Will do. I might go into work tomorrow and try it out (even though
I'm supposed to be off...)
Sent from my iPhone
On Sep 6, 2009, at 6:20 PM, Roy Stogner <roystgnr@...>
> On Sat, 5 Sep 2009, Derek Gaston wrote:
>> Well... I think there is a deeper issue here than just having
>> _closed not getting set to true.
> Yes, there definitely is. The failing is_closed() assert is not your
> real problem (it's asserting something which is true in essence, even
> if the bool is technically false); that assert is just masking
> whatever the real problem is.
>> Like I mentioned, I was investigating because I am getting somewhat
>> random segfaults. I ended up backing up to a revision just before
>> ghosted vectors were made the default (3444 I believe)... and that
>> completed fixed it.
> Don't bet on that being completely fixed. So far the ghosted vectors
> are batting 1.000 on "revealing bugs in other code" vs "having bugs
> themselves". In particular, it's possible that an incomplete
> send_list is leading to a situation where a ghosted vector would (in
> opt mode) segfault but a serial vector would just (perhaps very
> slightly!) corrupt your result. I would strongly recommend you run in
> dbg mode with ghosted vectors on now that that assert is fixed, and
> see if you still get to a segfault or if you get a failed assert in a
> ghost coefficient lookup instead.