Fix vector block data for single function space case #3924
Conversation
schnellerhase
changed the title
| @@ -123,7 +123,10 @@ def create_vector( | |||
| index_map, bs = maps[0] | |||
| ghosts = index_map.ghosts.astype(PETSc.IntType) # type: ignore[attr-defined] | |||
| size = (index_map.size_local * bs, index_map.size_global * bs) | |||
| return PETSc.Vec().createGhost(ghosts, size=size, bsize=bs, comm=index_map.comm) # type: ignore | |||
| b = PETSc.Vec().createGhost(ghosts, size=size, bsize=bs, comm=index_map.comm) # type: ignore | |||
| if kind == PETSc.Vec.Type.MPI: | |||
There was a problem hiding this comment.
What if kind == None?
There was a problem hiding this comment.
For the previous behaviour that would have been equivalent to the call create_vector(L) which did not attach block data, therefore we don't want to attach data.
There was a problem hiding this comment.
If creating vector for list of a single space, i.e. create_vector([V]) is almost the same as creating a vector on a single space create_vector(V), except that the former attaches block data, then why don't we remove the whole concept of single space create_vector(V) and always require a list of spaces, be it of length one. That would simplify meaning of kind, in my opinion, so either mpi or nest. @garth-wells ?
There was a problem hiding this comment.
I know I added the 'block data' concept but I can't recall the details ;). Does it introduce any memory overhead for the single space case?
In general, I'm in favour of as few special cases as possible, and has few interface types and possible.
There was a problem hiding this comment.
How about merging this, which recovers behaviour as before (and fixes down stream usage) and having a look into removing the difference in another PR?
With the new interface, the
create_vector([V], kind="mpi")case no longer holds block data. This adds the block data, allowing to treat a single function space vector as a block vector when passingtype="mpi". Caught downstream.