Don't follow libraries which won't be copied.

[HexDecimal]

Combines lib_flit_func and copy_flit_func behavior into one function.
This makes sure that a file which won't be copied will not have its dependencies analyzed.

Fixes #119

[HexDecimal]

In theory this can be back-ported to a Python 2.7 branch. Since attempting to modify system libraries is a rather critical error.

[matthew-brett]

Sorry to be slow to get to this one. Can you think of a test for this behavior?

[HexDecimal]

Sorry to be slow to get to this one. Can you think of a test for this behavior?

Not really, because this simplifies the behavior of how libraries are selected, so there's less to test than before. test_fix_plat_dylibs was modified to remove an assert because the functionality which allowed that assert to pass is what caused #119 in the first place.

[matthew-brett]

A suggestion - sorry to be slow.

And - for the test - how about a test where we make lib_filt_func filter out a library with a dependency, and show that neither the library, nor the dependency, gets copied?

[HexDecimal]

And - for the test - how about a test where we make lib_filt_func filter out a library with a dependency, and show that neither the library, nor the dependency, gets copied?

I've now added this test.

[matthew-brett]

Thanks for your patience.

[HexDecimal]

I'm still thinking about #119, that's a pretty bad error to have in one of the releases. I thought it might a good idea to make another Python 2.7 release but that might take more effort than we should be giving to Python 2.7.

@matthew-brett, I'd like you to consider yanking delocate==0.9 from PyPI. Unless you have a better reason not to.

[matthew-brett]

I think we've either got to let the bug stand for Python 2.7 (and leave the 0.9 release), or backport the fix to Python 2.7...

[HexDecimal]

If I make a python2 branch and backport this PR will you deploy it?

[matthew-brett]

Sure - or happy for you to; can give you PyPI permission.

[matthew-brett]

I've added a coverage check in the Travis-CI.org matrix, but it would be better to have one for the Github workflows, as Travis-CI is getting a bit hard to work with (.org shutting down very soon, credit crunch on .com).

https://staging.travis-ci.org/github/matthew-brett/delocate/builds/770146

[HexDecimal]

My PyPI account is https://pypi.org/user/HexDecimal/. I'll upload the backport and then yank the problematic version afterwards if I have access.

You have some encrypted data in your TravisCI scripts, and it's generally insecure to share that information with your testing tools. I'd recommend not adding any more external tools to the TravisCI script.

[matthew-brett]

I've invited you via PyPI. I've removed the encrypted stuff in the Travis-CI file; I wasn't using it, and the relevant service is long defunct. But you're right, it's best not to extend it at this stage.

[HexDecimal]

I should probably have been a PyPI maintainer, not an owner. Giving owner permissions is a lot of trust for this type of project.

[matthew-brett]

Happy to downgrade you but a) I trust you! and b) you have become the primary author of the package over the last few months. You now know the code-base a lot better than I do.

[HexDecimal]

Fine, I'll begrudgingly accept co-ownership.

[matthew-brett]

Totally happy to degrudge down to Maintainer if you prefer...