Re: [Gnu-arch-users] tla bash completion

[Jason McCarty]

Tom Lord wrote:
> Auto-generating completion should not depend on the "option language
> thingy" -- it should depend on the output formats of `tla help', and
> the outputs of various commands when given the `-h' option.
> 
> So I encourage you to try parsing those outputs, and if there are
> glitches, let's fix them.

Very well, I'll review the help outputs that I thought made reliable
auto-generation difficult when I have some time. IIRC one of those was
that some of the arguments are named based on what they're used for
(FROM, TARGET, LIMIT), while others are named based on what they are
(REVISION, DIR). I was hoping that the option syntax language would
specify the types of those arguments, which would be one less task to do
manually.

Currently the -H option usually tells you the argument types in prose,
which obviously isn't very parsable. A format that displays the
"used-for" names in the short output and the "has-type" names in the
long output might be a good compromise, even for human readers. And the
"has-type" declarations could probably even be written by the syntax
language interpreter. Or maintained manually while that doesn't exist.

Then some of the arguments have special cases: cat-log can accept an
unqualified patch-level for example. But there are few enough of these
that doing them manually isn't too much work.

>     > I agree; -v, -vv, etc might serve as alternatives.
> 
> -vv would violate the conventions for option syntax.  --vv would be
> better, and --very-verbose would be better still.  (If it's needed at
> all.)

OK, that sounds good. I think most commands would just need two or fewer
verbosity options (increase/decrease), so --quiet could be its opposite.

Jason