Re: PATCH: new utility function for mixing in extra values

[Oliver Kiddle <opk@xxxxxxx>]

On 16 Jun, dana wrote:
> two specific issues with menu select when there are only phony matches:
>
> - if you e.g. do `shutdown <TAB>` and then back-space, it doesn't
>   dismiss the menu as it should, the description and phony match remain
>   on the screen like they do with non-menu-select

I've not been able to reproduce this. How are you even getting menu
selection with only a single menu option? The one unusual thing that
_phony does is set compstate[insert]='' if it added phony matches but no
real matches were added and the command-line is empty. For me, this
means that, shutdown <tab> will print the description and a match list
(consisting of just "now"). A second <tab> is needed to insert the
match, or you need to enter at least "n" before pressing <tab>.

> - the behaviour where the phony match isn't inserted for an empty word
>   doesn't really work with menu select imo -- it's fine if it doesn't
>   immediately insert it but it's incredibly confusing that there's an
>   apparent item in the list that can't be selected *at all*

With your setup, what does $compstate[insert] contain before it is
emptied? Perhaps a further condition can be added to detect a setup
similar to yours and act differently.

> for example, with `_phony none _files` (like `ssh -F`), having 'none'
> unceremoniously mixed into the list of files is weird. i don't think i'd
> even notice it most of the time, tbh, and if i did it would mislead me
> into believing that such a file actually exists

It can depend on the context and very often it is better to use
_alternative as you suggest. But often that seems like a lot of effort
and people writing completions have simply not bothered. As an
interface, the wrapper makes them very easy to add.

_phony could do the equivalent of _alternative. The existing
separate-sections style seems applicable and $_comp_tags could be used
to supply a tag when looking it up. But we only have one description
provided in the interface, and often that same description is equally
applicable to both real and phony matches.

My preference is to find a way to apply special formatting to the phony
matches. So something like the list-colors style. I'm open to ideas on
how that might be best be configured.

Oliver