LGTM.

Applied to master (c646114, 122904f). Thanks!

(I'll backport to release-17.09 later.)

Commit f79d691 ("nix-bash-completions: lazy load aware install") hasn't yet been backported, but is a dependency of this PR. Is it correct to backport it?

@bjornfor it should work, but will source the script many times on startup. So might be bad for performance, might be better to wait.

Ah, misread it a bit. I'm guessing we can't backport this when f79d691 isn't packported too (which is the one that would cause bash startup time to increase). I'm guessing the best thing is to wait?

Wait for what? (I won't backport anything unless someone says it's OK.)

As in wait for #32534 to hit stable (either through backporting or 18.03). Did some timing and backporting f79d691 will increase bash startup time with 0.1s+ in the best case scenario (hot cache), which is far from ideal.

So..if we backport #32534, f79d6991, and this, things will be good?

(I think you're saying grabbing the commit without #32534 would be rather bad for startup performance, right?)

Is there a reason to wait to do this?

(Needs more testing, breaking change?)

So..if we backport #32534, f79d6991, and this, things will be good?

There's #33704 where the completion script have a name which isn't picked up by the lazy loading (relevant PR: #33717). Doing a quick search of of nixpkgs the package memo have the same problem.

(I think you're saying grabbing the commit without #32534 would be rather bad for startup performance, right?)

Yeah.

Is there a reason to wait to do this?
(Needs more testing, breaking change?)

Apart from the issue with taskwarrior and memo not that I know of, though there might be other packages which use a non-standard name for the completion script which escaped my quick search. So to not break anything we would need to backport fixes for taskwarrior and memo too.

Also the above stuff only applies to bash of course, so there shouldn't be any problems backporting the zsh completions.