fish vs carapace in 2024

[ccoVeille]

Request

I'm a fish user, and I'm new to caparace

I followed the step describe in https://carapace-sh.github.io/carapace-bin/setup.html

I was surprised to see

carapace --list | awk '{print $1}' | xargs -I{} touch ~/.config/fish/completions/{}.fish # disable auto-loaded completions (#185)

see

• fish: erased functions still get autoloaded #185

I can see the reference here

carapace-bin/docs/src/setup.md

Line 32 in 1b1ecb1

carapace --list | awk '{print $1}' | xargs -I{} touch ~/.config/fish/completions/{}.fish # disable auto-loaded completions (#185)

I found the code was moved via 6ad36ab and was initially added via 79e988e (via #880)

All these are pretty old, the #185 was mentioning

• Completions not erasing with complete -c <command> -e fish-shell/fish-shell#6716

Is there any progress around this ?

I saw the status of fish-shell#6716 but I was curious if there is nothing else that evolved in the last few years. I find pretty painful to create empty file it's a strange logic

I found this cli/cli#360 (comment)

Proposed solution

I would have expected carapace to:

• create the empty file via the go, not via an shell rule

• the empty file could have a better content than simply being empty

they could contain comment about why the file is there and empty

• use fish "complete -e" can't help ?

• maybe the fish file could contain the complete command instead of being empty

for example

~/.config/fish/completions/git.fish could contain this

complete -e 'git' 
complete -c 'git' -f -a '(_carapace_lazy git)'

Anything else?

I might be wrong about the things I'm asking or the point I'm raising, but I would like to know if we couldn't do something better for the fish integration

Thanks

Polar

[rsteube]

Well it is only a workaround and I expected fish to fix this fairly quickly (little did I know).

The problem is that this is a bug in fish and from the outside this was the only way to get around it.
Erasing existing completion doesn't work, because fish loads the system completion on first invocation (pressing tab).
It does so even if one is already defined which clutters it.

Shell startup time is critical which limits other potential workarounds.

Generating the files in go is possible, but it is quite an invasive step as it essentially kills the native completion of fish which might be unwarranted when one disables a completer from carapace.

To be honest, this just needs to be fixed in fish with the tombstone approach mentioned here.

[ccoVeille]

Thanks for your reply. I understand.

Do you think I could quote your reply in the involved fish discussions, and ask them for a status?

[rsteube]

Sure

[ccoVeille]

A fix was provided. Let now wait for it to be released

• khipp/fish-shell@5918bca

[azlekov]

Hello carapace is adding like 3 seconds to my fish prompt startup and I think is because of these completions. Is there any fix for this?

[rsteube]

Hello carapace is adding like 3 seconds to my fish prompt startup and I think is because of these completions. Is there any fix for this?

Even with the workaround it shouldn't be this bad 🤔.
The files in ~/.config/fish/completions just need to be created once for each new completer.
So the workaround doesn't need to be executed every time.

The complete command in fish also tends to be quite slow (which is a bit weird),
if you want the full startup speed you could also export the scripts instead of sourcing carapace.

carapace git fish > ~/.config/fish/completions/git.fish

But you'd need to update PATH and add the set-env/get-env functions manually.

[ccoVeille]

The point raised by @azlekov is legitimate, and the workrounds you provided could provide a solution.

But @rsteube said a few months ago fish needed to change something in a way to use something called "tombstone"

But the issue is now fixed

fish-shell/fish-shell@5918bca

Shouldn't the solution be now updated as fish should support it.

Please note, I'm aware there is no new fish release since March 2024

[rsteube]

Well, depends on how easy it is to install a nightly (and how stable the master is).
But I can add a notice to the instructions, yes.

Might slowly be time to have another look at this.
Invocation delay of carapace is in the range of 7-10ms, which I personally don't consider an issue.
But the slow performance of the complete command and the amount of completers registered due to the bridges makes the whole topic relevant again.

It's possible to have a near zero overhead for most shells, but it involves a slightly different approach (see above).