bash-completion: use sqlite cache when available (BZ: 1815895)

[rjarry]

Use /var/cache/dnf/packages.db to make SQL requests instead of parsing the repodata files at each tab completion. This makes completing package names much snappier.

This is an inspiration from zsh dnf completion: https://github.com/zsh-users/zsh/blob/zsh-5.8.1/Completion/Redhat/Command/_dnf#L7-L33

= changelog =
msg: Use sqlite cache to make bash completion snappier
type: enhancement
resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1815895

[rjarry]

Gentle bump :)

[kontura]

I like the change in general.
We have a plugin in dnf-plugins-core that creates and updates the packages.db cache but as far as I can tell we are not using it anywhere so it is kind of wasted.

It provides less correct completions though.
For example when running update it completes all installed packages but not all of then can have updates available.
Also when running with for example with --disable/enablerepo, --repo, --repofrompath or --installroot the result can be misleading. Maybe we could at least add a check for those options and not use the cache in that case?

Also also we already have a PR with similar approach: #1100

[rjarry]

I can send a changed version that bypasses the completion cache when --disable/enablerepo, --repo, --repofrompath or --installroot are used. I can do the same thing for update if you want.

[rjarry]

I had not seen #1100 which is indeed the exact same thing.

[rjarry]

I did not find a way to detect if --disable/enablerepo, --repo, --repofrompath or --installroot are used. I am not sure if the current implementation takes these options into consideration when completing package names. I could be wrong though.

I have bypassed the completion cache for update and its related commands. However, from a user experience point of view, I wonder if having less accurate completion may be better than having the terminal freeze for several seconds at each tab completion. In the worst case, users may try to update a package which has no update candidates. To see which updates are available, they can always use list updates.

I am waiting for your feedback before proceeding. Thanks.

[kontura]

I did not find a way to detect if --disable/enablerepo, --repo, --repofrompath or --installroot are used. I am not sure if the current implementation takes these options into consideration when completing package names. I could be wrong though.

I see yes you are correct. It doesn't take them into account either so doing the fallback wouldn't help much.

I have bypassed the completion cache for update and its related commands. However, from a user experience point of view, I wonder if having less accurate completion may be better than having the terminal freeze for several seconds at each tab completion. In the worst case, users may try to update a package which has no update candidates. To see which updates are available, they can always use list updates.

I also prefer faster feedback.

Ok, thanks for the contributions!