traceback is like this:

Debugger entered--Lisp error: (void-function dabbrev-capf)
  dabbrev-capf()
  completion--capf-wrapper(dabbrev-capf optimist)
  company--capf-wrapper(dabbrev-capf optimist)
  company--capf-data-real()
  company--capf-data()
  company-capf(prefix)
  apply(company-capf prefix)
  company-call-backend-raw(prefix)
  company--force-sync(company-call-backend-raw (prefix) company-capf)
  company-call-backend(prefix)
  company--begin-new()
  company--perform()
  company-auto-begin()
  company-manual-begin()
  company-complete()

Am I not setting it up right? Looks like dabbrev-capf doesn't take any args. Does it need to?

BTW the reason I want to add it at the capf level rather than in company-backends is because sometimes LSP modes or eglot modify things so it seems (?) like setting the default value of capf is a good way to ensure it's always used.

[dgutov]

Looks like dabbrev-capf doesn't take any args. Does it need to?

From the backtrace, it just says like dabbrev-capf is not defined in your session. That probably just means that you need to (require 'dabbrev) in your init file somewhere (this function was added in Emacs 29, but it's not autoloaded).

But the completion function itself doesn't seem to be well-written and debugged: I've just tried it, and it shows a backtrace after a backtrace. If somebody has used it successfully (@minad?), they're welcome to chime in.

You might want to try cape-dabbrev as alternative, but according to my measurements (#1411) it's overall slower than the company backends (either of the two).

BTW the reason I want to add it at the capf level rather than in company-backends is because sometimes LSP modes or eglot modify things so it seems (?) like setting the default value of capf is a good way to ensure it's always used.

I recommend instead customizing a variable like eglot-stay-out-of or lsp-completion-provider (setting the latter to :none). Depending on which of the LSP clients you use. And customizing company-backends to the preferred configuration.

[garyo]

That makes sense. I'll go that way, adding to company-backends; that also has the benefit of being able to separate results by backend. (Also: I did try cape-dabbrev but it gave weird/incorrect results.)