Add support for a `prescient.el` completion style.

[okamsn]

This is still a WIP due to surrounding features, but it seems to work well enough. I'm making the PR now in case people want to point things out.

---

This completion style can optionally modify a completion table's metadata to use prescient.el sorting, including prescient-sort-full-matches-first.

• Add faces prescient-primary-highlight and prescient-secondary-highlight. Add a function to apply them.
• Add user options prescient-completion-highlight-matches and prescient-completion-enable-sort.
• Make prescient-filter handle completion tables.
• Add filtering functions prescient-all-completions and prescient-try-completion, which use prescient-filter.
• Use C function all-completions inside of prescient-filter.
• Move handling prescient-sort-full-matches-first outside of prescient-filter.
• Add sorting functions prescient--completion-modify-sort, prescient-sort-full-matches-first, and prescient-completion-sort.

Logic:

1. Use triggers prescient-all-completions or prescient-try-completion.
2. These use prescient-filter to filter candidates.
3. That function stores the regexps used and whether case is ignored as text properties on all candidates in the post-filtering list of candidates.
4. Highlighting is done back in prescient-all-completions, but not in prescient-try-completion.
5. Emacs adjust the metadata in completion--nth-completion?
6. Sorting is then applied?

Since the built-in flex completion style modifies the sorting metadata to a function using completion input, this approach should be fine.

[okamsn]

Related things to do:

• Based on what is done with Selectrum and Prescient:
  • [x] Clean up prescient-sort-full-matches-first for Selectrum. This was implemented in prescient-filter, which assumes that prescient-sort was run first. This can't be assumed for completion styles and I shouldn't have done that.
  • [x] Clean up faces and highlighting now that Prescient needs its own faces for the completion style.
  • For Company and Prescient:
  • [x] The sort-full-matches-first works when Company-Prescient is disabled, but not when enabled.
    • Now works regardless. Try to not sort twice if we think the completion style forced the sorting.
  • [x] Highlighting in Company doesn't seem to work? Though it doesn't seem to work in Orderless either, so maybe that's OK.
    • From the Orderless README (here):

      The matching portions of candidates aren’t highlighted. That’s because company-capf is hard-coded to look for the completions-common-part face, and it only use one face, company-echo-common to highlight candidates.

    • [x] Try with Corfu.
    • Highlighting works in Corfu.

Questions to answer:

• [x] How does sorting work in Company? Does the Prescient transformer re-sort after whatever the completion table or filtering does?
  • It did, but not anymore (at least for the prescient style). Yes.
• [x] Why is prescient--anchored-quoted-regexp (from orderless--anchored-quoted-regexp) needed?
  • Because some completion tables will only produce candidates that match the input, and Orderless and Prescient try to pass an empty string to get all of the possible candidates. We want a way to communicate that a query should be part of the prefix.
  • [x] Test a filter method that satisfies this? Code removed.
• [x] What can we get/remove if the completion style uses Prescient completion by default? This is for Selectrum Prescient, Company Prescient, etc.
  • Selectrum removes the text properties from candidates, so we still need a custom refinement function. We can't just use selectrum-refine-candidates-using-completions-styles.
  • In Company, the completion style's sorting (and filtering) only applies to CAPFs. We still need the transformer, but we can check whether candidates were sorted by the completion style to avoid sorting them twice.
• [x] Can a list of candidates be re-arranged between filtering and sorting? The completion style currently only bothers propertizing the first filtered candidate to store information for sorting full matches first. Should it just propertize all to be on the safe side?
  • Now propertizing all candidates to support sort-full-matches-first in Company Prescient for Emacs less than 27. Company was re-arranging the propertized candidates before they got to the post-processing filter, since they didn't have an explicit sort order.
• [x] Can we detect when we're completing in region? Would like to have company-prescient-sort-length-enable for CAPFs in general.
  • Instead of this, just don't modify completion metadata by default. This allows for better control in the UIs themselves without needing to worry about redundant sorting.

In general:

• [x] names of things
  • Settling on "prescient-completion-" since "prescient-completion-style-" is a bit long for a prefix.

[okamsn]

@raxod502, do you have any thoughts on these things?

@minad, since this further enables https://github.com/minad/vertico/issues/237, would you please give this a once-over to see if anything is off? I haven't noticed anything yet, but I'm sure that there are cases I haven't considered.

[minad]

I didn't try yet, but I have a few comments.

[raxod502]

This looks wonderful, thank you.

[okamsn]

I think all of the kinks have been worked out now. @minad and @raxod502, is there anything you want addressed/changed before I merge this?

[minad]

Looks good!

[raxod502]

Nope, go ahead!