Update table and add information about staggered migrations

[abravalheri]

After setuptools added deprecation warnings for pkg_resources.declare_namespace there has been a lot of doubt in the community about how to migrate away from it, and concerns about carrying out this migration in large projects (see pypa/setuptools#3434).

The objective of this PR is to shed some light on the problem and help to answer the question "what happens if not all the packages sharing a namespace migrate at the same time?".

I took the opportunity to "modernize" the nox matrix, removing the deprecated installation method (using python setup.py ...) and adding some newer Python versions¹.

Please note that I did not add all active versions of Python to the matrix on purpose, because that would take too much time in my machine. So I am intentionally stepping away from this objective (if any one is interested in adding the missing Python version that would be a nice next step)².

Footnotes

1. The motivation for this is twofold: we are a few moths away from the EoL of Python 3.7 and leaving Python 2.7 and python setup.py ... in the table would be distracting. ↩

2. Another equally nice improvement would be using CI, but I am also intentionally removing this from the scope of this PR. ↩

[chrysle]

Thanks for the work! But is it now staggered or staged migrations? If it's not of importance, I'd prefer the latter.

[abravalheri]

Hi @chrysle, I am not a native speaker, so I am just replicating the words that were proposed by Jason in pypa/setuptools#3434.

[chrysle]

Me neither ;-)

Referring to that, the term staged migration came up in zopefoundation/meta#194 (comment). Since that sounds much more "planned", could you update the PR to it?

Moreover, I think Python 3.7 could be dropped now.

[chrysle]

@abravalheri Please let me know what concerns you have concerning the naming.

[abravalheri]

Hi @chrysle sorry I let this one slip...

No I am fine with whichever words people think will make it clearer.
(If any native speaker would like to chip in, that would also be great :P)

[chrysle]

Well, so I asked OALD, and to stagger seems to be a valid transitive (although that wasn't known to me), and the meaning of to stage I meant (of which I know it's possible) isn't even mentioned on the corresponding page....

So let's take the former ;) Could you rebase?

[abravalheri]

Hi @chrysle, I solved the merge conflicts and also took the opportunity to ensure all tests were using build-isolation and backend APIs (aka PEP 517+PEP 660) - by adding pyproject.toml everywhere.

Hopefully the CI will pass 🤞

[abravalheri]

Hopefully the CI will pass 🤞

Oh well, too soon.
I think that the CI might be failing in some rows of the column, e.g.:

nox > * session_cross_pkg_resources_pkgutil-3.8(command_b=('pip', 'install', '.'), command_a=('pip', 'install', '-e', '.')): failed

That is kind of expected isn't? Some combinations will fail indeed.
Not sure why the CI is expecting everything to pass...

[chrysle]

Thank you!

That is kind of expected isn't? Some combinations will fail indeed. Not sure why the CI is expecting everything to pass...

I actually made the CI pass in #24 (could you have a look at that?). I don't think there will be any more merge conflicts when that is merged. And @pradyunsg enabled that passing checks are required.

[abravalheri]

I think #24 is going to be a big conflict with this one, isn't it?

The original idea behind running nox right now is to generate an informative table that will include situations where combining thing will pass and other where it will not pass... Making all the sessions in nox pass kind of defeat this purpose isn't it?

[chrysle]

Sorry if I screwed this up, I didn't find any documentation here that would explain this purpose. On the other hand, if you are able to run all scenarios correctly, isn't that more helpful to you?

[abravalheri]

Oh no, I didn't mean you screw things up, sorry if I miscommunicated that.

I think that adding more scenarios is good. But we should not remove scenarios that don't work (showing that they don't work is kind of the point).
The assumption that CI should have all the nox sessions passing is not something that this repository was originally designed for.

[chrysle]

Looks good!

Should we migrate all the examples to pyproject.toml later on (probably keeping setup.pys as samples)?

[chrysle]

@abravalheri Gentle nudge to this, sorry for approving too early.

[abravalheri]

Hi @chrysle, sorry for the delay. I am away from my computer this December and less responsive than usual. I will be able to have a look as soon as I come back in January. But I think it might make sense to separate into 2 PRs.

[chrysle]

No problem, take your time and enjoy your holiday!

But I think it might make sense to separate into 2 PRs.

That's my suggestion too.

[abravalheri]

I rebased the PR and updated the table with the latest output of nox.

[chrysle]

Thanks a lot!