@Flamefire Can you update the scipy easyblock as well for this change?

Done. And with SciPy-bundle-2019.03-foss-2019a.eb I get a failure due to a precision issue -.- Same with

Those are very small failures, e.g. -0.00077271310043404 != -0.000772713100434114 (rdiff 9.569216326469691e-14) or "Arrays are not almost equal to 5 decimals" ... Max absolute difference: 1.5497208e-05

Numpy seems to be fine. Not sure what to do. I'd suggest I'll remove the scipy change so we have at least numpy tested

Let's put it in a separate PR so we can still try to address it

Was that test on POWER? Looks like this is not uncommon:
scipy/scipy#11181

No, this was on Intel and AMD x86 CPUs. Split out the SciPy stuff into #2241

Test report by @Flamefire

Overview of tested easyconfigs (in order)

• SUCCESS SciPy-bundle-2019.03-foss-2019a.eb
• SUCCESS SciPy-bundle-2019.10-foss-2019b-Python-2.7.16.eb
• SUCCESS SciPy-bundle-2019.10-foss-2019b-Python-3.7.4.eb
• SUCCESS Python-3.6.6-foss-2018b.eb
• SUCCESS Python-3.6.4-foss-2018a.eb

Build succeeded for 5 out of 5 (5 easyconfigs in total)
taurusi6257.taurus.hrsk.tu-dresden.de - Linux RHEL 7.8, x86_64, Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz, Python 2.7.5
See https://gist.github.com/0a047b845b27d654b8e0bbe381e00874 for a full test report.

lgtm

I' wondering if there should be a log message about inverting the return value, so if a future version of numpy changes the behaviour (again) and the test_steps fails although all tests pass, at least it would be immediately clear why (?)

I expect the shown error message to be clear enough for that case. I would not clutter the log in anticipation of stuff that may or may not change and may or may not be obvious what changed. I mean: The test command is pretty clear: Run a test function and exit with the inverse of the result. As long as you remember that exit takes an error value and not a success value of course.

Going in, thanks @Flamefire!