bug#23013: 25.0.92; assertion failure on marker outside text range in echo area

[Eli Zaretskii]

> From: Ken Raeburn <raeburn@permabit.com>
> Date: Fri, 22 Apr 2016 17:01:37 -0400
> 
> No more than half an hour after I sat here wondering how many days
> without a crash might suggest that maybe the change I reverted was
> actually related to the crash, Emacs crashed; the same way as before, an
> assertion failure on a marker position in " *Echo Area 0*", while
> switching between buffers; the last stuff I typed was
> 
>   C-x b e @ M-backspace return
> 
> (I.e., switch buffer, start typing a partial name, notice the default is
> what I want so delete what I typed, hit return.)
> 
> The w->pointm marker has charpos and bytepos set to 41 but the buffer
> text field has gpt, z, gpt_byte, z_byte, all set to 1.
> 
> I've been running commit a1f221b, with d82f24b reverted, configured
> with:
>   '--prefix=/permabit/user/raeburn/dev/emacs/emacs-25/lx/Inst-5725167'
>   '--with-x-toolkit=lucid' '--enable-checking' 'CPPFLAGS=-DMARKER_DEBUG'
>   '--without-sound'
> 
> The marker debugging code hasn't flagged anything.

MARKER_DEBUG is for hunting a different kind of problems, those with
converting byte positions to character positions and vice versa.  So
it's little wonder it didn't flag anything.

Needless to say, this never happened to me, although I use echo-area
buffers as much as anyone else.  Not sure what this tells us.

I think at this point, the only way to find this elusive bug is to
trace updates to markers when the buffer text of the echo-area buffers
is shrunk.  It is also possible that the problem happens because we
switch the echo-area buffers, but do not update the markers when we
do.  I would suggest to add some tracing code, perhaps check_markers
or similar, to the places that are involved in the above events, and
see what that tells us.

Thanks.