On Nov 29, 2010, at 4:23 PM, Jed Brown wrote:

> On Mon, Nov 29, 2010 at 23:21, Barry Smith <bsmith at mcs.anl.gov> wrote:
> Crap. That means that the output from this newly modifed function is not what 
> is needed by the creation routine. In fact there exists 
> DMDASetOwnershipRanges(). What are we going to do with that? Change the 
> meaning of DMDASetOwnershipRanges() arguments (and hence also the meaning of 
> the final optional arguments to the DMDACreate3d() and 2d)?  For uniformaty 
> we  need to change those also. But the changed form is more cumbersome and 
> less natural for users, is it not? Or is it ok to change all of them? We 
> don't have a concept of setting global ownership values for Vec and Mat 
> (cause the meaning is a little different). Maybe a name change is in order?
> 
> That was my original thought, either change the meaning of all these things 
> or change the name.

   Then why didn't you say that? Your original email didn't make clear that 
lots of PETSc code needed to be changed (to perhaps a more cumbersome model). 
When I thought basically it was only user code that needed much change I was 
perfectly happy to make them change it, but now that it involves PETSc code I'm 
not so sure :-)

   BUT given my hatred of new concepts and the fact that it basically fits the 
VecGetOwnershipRanges() concept I still think the correct fix is to change all 
the PETSc code but leave the same names. Never introduce a new concept to save 
changing code, only introduce a new concept when it is really needed and here 
it is not really needed!

   Barry

> 
> Jed


Reply via email to