python-packaging: honor add_context.include for non-stdlib extensions

[dae]

The setting was previously indiscriminately ignored, meaning it wasn't
possible to prevent third-party extension modules from being included
in the build.

Related: #405

[indygreg]

Thanks for tracking this down.

I agree a partial solution here is better than none.

I'm going to rebase this and amend it slightly before pushing. I'll likely also tack on some tests in a follow-up commit because this behavior warrants testing!

[indygreg]

As I was implementing tests, I found a few issues with how extension modules are handled. I'd like to get the tests shored up before landing this so I can be confident of what the behavior is. I may not get this landed tonight. But it is on my short list.

[indygreg]

OK. So I've changed the API for the resource collector to return objects describing what actions were taken. I did this because as part of implementing tests for this, I encountered some weird behavior and found the existing logging around resource actions... lacking. So the new API enables us to print more useful log messages about what's going on.

This unfortunately introduced a merge conflict. The return Ok(None); in your PR should be changed to

return Ok((
    vec![AddResourceAction::NoInclude(extension_module.description())],
    None,
));

However, when I did this locally, a bunch of tests unrelated to the new ones started failing! I suspect the issue here has something to do with the original comment mentioning built-in extension modules and the new code only filtering on stdlib. The built-in set and the stdlib set aren't exactly the same and I think that is causing tests to fail.

This is a thorny situation. Having looked at this code, the tests I wrote, and the new log output (which appears to show duplicate extension adds), something is clearly off somewhere. I'll try to take a look at this again this week.

[dae]

No rush - presumably it could be worked around by users for now by installing into a venv and pruning the unwanted packages before running pyoxidizer.