Use separator character for boundary quit logic and prefix sort

[jdtsmith]

Implement a component separator character and associated insertion command (corfu-insert-separator-char), used to inhibit quitting at completion boundaries (see #119). This character is also removed from prefix sorting, superseding #118.

This is useful for multi-component completion styles such as orderless. A convenient key binding for corfu-insert-separator-char such as M-SPC can be used to insert the initial completion component separator.

Obsoletes corfu-quit-at-boundary (for v0.19). Setup documentation is provided.

[minad]

Very good, I will try this in the next few days. So if we bind SPC to corfu-insert-separator we get the current quit-at-boundary behavior? If we bind M-SPC instead we get the enhanced behavior? I guess we could bind M-SPC by default.

[minad]

I gave it a quick review. This looks very good, nice and simple. There are only a few minor details. Then there is the question if we want to support disabling the separator character completely - by setting it to ?\0 or nil.

[jdtsmith]

So if we bind SPC to corfu-insert-separator we get the current quit-at-boundary behavior?

I suppose that's right. Or, if you (or the user) binds nothing at all to corfu-insert-separator-char, you will simply fall through to the predicate boundary check. So the default could become the less intrusive option.

if we want to support disabling the separator character completely

I tried to think of a real usage case for disabling the behavior of the separator. Unless you invoke M-SPC, the separator is benign. The only thing I could come up with:

1. You do not use orderless or a multi-component completion style.
2. Your mode's CAPF happily includes the separator char within the completion boundary (I suppose more likely if it's not space).
3. You want other non-word chars to quit completion (e.q. punctuation like , or =).

I don't know if 2 is occurring anywhere in the wild for the default space separator. But allowing this to be nil is perhaps easy insurance for any cases we are missing. I'd still recommend keeping space as the default though.

Also, regarding the binding, should we bind it ourselves or let the user pick? I've also tried S-SPACE which works well.

[minad]

I just want the ability to disable this out of principle, but I agree that having a separator like space (also having it enabled by default) is benign

[minad]

Also, regarding the binding, should we bind it ourselves or let the user pick? I've also tried S-SPACE which works well.

We should add the binding M-SPC by default. It is easy to unbind this in a user configuration. We also preconfigure separator=SPC.

[minad]

This looks good, thanks!