|
|
|
|
|
|
|
Importer Wall Of Shame
|
|
|
|
----------------------
|
|
|
|
|
|
|
|
The following modules and packages violate protocol or best practice in some way:
|
|
|
|
|
|
|
|
* They run magic during ``__init.py__`` that makes life hard for Mitogen.
|
|
|
|
Executing code during module import is always bad, and Mitogen is a concrete
|
|
|
|
benchmark for why it's bad.
|
|
|
|
|
|
|
|
* They install crap in :py:data:`sys.modules` that completely ignore or
|
|
|
|
partially implement the protocols laid out in PEP-302.
|
|
|
|
|
|
|
|
* They "vendor" a third party package, either incompletely, using hacks visible
|
|
|
|
through the runtime's standard interfaces, or with ancient versions of code
|
|
|
|
that in turn mess with :py:data:`sys.modules` in some horrible way.
|
|
|
|
|
|
|
|
Bugs will probably be filed for these in time, but it does not address the huge
|
|
|
|
installed base of existing old software versions, so hacks are needed anyway.
|
|
|
|
|
|
|
|
|
|
|
|
``pbr``
|
|
|
|
=======
|
|
|
|
|
|
|
|
It claims to use ``pkg_resources`` to read version information
|
|
|
|
(``_get_version_from_pkg_metadata()``), which would result in PEP-302 being
|
|
|
|
reused and everything just working wonderfully, but instead it actually does
|
|
|
|
direct filesystem access.
|
|
|
|
|
|
|
|
**What could it do instead?**
|
|
|
|
|
|
|
|
* ``pkg_resources.resource_stream()``
|
|
|
|
|
|
|
|
**What Mitogen is forced to do**
|
|
|
|
|
|
|
|
When it sees ``pbr`` being loaded, it smodges the process environment with a
|
|
|
|
``PBR_VERSION`` variable to override any attempt to auto-detect the version.
|
|
|
|
This will probably break code I haven't seen yet.
|
|
|
|
|
|
|
|
|
|
|
|
``pkg_resources``
|
|
|
|
=================
|
|
|
|
|
|
|
|
Anything that imports ``pkg_resources`` will eventually cause ``pkg_resources``
|
|
|
|
to try and import and scan ``__main__`` for its ``__requires__`` attribute
|
|
|
|
(``pkg_resources/__init__.py::_build_master()``). This breaks any app that is
|
|
|
|
not expecting its ``__main__`` to suddenly be sucked over a network and
|
|
|
|
injected into a remote process, like py.test.
|
|
|
|
|
|
|
|
A future version of Mitogen might have a more general hack that doesn't import
|
|
|
|
the master's ``__main__`` as ``__main__`` in the slave, avoiding all kinds of
|
|
|
|
issues like these.
|
|
|
|
|
|
|
|
**What could it do instead?**
|
|
|
|
|
|
|
|
* Explicit is better than implicit: wait until the magical behaviour is
|
|
|
|
explicitly requested (i.e. an API call).
|
|
|
|
|
|
|
|
* Use ``get("__main__")`` on :py:data:`sys.modules` rather than ``import``, but
|
|
|
|
this method isn't general enough, it only really helps tools like Mitogen.
|
|
|
|
|
|
|
|
**What Mitogen is forced to do**
|
|
|
|
|
|
|
|
Examine the stack during every attempt to import ``__main__`` and check if the
|
|
|
|
requestee module is named ``pkg_resources``, if so then refuse the import.
|
|
|
|
|
|
|
|
|
|
|
|
``six``
|
|
|
|
=======
|
|
|
|
|
|
|
|
The ``six`` module makes some effort to conform to PEP-302, but it is missing
|
|
|
|
several critical pieces, e.g. the ``__loader__`` attribute. This not only
|
|
|
|
breaks the Python standard library tooling (such as the :py:mod:`inspect`
|
|
|
|
module), but also Mitogen. Newer versions of ``six`` improve things somewhat,
|
|
|
|
but there are still outstanding issues preventing Mitogen from working with
|
|
|
|
``six``.
|
|
|
|
|
|
|
|
This package is sufficiently popular that it must eventually be supported. See
|
|
|
|
`here for an example issue`_.
|
|
|
|
|
|
|
|
.. _here for an example issue: https://github.com/dw/mitogen/issues/31
|
|
|
|
|
|
|
|
**What could it do instead?**
|
|
|
|
|
|
|
|
* Any custom hacks installed into :py:data:`sys.modules` should support the
|
|
|
|
protocols laid out in PEP-302.
|
|
|
|
|
|
|
|
**What Mitogen is forced to do**
|
|
|
|
|
|
|
|
Vendored versions of ``six`` currently don't work at all.
|