Can company configured to fallback to other backends automatically when no matches found?

[adamcavendish]

Can I fallback to dabbrev-code for example, if company says that there's no match found?

[dgutov]

See #142.

The short answer is "no", but there's also a third-party package (mentioned in the discussion linked above) which gives one alternative.

For manual invocation, you can also try the built-in command M-x company-other-backend and bind it to a key.

[dalanicolai]

So can you suggest the best option please?

But let me first mention the following. I am using Spacemacs where, e.g. for emacs-lisp-mode, company-capf is configured as the first backend. Also, company-dabbrev(-code) is configured but company-capf hijacks most or all completions. So now the problem is that symbols that occur already within a buffer do not get suggested by autocompletion because company does not proceed to use the dabbrev backend.

On the other hand, in python-mode, already occurring symbols do get suggested already by company-capf which uses lsp-auto-completion-at-point.

To make things work in emacs-lisp-mode, I can group company-capf with company-dabbrev(-code) to improve the situation, but I am not sure if that is the "correct/recommended" solution.

To make things more consistent (with python-mode), maybe the previous occurring words should also get suggested by company-capf/elisp-completion-at-point in emacs-lisp-mode? So then, should I report an issue for it at company-capf/emacs-bugs?

[yugaego]

I can group company-capf with company-dabbrev(-code) to improve the situation, but I am not sure if that is the "correct/recommended" solution.

That's a fine solution (that can be tuned per exact setup/use case). It had been suggested ("recommended") a number of times, f.i. #484, #1041 (there're more but I didn't quick-find them, so you may look up for more examples).
Moreover, rhetorically speaking, if it works for you, doesn't it make it the "correct" solution (for you)?

I don't have sure/exact answers to the finalizing questions you listed, but they might be given later by (the maintainer) dgutov.

[dalanicolai]

Thanks for the quick answer. I was asking for it because I don't know how company's design is intended to work, and maybe you have some preferred design. But it seems that indeed it is recommended to group company-dabbrev-code with company-capf when the latter does not already provide the suggestions as provided by the former.

So I guess all is fine then, and the later questions have been answered implicitly.
Thanks again!

[dgutov]

To make things more consistent (with python-mode), maybe the previous occurring words should also get suggested by company-capf/elisp-completion-at-point in emacs-lisp-mode? So then, should I report an issue for it at company-capf/emacs-bugs?

Maybe only in certain well-defined contexts? Otherwise the current implementation tries to make "smart" suggestions: offer functions where you need a function, offer variables where you need a variable, and so on.

[dgutov]

Simply mixing in all symbols from all buffers would mess with that goal.