There are a few testing-related things going on with this PR. First, with Travis CI:
Change language from c to python to take advantage of built-in features for that language; there is no advantage of using c (also fortran is not a supported language)
Use FC environment variable to control gfortran version and symbolic link identified by pymake
Replace miniconda with xenial's Python 3.7
And with the Python testing:
Replace assert np.allclose with np.testing.assert_allclose, which provides a nice concise output of differences (if necessary). This required specifying the rtol=1e-05 in a few places for some tests to behave the same as np.allclose.
Simplify UZT_NonEq/insert_stopflow_period.py to use conventional Python features. It also now works with Python 2.7 (if that's needed)
There are a few testing-related things going on with this PR. First, with Travis CI:
c
topython
to take advantage of built-in features for that language; there is no advantage of usingc
(alsofortran
is not a supported language)apt: true
cacheFC
environment variable to control gfortran version and symbolic link identified by pymakeAnd with the Python testing:
assert np.allclose
withnp.testing.assert_allclose
, which provides a nice concise output of differences (if necessary). This required specifying thertol=1e-05
in a few places for some tests to behave the same asnp.allclose
.UZT_NonEq/insert_stopflow_period.py
to use conventional Python features. It also now works with Python 2.7 (if that's needed)