completion-styles for custom dynamic completion capfs?

[mooseyboots]

i'm have a dynamic completion capf for user-mentions autocomplete (we fetch candidates from an REST API).

when doing normal completion with corfu, e.g. programming elisp, i can enable a completion style (substring or orderless) so that a match can be anywhere in a string.

but with my capf, i have noticed that the popup only appears if the prefix matches exactly the first letters of some candidates. is it possible to have the more tolerant substring style in this context too? the matches would be less likely to be correct, but at least we would look like we were trying to provide something, which is better than silently/invisibly failing.

[minad]

It depends on how the Capf is implemented. Your description sounds as if the Capf would only produce candidates starting with the input prefix. See the Cape package for example Capfs which produce candidates dynamically, e.g., cape-dict. There the input is used to produce the candidates but the input can appear anywhere in the candidates.

[minad]

Another reason could be that you have configured (basic substring) as completion styles. Then prefix matches will be preferred. Only if there are no prefix matches, then substring will take over.

[mooseyboots]

i have tried with completion styles first entry being substring and orderless.

[mooseyboots]

Your description sounds as if the Capf would only produce candidates starting with the input prefix.

This is true, but I don't know where this is determined or how to avoid it.

Is your cape--dynamic-table what perhaps makes avoiding this possible?

Its docstring concludes:

"This way the dynamic candidate computation is compatible with non-prefix completion styles like substring' or orderless', which pass the empty string as first argument to the completion table."

I also thought I read about dynamic-completion that completion styles can't be modified for dynamic completion.

my capf maker is here:

https://codeberg.org/martianh/fedi-http/src/dev/fedi-post.el#L231

(you helped me out with it way back)

an instance of it is here: https://codeberg.org/martianh/fj.el/src/dev/fj.el#L4060

[mooseyboots]

sorry, i'm confused now, as above you said "In contrast completion-table-dynamic uses only prefix completion." now you seem to suggest it is possible to change the completion style?

[mooseyboots]

i'll try the debug function out, thanks. i was going to ask how the hell you debug/work with dynamic completion!

[mooseyboots]

debugging seems to confirm that it doesn't work: i confim my category is used, completion-category-overrides sets styles to substring (or orderless). the table returned contains non-prefix results as expected, but then completion-table-dynamic filters them out. it's ok, good to learn the debug stuff, and to have a better idea of what is possible. i can live with it not working.

[mooseyboots]

thanks for the responses/suggestions/

[minad]

Yes, completion-table-dynamic is broken for non-prefix completion. Prefix completion is all it supports. The input string argument passed to the dynamic table function is only provided by prefix completion (the completion styles basic, emacs21, emacs22). For other completion styles the input argument will be empty (and you likely don't generate candidates in that case?). This means that if you have configured your completion styles as (substring basic), or if you have configured a substring (or orderless) override, the filtering will still fall back to basic completion in the end. Then your function will receive the input, but completion-table-dynamic will filter the returned candidates by prefix afterwards. See completion-table-dynamic, which calls complete-with-action, which calls all-completions, where the filtering happens.

[mooseyboots]

i tried setting a category in my capf maker, and then using completion-category-overrides to make it use substring:

((fj-mention (styles substring)))

to no avail (but i'm not sure i know how to add a category...

[minad]

You can specify a :category key in the metadata of the Capf result.

[mooseyboots]

that's what i tried yeah, but it's hard to know if it even exists or not in the eyes of completion-category-overrides.