small sparse test fixes #30874
small sparse test fixes #30874
Conversation
ViralBShah
added
sparse
| rowval = [1,2,3] | ||
| nzval1 = Int[] | ||
| nzval2 = [1,1,1] | ||
| A = SparseMatrixCSC(n, n, colptr, rowval, nzval1) |
There was a problem hiding this comment.
Actually, this test seems bogus altogether (it's explicitly constructing a bad matrix). I missed the point initially, but I would just eliminate this test (and eventually add throws on construction forcing consistency between buffers).
There was a problem hiding this comment.
I just read through the discussion in #20464 to refresh my memory. The symbolic factorizations in SuiteSparse don't reference the nzval buffer so this might be related to that. I don't remember if any of SuiteSparse wrappers currently make use of this feature. We should check that. We might have allowed an empty nzval such that people could do symbolic factorizations without the non-zeros but that might not be a good idea to allow such invalid instances of SparseMatrixCSC. Expert users should be able to work directly with CHOLMOD.Sparse for that.
|
The changes are all good. @abraunst Do you want to do something about that check buffers test? I would be ok to merge this as is. |
No, it's fine. We can change/remove the test when/if we decide for more strict enforcing of buffer lengths. Thanks! |
This fixes in the
sparse.jltest fileall(...)SparseMatrixCSC