Updating GitHub actions

[hoxbro]

I have been working on updating the test GitHub Actions and I have come to a point where I want to share it. It is not necessary that this PR is to be merged, but the changes I have made could be added to a future PR.

Overview over the changes I have done

• Only accept Python 3.6, 3.7, and 3.8. Maybe this is a bit harsh, but as I understand the 1.14 release will be the last to support 2.7.
• Added conda-forge to channels. This is maybe not needed anymore.
• Loosen some pins, which does not seem to be needed anymore.
• Updated matplotlib/datetime related unittest to be compatible with matplotlib 3.3, the changes is because matplotlib have changed their date2num from a custom format to UNIX time, see example code below. Maybe matplotlib should be pinned for this?
• Changed values of 2 others related to matplotlib update. I would like to know if this is ok?
• I added mamba to replace conda install but removed it again. I think it could be a good option to add to pyct/pyctdev.

Example code of change to matplotlib

# located at https://matplotlib.org/3.3.0/_modules/matplotlib/dates.html#_from_ordinalf
# changed the functions a bit to be more compact.

SEC_PER_DAY = 60 * 60 * 24

def new_dt64_to_ordinalf(d):
    # From matplotlib version 3.3.0
    dseconds = d.astype('datetime64[s]')
    extra = (d - dseconds).astype('timedelta64[ns]')
    t0 = np.datetime64("1970-01-01T00:00:00", 's')
    dt = (dseconds - t0).astype(np.float64)
    dt += extra.astype(np.float64) / 1.0e9
    dt = dt / SEC_PER_DAY
    return dt

def old_dt64_to_ordinalf(d):
    # From matplotlib version 3.0.3
    extra = d - d.astype('datetime64[s]').astype(d.dtype)
    extra = extra.astype('timedelta64[ns]')
    t0 = np.datetime64('0001-01-01T00:00:00').astype('datetime64[s]')
    dt = (d.astype('datetime64[s]') - t0).astype(np.float64)
    dt += extra.astype(np.float64) / 1.0e9
    dt = dt / SEC_PER_DAY + 1.0
    return dt

x = np.datetime64('2016-04-04', 'ns')
old_dt64_to_ordinalf(x), new_dt64_to_ordinalf(x), old_dt64_to_ordinalf(x) - new_dt64_to_ordinalf(x)
# (736058.0, 16895.0, 719163.0)

[jlstevens]

Thanks for the PR!

Only accept Python 3.6, 3.7, and 3.8. Maybe this is a bit harsh, but as I understand the 1.14 release will be the last to support 2.7.

That is correct though we expect at least one more release in the 1.14 series.

Added conda-forge to channels. This is maybe not needed anymore.

Generally we avoid using conda-forge unless necessary as including it slows down solve times.

Thanks for the matplotlib 3.3 compatibility updates!

[philippjfr]

Thanks @hoxbro! I had deliberately not updated the matplotlib tests so I could keep them passing on py2 and py3 builds. Having dropped the py2 builds updating them is the correct thing to do.

[philippjfr]

This needs rebasing but I'm quite happy to now make the master branch drop Python 2 support and backport commits to a 1.14.x branch if necessary.

[hoxbro]

I have made a rebase of the PR.

If this PR is being merged I need to do some small clean ups and if needed update the code to work with matplotlib > 3.4.

[hoxbro]

I have no idea why the unit test was skipped on Pytest on 3.8, windows-latest, maybe just a coincidence?

[hoxbro]

This PR should be ready for review. Some comment/notes since the initial PR:

• I have updated the code to be compatible with matplotlib 3.4. This should fix Matplotlib 3.4.1 compatibility #4886.
• I have also been wondering if the testing should be done only for python 3.7 and python 3.8, what are you thoughts on that?
• For reason unknown python 3.8 on windows skips the unit test without failure.

[jlstevens]

I've not done a detailed review yet but at a glance this PR looks good!

I have also been wondering if the testing should be done only for python 3.7 and python 3.8, what are you thoughts on that?

I think we should keep 3.6 tests at least (3.5 I am happy to ignore as that was a problematic release).

[philippjfr]

I think we should keep 3.6 tests at least (3.5 I am happy to ignore as that was a problematic release).

I think if we're dropping 2.7 we should also drop 3.5, it's been 5+ years and most libraries I know have dropped support for it at the same time as 2.7.

[philippjfr]

Will merge. Thanks for your efforts @hoxbro!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.