Re: Segmentation fault with core dump

Glauco Torres <torres(dot)glauco(at)gmail(dot)com> writes:
> (gdb) bt
> #0 ckpt_buforder_comparator (pa=pa(at)entry=0x7f6fa9ef4b2c,
> pb=pb(at)entry=0x1be06d2d06644)
> at bufmgr.c:4137
> #1 0x0000000000801268 in med3 (a=0x7f6fa9ef4b2c "\177\006",
> b=0x1be06d2d06644 <Address 0x1be06d2d06644 out of bounds>,
> c=0x2fc9dfbb1815c <Address 0x2fc9dfbb1815c out of bounds>, cmp=0x6a4d20
> <ckpt_buforder_comparator>)
> at qsort.c:107
> #2 0x0000000000801621 in pg_qsort (a=0x7f6fa9ef4b2c, a(at)entry=0x7f6fa9ea8380,
> n=<optimized out>, es=es(at)entry=20, cmp=cmp(at)entry=0x6a4d20
> <ckpt_buforder_comparator>) at qsort.c:157
> #3 0x00000000008015e2 in pg_qsort (a=0x7f6fa9ea8380, n=<optimized out>,
> n(at)entry=111473, es=es(at)entry=20, cmp=cmp(at)entry=0x6a4d20
> <ckpt_buforder_comparator>) at qsort.c:203
> #4 0x00000000006a81cf in BufferSync (flags=flags(at)entry=128) at
> bufmgr.c:1863

Hm. I'm not normally one to jump to the conclusion that something is a
compiler bug, but it's hard to explain this stack trace any other way.
The value of "n" passed to the inner invocation of pg_qsort should not
have been more than 29914, but working from either the value of d or the
value of pn leads to the conclusion that it was 0x7f6fa9f3a470, which
looks a lot more like an address in the array than a proper value of n.

I suppose this might be due to a corrupted copy of the postgres executable
rather than an actual compiler bug. Did you build it yourself?

BTW, I notice that ckpt_buforder_comparator assumes it can't possibly
see the same block ID twice in the array, which I think is an
unsupportable assumption. But I cannot see a way that that could lead
to a crash in pg_qsort --- at worst it might cause a little inefficiency.

regards, tom lane