Re: [Libmesh-devel] Ghosted solution vector not closed in update()

[Roy S. <roy...@ic...>]

On Fri, 10 Jun 2011, Derek Gaston wrote:

> I'm still doing testing on System::update() with the large systems
> that are giving us problems.  I haven't been able to isolate the
> problem yet.... but I'm working on it.

Thanks!

Any chance I could get a look at the (tail of the) stack trace to
help?  I'm anxious to at least know whether the segfault hits us in
localize() itself, _restore_array, a PETSc function, or elsewhere in
the code.

> I am thinking that the issue might actually be elsewhere but it's
> causing the failure in System::update().... I'm somewhat hopeful
> that after I track it down we might be able to revert
> System::update() to the previous algorithm.

> Sorry it's taking so long.... just hard to debug these big problems.

Understandably so.

And for what it's worth, I think I've massaged the code into more
tolerable shape now; we've gone from "ugly workaround that will
probably break on other localize() uses and that definitely breaks on
PETSc+noMPI or Trilinos" to plain "ugly workaround".  Still definitely
worth investigating and probably worth reverting when we can, but now
that it's no longer turning a bunch of regression tests red I'm no
longer as anxious about the timeframe.
---
Roy