Provide completions for s3cmd

[mqudsi]

No longer auto-generated. Everything has been summarized. Supressing
file completions for initial command, providing list of valid initial
commands, filtering --options by subcommand.

[faho]

You'll usually want to use the "-q" flag for string match.

It's okay here since this is only used as a complete condition, which has its output already silenced.

[faho]

string match -q -- $argv $__tokens[-1] - the echo is unnecessary.

[faho]

Also, this is basically __fish_seen_subcommand_from.

[faho]

This is a bit unorthodox.

Strictly, this would be set -q __tokens[1]; and not set -q __tokens[2]. You could probably get by without the first check, depending on what exactly you want to do with this.

Alternatively, test (count $__tokens) = 1.

[faho]

There is also __fish_is_first_token already. That one skips options.

Personally, for subcommands, I prefer not __fish_seen_subcommand_from $cmds (see e.g. the systemctl completions). That allows for options and option-arguments, as long as they don't look like a command.

For a completely correct solution, you're going to need to do something like the git completions, or use argparse (which I'm converting both the git and systemctl completions to).

[mqudsi]

@faho thanks for reviewing; I knew there would be a significant amount of feedback if I opened this a PR rather than merging directly despite the fact that it works.

I had a hard time finding a good reference on the best practices for writing completions and each completion file I looked at was different than the other (though of course I didn't go through too many). Perhaps we can add a more-indepth article about writing completions to the online resources, as the existing seemed to only cover the very basic usage of complete.

I didn't look at the completions for sysctl, I tried to base mine off of apt-get but found it too hard to follow and so ended up reinventing the wheel :)

It doesn't seem like it should be too hard to take these suggestions into account (I'm going to hold off on argparse at this time, if it's OK with you); let me review and push.

[faho]

I had a hard time finding a good reference on the best practices for writing completions and each completion file I looked at was different than the other

Yes, they are. They were written at different times by different people. Obviously I prefer my own style.

Perhaps we can add a more-indepth article about writing completions to the online resources

That would be nice to have, yes.

I didn't look at the completions for sysctl

Note: systemctl, which is basically systemd's main command. sysctl is that thing for tweaking kernel knobs.

The completions for it are nice (except for the use of eval), and easier to follow than the massive git script.

I'm going to hold off on argparse at this time, if it's OK with you

Sure.

let me review and push.

Yes. You can push directly to Integration_2.7.0 if you want - no need to review again, unless you prefer that. In its current state this is okay, my objections are stylistic.

[mqudsi]

OK, thanks again.