This logic isn't immediately clear to me. If the shape of the matrix is known ahead of construction, does it matter that there are full zero columns? They shouldn't have any information in the sparse matrix structure.

Can you add some testing for this logic?

When I ran it originally I got an error. As I understand it, scipy.sparse.csc_matrix expects indptr to be length n_cols + 1, even when some columns are all zero. In the depmtx files the matrix size includes columns that never appear in the A matrix, so those columns are implicit zeros. Padding indptr just makes the sparse structure consistent with the declared shape and avoids the pointer-size error.

Similar comments from #512

1. Please add this to the Attributes section of the docstring, when it may exist or not, what purpose it serves, etc.
2. Please consider using the full name self.flux

The attribute is now named flx so this will need an update. If we just want to make sure we have something, we can instead use

changed!

I would greatly appreciate a docstring here as to what this test is accomplishing. At a certain point, it feels like we're testing the behavior of the scipy sparse arrays and I'm not sure how much value that adds.
But, the value is we want to make sure the indptr stops advancing for the last two columns. That's captured in the comments, but I would like it emphasized and stated in the docstring

added blurb, let me know if you were thinking of something more detailed

Hmm would using a NamedTemporaryFile be easier? Then we don't need to worry about the file descriptor object nor calls to os.close. Something like

@classmethod
def setUpClass(cls):
    content = """..."""
    cls._tmp = tempfile.NamedTemporaryFile(mode="w")
    cls._tmp.write(content)
    # Set file pointer back to start of file so we read from the top
    cls._tmp.seek(0)
...
@classmethod
def tearDownClass(cls):
    cls._tmp.close()

This may still require the try/except pattern during removal but I'm not sure.

changed. because i'm on windows i have set the delete flag to False (per https://docs.python.org/3/library/tempfile.html ), but that should be okay with linux and macs?

Hmm this is new to me!¹

Since Pytest 2.6, users can prevent pytest from discovering classes that start with Test by setting a boolean __test__ attribute to False.

Looking at the two concrete classes below, could we have the "base" class be SparseOptionalPreambleTester and a subclass for dense? The only difference in the two is handling of the sparsity. Then we have two classes and don't need to modify the __test__ attribute