[petsc-dev] MatGetLocalSubMatrix, MatLocalReference implementation

[Jed Brown]

Your function is returning objects with global semantics (they live on the
global comm). The other local functions return objects that are semantically
serial, although they may be logically part of something bigger. But why
return an IS at all, why not an ISLocalToGlobalMapping? Too many characters?

I need a name for a local IS that holds indices in a local numbering. This
thing has no parallel semantics, just like a local Vec. The term "local IS"
is not currently taken, if we don't count you man page. Would it be
acceptable to use it for my purpose?

Jed

On Nov 20, 2010 4:15 AM, "Barry Smith" <bsmith at mcs.anl.gov> wrote:


On Nov 19, 2010, at 6:43 PM, Jed Brown wrote:

>
> Jed
>
> [1] Barry, what is going on with DMCom...
  Looks like code that was started (maybe from cut and paste) and never
finished. The manual page for function has a different name then the
function, like I just stopped in the middle of editing and forgot about it.

> The former has no implementation and the
> latter has no man page. I do not think that "Local IS"...
  Well no. Local always refers to things with ghost points, while global
means without ghost points. You are asking to introduce another term
"Ghosted" to mean with ghost points.  Confusing/bad to introduce more
terminology.

  Now you could argue that using local and global was bad originally and we
should use ghosted and global (or nonghosted) as our two terms. But
introducing ghosted in just one place is not the solution, we would have to
do it systematically through PETSc source and documentation. Though I think
local and global are fine as is and don't see a need to change from local to
ghosted.

 Or I could be misunderstanding what you want to change.

  Barry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101120/9af9a1fc/attachment.html>