Re: MPS: Win64 testers?

[Kien Nguyen]

> Did this happen randomly, or is it reproducible?

It happens randomly for me.

> What I suspect is that we failed to trace an object, probably a face
> (but maybe it was an ordinary vector), during the previous GC.  I don't
> see how that could have happened, yet...

> Were you using non-ASCII faces at the time?

Yes, I'm using nerd-icons which is full of non-ASCII characters.

> Slow because of the hook or because of other reasons?

It doesn't seem to be because of hooks as I can still type, a tad slower
than normal.  On the other hands, whenever a compiled package is loaded,
it's really slow.  I do delay loading a lot so using Emacs under debugger
has not been enjoyable so far.

> You should realize that attaching to Emacs when it crashes many times
> misses important information, because when Emacs is not running under
> the debugger, fatal exceptions go to the top-level exception handler,
> which at times loses important information, like exactly which code
> caused the crash.  So running an unstable Emacs build outside of GDB
> is not recommended on Windows for several good reasons.

The mps build is quite stable for me and only crash once in a while so
...

> In any case, you should be able to invoke remove_w32_kbdhook from the
> debugger, as soon as you attach to Emacs.  I hesitate to make the
> change you proposed because emacs_abort is called not just when Emacs
> is under a debugger, so the change will affect many users who don't
> debug Emacs.

Got it, I will try that next time.  It will need some setups as at the
time the debugger is attached, typing is really slow so it's impossible
to type anything long.
Btw, I think that at the time emacs_abort (for Windows) is called, the
process is going to crashed anyway unless the debugger is attached and
do some hot patches.  The function is either abort or wait for a
debugger to attach anyway.  Is there another way to recover and continue
to use Emacs?