bug#71224: 30.0.50; SIGSEGV in start_display

[Daniel Clemente <n142857@gmail.com>]

> I could reproduce it again. I don't know what's involved, but this is what I did when it crashed.
> Run M-x debug
> Go to other window, and eval (recurse) there
> Open new frame
> Close it, and close the first frame too (close all frames)
> I got to the point when I couldn't open a new emacsclient frame. The server complained (Lisp nesting exceeds
> ‘max-lisp-eval-depth’: 1633) and the new emacsclient waited (black window) but couldn't start and finished.
> After trying to open still another one, it crashed.
>
> Yes, w->desired_matrix is null.
>

I have tried to reproduce this crash in a newer Emacs
(d719a8140917e3b1dd6889f064ffb8e8b0621c7d, from yesterday) and I can't
anymore. When I do the the "Open new frame" step, emacsclient refuses
to open:

$ emacsclient -nw
*ERROR*: Lisp nesting exceeds ‘max-lisp-eval-depth’: 1622

If I close all frames, I can't open any frame at all, and the only
valuable next step is to kill the Emacs daemon. It makes sense, since
I caused the infinite recursion. It could be improved, but that's
another issue.

I suggest closing this bug, and I'll open another one if I find
another case in which w->desired_matrix is null. This may still happen
now (since we didn't locate or fix the bug) but the situations that
cause it are contrived and apparently unreachable now.
One of the crashes in this bug report was already fixed by Eli.