Re: need help debugging cvs completion problem

[Greg Klanderman <gak@klanderman.net>]

>>>>> Peter Stephenson <p.w.stephenson@ntlworld.com> writes:

> That sounds plausible---I think you're on the right lines.  You might
> need a -U argument to compadd in that case, too.

> Basically, you are (anybody is) on your (their) own in this type of
> thing---lots of us have been somewhere similar once or twice and it was
> so complicated we've forgotten and won't remember again without devoting
> hours to remembering why we thought it was better to forget in the first
> place.

Hi Peter, sorry I had to put this aside for a bit, and am still not
quite ready to dig back into it.  Seems like to fix this right, I'm
headed down a path where I have to reinvent much of the cruft that's
already in _path_files, which is probably why the logic had been
trying to leverage that, albeit in a somewhat hacky way.  My next step
is to understand what was actually broken that you fixed in 2003, and
if there's not some way to go back to using _path_files.

Unfortunately, it seems that when the new completion system was
designed, it did not abstract stuff like handling parameter
references, quoting, etc like the compctl stuff did.  So if you're
doing file completion, _path_files has all that logic (at the expense
of being completely incomprehensible) and it mostly works, but other
sorts of completion either have to reimplement that logic (sometimes
partially and/or buggily), or just not handle these complications.

cheers,
Greg