Re: Bug in completion with curly braces?

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

On Sat, Nov 21, 2020 at 6:58 PM Felipe Contreras
<felipe.contreras@xxxxxxxxx> wrote:
>
> So the options are:
>
> 1. Fail the completion
> 2. Complete something I didn't ask for (but it's close)
> 3. Complete correctly what I asked for
>
> I'm saying if 3 is not feasible, then perhaps 2 would make sense
> rather than 1. It's at least something.

It's up to your completion function to tell it that #2 is OK, and what
to offer in that circumstance.  That's one thing e.g. _alternative is
for.

In this specific example, though, any time an unquoted "{" appears on
the command line at the time completion is invoked, it's being treated
as the start of a brace expansion and therefore removed from the match
comparisons so that completion can try to fill in a comma-separated
list of replacements.

This does appear to be related to a couple of actual bugs, because
"setopt ignorebraces" is supposed to disable this, but does not do so
correctly.  Before I get into that, here's a hack to take advantage of
the brace-completion behavior to get something that works the way I
think you would like:

_foo() {
  local stashes=('stash@{0}' 'stash@{1}');
  if [[ $words[CURRENT] = *\{* ]]
  then compadd -Q -d stashes ${stashes//\{/}
  else compadd -Q -a stashes
  fi
}

This just says "if there's already a brace on the line, pretend this
is a brace expansion but include the braces in the menu display".

As to the bug ... there are actually a few of them, all inter-related.

1)  _comp_options includes "NO_ignorebraces" so isset(IGNOREBRACES) is
never true in zle_tricky.c:get_comp_string
2) Even if that's removed from _comp_options (and set in the caller
context), when the brace is added to $PREFIX it is escaped with a
backslash, so it no longer matches the string on the line and
completion fails
3) When IGNOREBRACES is set, the $debug_indent string in
_complete_debug is mis-handled
4) Possibly as a consequence of that, invoking _complete_debug from
outside GDB with ignorebraces causes the shell to exit (no reported
error, it just exits)

#3 is relatively easy to fix but I won't yet in case someone can track
down #4 (I've had no luck).