Re: Crash with tests

[Paul Eggert]

On 04/14/2015 09:37 AM, Eli Zaretskii wrote:
> > (gdb) p/x p
> > >$1 = 0x4000000001013ffb
> > >
> > >And obviously, memory is not readable at this location.
> Yes.  So does it mean the vector-like object from which this came is
> corrupted?
>
> > >When you say you don't see it: is it using 32bits or 64bits emacs ?
> 32-bit, the only one I have.
I'm not seeing the problem in a 32-bit build either, or in a 64-bit.  
And '0x4000000001013ffb' is not a 32-bit value, so I'm puzzled by the 
above GDB output.

Fabrice, what does the GDB command 'p sizeof p' say?  On my 32-bit 
build, it says 4.

The symptoms you report are consistent with being at commit 
b80c5ebc4f6afd54597012583c6a1390db0ade9c dated Mon Apr 13 23:26:13 2015 
-0700, which has an (in-hindsight) obvious bug that can trash 
save_window_state like that.  This bug should be fixed by the next 
commit 7744cc7e95bcae1df6911be5a9a941d73ea207f7 dated about an hour 
later.  Perhaps you got unlucky and did a 'git pull' during the bad 
hour?  But in that case I don't understand the "commit
b5919771aea2d43c64be381b7d7b395eeb1191bd" in your first email, as that's 
dated after both of the other two commits.

Maybe you are in an inconsistent build state?  That would explain both 
the wrong size and the commit hash discrepancy.  Can you reproduce the 
problem in a fresh checkout and build from scratch?