Even though such configuration is wrong...

extends can be interpreted as inheritance in OOP, but the original
behaviour was against what is commonly (always?) seen everywhere.

It is however good practice a file extends all files it needs directly
(and only them). Then if two files A & B (possibly unrelated) extends
the same third C, A was unable to overrides C values. It was even
error-prone because someone who don't use B yet could override C values
in A and later extending B would break A.

For some of our common use cases, this new algorithm is also 9x faster
(time to annotate: ~2.3s with -> ~.29s).

Other changes:
- ~/ is now expanded for non-url extends.
- An absolute (non-url) path is not longer treated like a local path
  if the base is a url.
- Better path/url normalization.

Rebase instructions:
- squash with "Chomp ../ from beginging of filenames"
- split and apply "Support ${:_profile_base_location_}." after

Sqaushed with

Chomp ../ from beginging of filenames.

In order to have as canonical as possible paths, chomp ../ from filenames and
recalculate base.

--setuptools-version and --buildout-version options in bootstrap script still have the priority.

One big issue in commit e129e187 is that the
initial value of buildout_and_setuptools_path can cause buildout to use a lot
of CPU at startup. The following change would help:

    def buildout_and_setuptools_path():
        r = [get_python_lib(standard_lib=1)]
        for d in pkg_resources.working_set:
            d = d.location
            if d not in r:
                r.append(d)
        return r[1:]
    buildout_and_setuptools_path = buildout_and_setuptools_path()

but initializing buildout_and_setuptools_dists also takes time to compute,
and it's only used for bootstrap.

Because the commit is not required anymore since setuptools 36, it's easier
for us to revert partially, so we can keep applying our patches.

(cherry picked from commit 18964211)

(cherry picked from commit 73b1377c)

PyPI redirects HTTP to HTTPS by default now so using HTTPS directly
avoids the potential for that redirect being modified in flight,
helping prevent MITM attacks.

Fixes #114.

(cherry picked from commit fac9979f)

[ci skip]

Fixes #326

Mainly to disable it when just trying to get distributions for
buildout and setuptools.

That I can no longer reproduce.

NFC

Fixed a bug introduced in 2.6.0:
zc.buildout and its dependeoncies were reported as picked even when
their versions were fixed in a ``versions`` section.  Worse, when the
``update-versions-file`` option was used, the ``versions`` section was
updated needlessly on every run.

[ci skip]

[ci skip]

Added a buildout option, ``abi-tag-eggs``

As suggested in: https://github.com/buildout/buildout/pull/325

Also simplified _setup_directories to not bother with the eggs
directory, since that's handled elsewhere.

that, when true, causes the `ABI tag
<https://www.python.org/dev/peps/pep-0425/#abi-tag>`_ for the buildout
environment to be added to the eggs directory name.

This is useful when switching Python implementations (e.g. CPython
vs PyPI or debug builds vs regular builds), especially when
environment differences aren't reflected in egg names.  It also has
the side benefit of making eggs directories smaller, because eggs
for different Python versions are in different directories.

Python 3.6

Update to work with setuptools 34

whatever

Editing while reading

[ci skip]