Re: [Libmesh-devel] mesh_transfer directory

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

On Fri, 25 Jan 2013, Kirk, Benjamin (JSC-EG311) wrote:

> On Jan 25, 2013, at 4:21 PM, "Roy Stogner" <roy...@ic...> wrote:
>
>> On Fri, 25 Jan 2013, Derek Gaston wrote:
>>
>> Oh, as long as I'm chatting to -devel: I've got a pretty massive
>> commit coming up soon; if anyone else also has current files in flux
>> I'd suggest committing ASAP so that I have to deal with the merge
>> rather than you.
>
> I'm good. Sounds ominous though - something for 0.9.0?

Ominous only in the sense that the old assert()->libmesh_assert()
change was ominous: it touches a ton of lines.  I'm finishing up the
implementation of that processor_id_type and dof_id_type stuff, so
soon we won't have to be incompatible with >32K processor or >2G dof
systems.

None of that requires any API-incompatible changes for anybody who
doesn't configure with a non-default id size.

I actually was making one API-incompatible change while I was at it,
though - passing a vector<dof_id_type> (default unsigned int) rather
than a vector<int> to SparseMatrix::zero_rows.  The original design
appears to have been motivated by "Petsc uses int, and static_cast is
unpleasant", which are both true but not really good enough reason to
have an API which uses different index signed-ness from the whole rest
of the library.

I can revert that change if people prefer.

As far as 0.9.0 goes, the only hold up from my POV is that Vikram and
I are still finding and fixing bugs in the SolutionHistory stuff for
transient adjoint solves.  We've given up on getting the whole API
nailed down before 0.9.0, but we'd like to make sure the stuff that's
in there is working correctly.
---
Roy