@rouille agreed, this could be much more time-efficient by combining several of these tests into one function (only defining ref_grid once, copying as necessary, with several individual asserts), or in a unittest.TestCase class with a setUp function. I PR'ed more of an MVP so that this branch could be used right away to aid in #182. I'll clean it up further before merging.

I refactored the changes to use one copy of the Texas grid and one copy of the Western grid, via @pytest.fixture(), calling copy.deepcopy as necessary. This cut the total test runtime in half, from ~35 seconds to ~17 seconds. For reference, the tests currently in develop take ~14 seconds, so the pytest fixture cut down the added time by a factor of ~6.

For posterity's sake: building these tests using the unittest.TestCase framework, with a setUp method that pre-built a Western grid and a Texas grid, was no faster to run in pytest than the original version where tests shared no data. I suspect that pytest was running the setUp before each test, rather than running it just once, hence the use of @pytest.fixture().