Well, the completion file works for me: Zsh should parse anything in this folder, whether it has a compdef or not.
I wonder why you have to source it manually. Can you confirm?

It looks like fzf is not using the conventional completion system but binding on TAB instead, for whatever reason.
I don't know if this is considered bad practice for Zsh.

Hi Pierre!

> Well, the completion file works for me: Zsh should parse anything in this folder, whether it has a compdef or not.

Can you confirm that the file really works for you? Zsh doesn’t automatically *source* stuff from that directory. It’s just meant for autoloaded *functions*, which are either declared via the autoload builtin, or by adding an autoload tag to the first line of the file. In any case, fzf’s completion.zsh does not contain an autoloaded function. You can confirm this via `which fzf-completion`. The way fzf is currently packaged, you should get `fzf-completion not found`.

completion.zsh is a bad choice for the filename, so it’s an easy mistake to make. What completion.zsh does is define a ZLE widget (*not* a completion command) and binds the widget to the TAB key. The widget will be triggered when you end a command line with `**<TAB>`. You can see that the currently packaged version doesn’t enable this by doing `ls /tmp/**<TAB>`. If you just get the expanded filenames, then it’s not working. completion.zsh will instead open fzf. Same thing for `ssh **`, where it should show an fzf dialog listing all your SSH hosts.

completion.zsh is meant to be sourced from the user’s config, in the same way keybindings.zsh is meant to be sourced. It’s just an additional widget and a keybinding, not a command completion for the fzf command.

Best regards,

My bad, you were right. Explanation: the previous maintainers of the package included a patch to automatically parse the completion file when the keybindings are loaded.
I don't like tangling configuration together and I'd prefer to separate keybindings from completion, which would also follow your recommendation.

Does anybody see a reason to keep the patch?

Ah, I see. I didn’t look at the PKGBUILD, this might have alleviated the confusion. The patch is a bit odd. I’d opt for dropping it. «auto-complete.zsh» doesn’t really belong in the site-functions directory anyway. I’d just install it to /usr/share/fzf along with the other keybinding script. Users can then choose to source whichever parts they want (e.g., some people might want to source the default keybindings, but not use the «**-TAB» thing, which is more intrusive than the other bindings).

Agreed. I'll commit your recommendation soon unless somebody comes with a better offer.