Drop Python 3.6 and test with Python 3.9

[SylvainCorlay]

Dropping Python 3.6

(Fixes #1540.)

[SylvainCorlay]

@MSeal what do you think about dropping Python 3.6?

[agoose77]

@SylvainCorlay are we not forced to drop Python 3.6 support given that we currently depend upon traitlets >= 5.0 which drops support for Python 3.0-3.6?

[mriedem]

If this is really dropping support for python 3.6 then shouldn't the python_requires version in the setup.py be updated to be at least 3.7?

nbconvert/setup.py

Line 191 in 9689b2c

python_requires = '>=3.6',

And if that's the case, the changelog should probably be updated as well right?

https://github.com/jupyter/nbconvert/blob/master/docs/source/changelog.rst

[mriedem]

If this is really dropping support for python 3.6 then shouldn't the python_requires version in the setup.py be updated to be at least 3.7?

I just noticed this in the README:

https://github.com/jupyter/nbconvert#python-requirement

So I guess 3.6 is supported but not all that strongly, and will be dropped in 6.1, so it seems this would be fine if it also triggers a 6.1 release.

[agoose77]

@mriedem I don't think we can support 3.6 at all, because we have a hard dependency on traitlets>=5. and traitlets has python_requires>=3.7

However, I don't know what the protocol is when it's a dependency that is enforcing our version constraint. I would assume we drop support for 3.6?

[mriedem]

However, I don't know what the protocol is when it's a dependency that is enforcing our version constraint. I would assume we drop support for 3.6?

Yeah I'd think so as well. I'm definitely not a regular contributor here, I'm just interested in getting #1471 merged since we need that downstream (we're capping nbconvert<6 as a result which is causing some other issues, i.e. jupyter/jupyter_client/issues/637).

CI has been broken for a few weeks and this PR has been up for about 3 weeks now so just trying to figure out how to proceed with getting CI passing again so things can start flowing for non-maintainers.

[mriedem]

Also, I'm personally fine with dropping support for python 3.6, my -1 is just that I don't think we should probably drop CI testing for 3.6 without also officially dropping support for 3.6 which I think at least means:

• bumping the python_requires version in setup.py
• adding a changelog entry
• removing that section of the README about python 3.6
• release a 6.1 version

[meeseeksmachine]

This pull request has been mentioned on Jupyter Community Forum. There might be relevant details there:

https://discourse.jupyter.org/t/current-state-of-nbconvert-drop-support-for-python-3-6/8882/1

[agoose77]

OK, I have somehow made a mistake here - the current released versions of nbconvert use traitlets 4.x, so we're fine in terms of the release install_requires.

The PR that introduced traitlets 5 is #1498

[mriedem]

The PR that introduced traitlets 5 is #1498

Yup and it failed CI but was merged (I can't see settings on the repo but seems failing CI should block a change unless an admin overrides it).

[agoose77]

The PR that introduced traitlets 5 is #1498

Yup and it failed CI but was merged (I can't see settings on the repo but seems failing CI should block a change unless an admin overrides it).

Right, then I think it's just a case of reflecting the implied changes in the various metadata as you've outlined above 🥳

[MSeal]

@SylvainCorlay Sorry I missed the notification on this originally. 3.6 technically has EOL from Python org until 12/21, but if it's desired to drop it given upstream deps dropping 3.6 it's probably fine, as 6.0.x is stable for 3.6 users anyway.

bumping the python_requires version in setup.py
adding a changelog entry
removing that section of the README about python 3.6
release a 6.1 version

We should probably get a couple other hanging PRs merged, e.g. extension for webpdf export fix, before the 6.1 release step. I've been a bit time constrained so haven't been able to put major effort into helping PRs get patched to a mergable state recently.

[meeseeksmachine]

This pull request has been mentioned on Jupyter Community Forum. There might be relevant details there:

https://discourse.jupyter.org/t/nbconvert-6-1-0-release/9731/1