Improve Fish's completions

[kidonng]

This PR brings two improvements to Fish's completions, when completing {{cmd}}:

• Let Fish only complete directories and excludes files

• If there is no match in the current directory, Fish will query zoxide with current token and serve them as completions:

  

  This is approximately the same as using zoxide query -i <current token>, but more integrated with Fish.

Let me know if you have any suggestion over these two changes.

[ajeetdsouza]

Thanks! Two small issues I had when trying this:

• If I type in z foo bar and hit Tab, I only get results for bar (foo is ignored).
• If I type in z foo bar and use the autocomplete, it will complete only bar, leaving me with z foo /home/bar. Since this contains multiple arguments, it doesn't cd correctly.

[kidonng]

• If I type in z foo bar and hit Tab, I only get results for bar (foo is ignored).

Only the token under the cursor is searched, see reasons below.

• If I type in z foo bar and use the autocomplete, it will complete only bar, leaving me with z foo /home/bar. Since this contains multiple arguments, it doesn't cd correctly.

That's just how tab completion works. You can't expect it to change anything else than the current token. It's also why it doesn't take any other token into consideration, which just won't make sense.

If people do want to search with multiple tokens, they should use zoxide query -i instead.

[ajeetdsouza]

I understand that, but IMO completions should never result in a command that is invalid, which is what's happening here.

Btw, merging this PR might take a while - I want to try a couple of different solutions to see what works best.

[kidonng]

but IMO completions should never result in a command that is invalid

You can easily reproduce the same issue with regular cd, and I don't think anyone have an issue with that.

To achieve what you expect, every completion needs to implement a full-fledged parser, which is obviously not practical.

Technically we can prevent providing completion if the user is completing a second argument, for this issue only though. But IMO that's just unnecessary.

[ajeetdsouza]

@kidonng if you have time, do take a look at #257. I've only implemented Bash completions at the moment, but I'd love to know what you think:

• Completions only work when there is one word (therefore completions are always valid).
• By default, it uses the same completions as cd.
• If the query ends with **, it opens up fzf and does an interactive search (similar to fzf's builtin completions).

If it works well, we can hopefully write completions for other shells in line with this.

[ajeetdsouza]

Hey @kidonng, I've updated the PR. This is what the new completion function does:

• It completes directories by default.
• If there is more than one argument, or if the last argument ends in **, it makes an interactive query via fzf.
• Early exit from interactive query removes the ** if present.

I'd appreciate any thoughts on this.