TST: fix sparse constructor test with large memory footprint

[dschult]

Fix test in test_construct that broke from #18862 because of change in private function.

This wasn't seen by that PR because it doesn't run on a 32-bit machine.
Thanks @WarrenWeckesser !!

[WarrenWeckesser]

Shoudn't it be True to maintain the existing behavior of the test?

Also, as long as we're touching this code, let's make it nicer...

Suggested change

	C = construct._compressed_sparse_stack((A,B), 0, False)
	C = construct._compressed_sparse_stack((A, B), axis=0,
	return_spmatrix=True)

(assuming you agree that it should be True).

[dschult]

In #18862 we changed the test to use csr_array so I think this should be False here.
And -- yes, we should make it prettier.
I've been aiming for 88 chars or less on a line (usually less). is there a tradition in scipy for what to aim for?

[WarrenWeckesser]

SciPy's policy is still PEP 8, which says 79 characters max. There has been no explicit policy change that I know of, so that is what we should shoot for.

The degree to which our CI checks this has changed over the years, and we are now using fewer automated checks than we used to. Some time ago, ruff replaced pycodestyle. ruff does not check a lot of basic whitespace issues--ruff expects that black is used for that, but we have not adopted black, so our automated style checking ended up weaker than it used to be. ruff does check line length, but its default max is 88 characters. Allowing lines that long seems be an accident of the change to ruff, and not an explicit decision by SciPy devs to increase the recommended line length. I'm sure a lot of devs are OK with it, and I'm trying to be less pedantic 😃, but I still think we should shoot for 79 characters, and only exceed it in exceptional cases (and this does not look like an exceptional case; an extra line break is no big deal).

[WarrenWeckesser]

In #18862 we changed the test to use csr_array so I think this should be False here.

OK, I'll take your word for it.

[tylerjereddy]

Amazingly, when I run python dev.py test -v -j 32 -- -k "test_concatenate_int32_overflow" -rsx on current main, I get:

SKIPPED [1] scipy/_lib/tests/test_array_api.py:12: Array API test; set environment variable SCIPY_ARRAY_API=1 to run it. I'll shift that over to gh-19190.

Using the other incantation over there, I do indeed get the test passing here and failing on main.

[tylerjereddy]

I checked the test at two points in time re the C object type change:

• on maintenance/1.11.x:

type(C):
 <class 'scipy.sparse._csr.csr_matrix'>
C.shape:
 (66000, 33000)
C.getnnz():
 2178000000

• on the PR branch:

type(C):
 <class 'scipy.sparse._csr.csr_array'>
C.shape:
 (66000, 33000)
C.nnz:
 2178000000

The change to csr_array is described as intentional above, from the other listed PR that was reviewed by CJ/sparse folks, so I think we're good to go with this test repair now. I'll probably squash-merge with the two small linting commits.