On Mon, 23 Jun 2008, Lorenzo Botti wrote:

> it seems that ES already has a call to update() in the end of read().
> This should avoid the user to call  System::update().

Yes; as it turned out, my problem was that the rhs vector wasn't being
properly closed during a restart.  Doing an explicit close in my
NewtonSolver before use was a sufficient workaround, but I'd like to
put in a real fix.

> But if you are trying to restart a transient system in parallel the problem
> is the lack of an update() method for such a system.
> I posted a message on the list about this problem some time ago.

Sorry I let these slip by me.

>> I have finally find out why I couldn't restart properly a transient
>> system in parallel. The problem is that transient system doens't have
>> an update(...) method.
>> So I have added one coping it from the re_update(...) method. It works
>> for me but it will be good if someone could check it and add it to the
>> svn.

I'll want to hear from Ben or John before doing this; there's got to
be some reason for the update vs. re_update distinction in the first
place.

>> Another problem is that we should avoid the reading of the RHS_Vector
>> becuase as I said in a previous message Petsc doesn't allowed to set
>> the value of a Petsc_vector an then zero it without calling close().
>> This now is happening in ImplicitSystem::init(..). I don't have a good
>> solution for this... of course we can call close() before zero() but
>> this is an horrible hack.

What makes this a horrible hack?  It's always possible that user codes
are going to want to overload rhs for some reason, and as long as
we're saving it anyway I'd rather let it be read.  The overhead of an
extra close() call doesn't seem to be significant compared to the
overhead of reading from disk in the first place.
---
Roy

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://sourceforge.net/services/buy/index.php
_______________________________________________
Libmesh-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to