Escape and complete -C (for fzf completion widget)

[Ambrevar]

fish 2.3.1, Arch Linux, urxvt 9.22

I am trying to write a widget to replace fish completion menu with fzf. (Awesome, right? ;) )

The idea is simple:

## Boilerplate...
complete -C | fzf... | read result
## Some code to replace the token with $result.

Problem is, complete -C unescapes the tokens a bit too much, which makes variables indistinguishable from escaped dollars.

Example:

set HFOO 42
echo \$HOME-$H<binding>

Replace <binding> with a binding to a function that prints out complete -C. Result:

$HOME-$HOME
$HOME-$HFOO

The result is unusable since there is no way for the calling function to know what should be escaped.

The normal completion does preserve the escaped token though:

\$HOME-$HOME
\$HOME-$HFOO

I think complete -C should do the same.

This might be related to #1127.

[faho]

Yes, can reproduce.

We have a bunch of issues with escaping (because it's really confusing), but I can't find this particular one.

[ViViDboarder]

I've been working on something like this with fzf for certain completions.

I put together the following function to use, but I see issues as well.

Not every time, but sometimes it will clear the entire command line (including prompt) from the view. Other times it will call fzf multiple times and try to do multiple completions. So far I haven't figured the issues or why they are only sporadic.

Interested to see if you've had any more luck!

[Ambrevar]

Correct me if I'm wrong, but I think you are referring to a different issue, namely process management and/or file descriptor management. fish has had several issues in that area in the past. You might want to keep looking in the issue tracker.

The issue from the original post related to the badly escaped output of the complete -C command.

[krobelus]

For complete -C'\$HOME-$H', I can imagine two possible behaviors:

1. \$HOME-/home/johndoe

That is, expand the completed token. This may be worthwile, assuming that completions are short-lived.

2. \$HOME-$HOME

I have an implementation of this one at escape-complete-C.
This would break third party scripts using complete -C (mostly used for completing subcommands I guess).

[nzig]

@krobelus Could this be fixed by taking the implementation in your branch, but gating it behind a --unescaped flag as suggested in #4570 (comment) and discussed in #8441? This would avoid the problem of breaking scripts using complete -C, but allow the use case in this issue.

I will open a PR for this.