Re: PATCH: minimal dropbox command line completion (and $compstate[nmatches]/$ret)

[Daniel Shahaf <d.s@daniel.shahaf.name>]

Oliver Kiddle wrote on Sat, 29 Aug 2020 13:45 +0200:
> Daniel Shahaf wrote:
> > I guess merging [1] and extending it would be the best option
> > technically.  However, that file appears to be under the GPLv3 (per the
> > LICENSE file in its repository).  Is that a problem?  
> 
> I think we should avoid doing that if at all possible. Where we've had
> functions under different licences it can create more work for people
> packaging zsh than it would be to just rewrite the function. Different
> distributions and operating systems have different rules, some remove
> them, some add complicated merged licence files and many ignore it.
> I should probably revisit that idea of having a separate Contrib
> directory for some functions. It was partly stalled on the naming.
> 
> To me it would seem bad form to simply take something found elsewhere
> without successfully contacting the author. If you get an answer,
> chances are they'll be more than happy to contribute it under the zsh
> licence.

To be clear, I wasn't proposing we just copy code from github to
zsh.git without contacting authors.  I was analyzing the problem from
different angles (technical, legal, social), each one separately, and
then combined the answers.

> When handling subcommands, it is always a good idea to fallback on
> calling _default wherever you have a subcommand for which there is no
> special treatment. Don't break file completion probably the most
> important rule of writing custom completions.

Add this to the documentation somewhere?

> > Aside: I wonder why we don't use a «_call_completion_function()
> > { readonly -i n=$compstate[nmatches]; "$@" && (( $compstate[nmatches] > $n)) }»
> > wrapper around calls to $_comps[foo]: that'd save every single
> > completion function having to manage $ret explicitly.)  
> 
> The original concept was that there would be cases where you explicitly
> want to allow completion to continue despite having added matches.
> In practice this is very rare but does occur. Also, we rely on the
> return status in many more places than just the calls of $_comps[foo] in
> _dispatch. Getting it wrong tends to be most apparent from _approximate
> acting when it shouldn't but it can break other things such as tag
> loops.
> 
> It could be possible to redesign this if we allow breaking changes
> sometime. Perhaps with keys in $compstate to indicate continuation.

Thanks for the explanation!

Cheers,

Daniel