bug#9087: Crash reading from minibuffer with icomplete-mode

[Juanma Barranquero <lekktu@gmail.com>]

On Mon, Jul 18, 2011 at 18:45, Eli Zaretskii <eliz@gnu.org> wrote:

> It wasn't.  This is the crash about which Martin said that it's with a
> window that has no children and no buffer.  So it looks like a
> different crash.

I cannot dismiss the possibility of having found two different bugs at
once, but as I repeat the same steps (allowing for the variation that
the bug does not always happen at the exact same step), it seems like
a long shot.

> Anyway, could you propose an exact recipe to reproduce this?

I can not promise that it will be exact, but the one I give works for
me every time.

>  I tried
> using the one you gave originally, but I don't seem to have the fonts
> you used, or maybe I'm not following the recipe correctly.  A step by
> step recipe using "normal" fonts would be appreciated.

OK, though I'm not sure whether the bug is affected by the contents of
the completion somehow (which surely depends on the fonts installed).

emacs -Q -f icomplete-mode
M-x set-face-font <RET> font-lock-comment-face <RET>
-ou<TAB>Cou<TAB>bo<TAB>no<TAB>iso1<TAB><RET>
M-x set-face-font <RET> M-p <RET> M-p    ; now you have back the font spec
M-f M-f M-f  ; to move just after "Courier New"
M-backspace M-backspace M-backspace  ; to delete "outline-Courier New"

then try to complete with ou<TAB>, then M-backspace to delete
"outline", and repeate a few times. Most of the time it crashes after
a few tries. You can also complete outline and then try with another
font family, but I suggest using one that does not have bold variant,
so the completion does not match.

I think the specific font, and wheter it is bold, normal, or whatever
does not really matter. The only thing that seems to matter is doing
non-matching completions.

Hope you're able to reproduce it.

    Juanma