When running the command install_lib, the command build_clib is not run.
If a package contains clibs they can be dependencies for extension libs,
at least that is stated in the module docstring of build_clib.

In real life you can see it when trying to run the command install_lib
on the current reportlab2.3.

The patch provided needs do be applied in the distutils dir directly with

patch install_lib.py dependency.patch

As requested, I asked on the mailing list first, whether this really is
a bug.
http://mail.python.org/pipermail/distutils-sig/2009-February/010974.html

I am half finished writing a test case, that I'll append to this ticket
later.

be careful to write your test in distutils (setuptools is not part of
Python)

The attached test works breaks without the patch and passes with the patch.

I am not sure whether the test itself is written following best practice.
I would be more than happy for a review, with hints what should be done
better.

The patch looks good.

Minor tweaks :

• sys.path should be set to its initial state after the test
  (setUp/tearDown might be helpfull here)

• ensure_finalized() should be called right before run()
  so it acts like the real call

Thank you, tarek

The sys.path should have been obvious.

It looks like there's a missing part in the patch: get_outputs(), and
get_inputs() should also return files created in build()

Now something I don't understand is how reportlab distribution is doing
to copy the compiled extensions when "install" is called.

Could you point me please to the package so I can test that ?

Thanks !

Hi Tarek,

I think clib stuff is installed in the right python directory and the
ext install step then just finds them.

To reproduce the issue, run
the "old" easy_install in a virtualenv.
easy_install ReportLab
It will then fail because of the missing library.

I can also show it during Plone Conference, just look for me in irc.

Thanks,

  Patrick

Adding back distutils1. This is a bug, not a feature.

Has the patch been applied on distutils(1/2) ?

If we haven’t closed it, it’s because it has not been.

> Now something I don't understand is how reportlab distribution is doing
> to copy the compiled extensions when "install" is called.
I think clib stuff is installed in the right python directory and the
ext install step then just finds them.
Yep, the install* commands just copy all files from the build dir to the install dir.

The patch is straightforward and the test looks good. However, I worry about backward compatibility. For projects without C libs, this won’t change anything; for projects already having a workaround, what will happen? While this is clearly a bug, I’m not sure fixing the existing behavior is okay in distutils.

For packaging (distutils2), I will commit this shortly.

for projects already having a workaround, what will happen?
I think I have the answer to my question: if build_clib is called twice, the registry in Distribution will know that it has run and won’t run it again. I’ll test manually and if I confirm, I will commit this change.

This issue has been open for some time, looking at the install_lib in master, there have been no changes to call 'build_clib' so the issue documented here would still apply.

Distutils is now deprecated (see PEP-632) and all tagged issues are being closed. From now until removal, only release blocking issues will be considered for distutils.

If this issue does not relate to distutils, please remove the component and reopen it. If you believe it still requires a fix, most likely the issue should be re-reported at https://github.com/pypa/setuptools