Test coverage for scipy.signals.windows needs improvement

[ankit-maverick]

Many of the window functions do not have unit tests : https://github.com/scipy/scipy/blob/master/scipy/signal/tests/test_windows.py

[rgommers]

They're all exercised in a very basic sense by https://github.com/scipy/scipy/blob/master/scipy/signal/tests/test_windows.py#L96, but that test coverage is indeed pretty poor.

[endolith]

Harris page 176 has a table of properties that could be tested against the window functions

For instance, highest sidelobe level of rectangular window would be

20*log10(abs(fft(boxcar(8), 16)[3]/8))
Out[174]: -12.956579111040664

[rgommers]

nice idea! definitions of what's in the table are in the text (a bit hidden in some cases), so this should be doable also for a newcomer -- added the easy-fix label.

[endolith]

I ran the table through OCR and error-checked it:

https://gist.github.com/endolith/c4b8e1e3c630a260424123b4e9d964c4

[endolith]

https://holometer.fnal.gov/GH_FFT.pdf#page=29 has a table of properties too

[larsoner]

Looks like this was taken care of in #6407 but if it's not sufficient I'm happy to reopen

[endolith]

#6407 just adds a few basic tests of length and dtype, it doesn't do the correctness tests that we were talking about