On 4/8/19 11:33 AM, Stefan Monnier wrote:
> We could also use the last Lisp_Object element as the anchor (taking its
> offset plus its size as the "end offset").

Yes, that's a simple way to fix the problem. I was working on this idea
when I read your email. Also, I noticed that the problem could
potentially happen to any pseudovector, not just to buffers. I installed
the attached into master to fix the problem with other pseudovectors too.

This patch doesn't solve the general portability issue in this area, as
the C standard guarantees only that struct members are aligned and don't
overlap and are in increasing address order. But it should be enough to
fix the immediate problem.