Hello Filipp,

Filipp Gunbin <fgunbin@fastmail.fm> writes:

>> [...]
>> This reads as if t ("do nothing") vs. 'move ("move to the limit of
>> search") should make a difference? A few quick experiments seem to
>> indicate that in practice both seem to behave the same though. In this
>> light, I'm fine with changing 'move to t.
>
> Well honestly I was fooled by save-excursion, and didn't notice the
> return value of (point), with which the value we're talking about _may_
> matter.  However, your regexp will always match, due to \\|^ branch
> which will find bol, so you can just do:
>
> (save-excursion
>   (re-search-backward "\\([:,]\\|^\\)[ \t]*")
>   (match-end 0))

I tried without the surrounding save-excursion, and even then I didn't
notice any difference in behaviour between t and 'move for NOERROR.

Your suggested simplification works just as well, of course. Thanks for
pointing out! Updated patch below.

>> [...]
>> That said, it would probably be desirable for message mode to have
>> different values for both, completion-at-point-functions and
>> completion-styles, depending on where point is (email header, newsgroup
>> header, message body, etc.). But this seems like a wider discussion
>> about the architecture of message.el rather than this patch.
>
> Yes, sounds like another use case for completion-category-overrides.

Hadn't come across that one yet; thanks for the pointer.

Message.el does this:

(add-to-list 'completion-category-defaults '(email (styles substring)))

It seems to me you suggest amending that line in message.el, rather than
having the setq-local in my capf function? I have included this in the
below patch, too.


Many thanks and looking forward to your thoughts,

  --alexander