Done: https://hg.savannah.gnu.org/hgweb/octave/rev/13a2062926b6

It took some experimenting but it finally passes with an empty second argument and fails with a non-existent variable as an argument.

Closing as Fixed but it can be reopened if necessary.

This works properly, but it looks complicated. Is there a simpler way?

%!test
%! lasterror ("reset");
%! unwind_protect
%!   hf = figure ("visible", "off");
%!   imagesc (reshape (1:100, 10, 10));
%!   colorbar;
%!   colormap (flip (gray (32)));
%!   colorbar;
%! unwind_protect_cleanup
%!   close (hf);
%! end_unwind_protect
%! assert (isempty (lasterr))

Gives:

octave:1> test colorbar
error: 'd' undefined near line 491, column 21
error: called from
    colorbar>cb_colormap at line 491 column 7
    colormap at line 133 column 10
    __test__ at line 8 column 4
    test at line 682 column 11
***** test
 lasterror ("reset");
 unwind_protect
   hf = figure ("visible", "off");
   imagesc (reshape (1:100, 10, 10));
   colorbar;
   colormap (flip (gray (32)));
   colorbar;
 unwind_protect_cleanup
   close (hf);
 end_unwind_protect
 assert (isempty (lasterr))
!!!!! test failed
assert (isempty (lasterr)) failed

Thanks for the quick fix.

Do you think it would make sense to add a BIST that ensures that this post of the code won't break again in the future? Maybe (a variation of) the motivating example on Discourse?

Calling colorbar and then changing the size of the colormap causes an "undefined variable d" error inside colorbar. See https://octave.discourse.group/t/colorbar-error-from-colormap/4591 for details.

This is to get a bug number for the fix, which I'll push next.