As described in #9, noload was broken by the fix to Python bug 1101399, which was merged into Python 2.7. This in turn breaks code such as zc.zodbdgc.
This PR makes noload return to its 2.6 roots, but keeps the bug fix from upstream: intrinsic objects are loaded and populated, but subclasses of them are ignored.
Furthermore, because noload was only implemented in C for Python 2, it couldn't be used from PyPy, so it also adds noload to the Python implementation for Python 2.
The Python developers seem to consider 'noload' an undocumented extension, and it seems to be removed in Python 3 anyway, so to me it seems that diverging on the behaviour is probably okay.
I tested under Python 2.6, 2.7, 3.2 and 3.3, along with PyPy 2.5.1. (The tests are broken under Python 3.4, I'll try to submit another PR that adds support for 3.4 later.)
tseaver
commented
9 years ago
As described in #9,
noloadwas broken by the fix to Python bug 1101399, which was merged into Python 2.7. This in turn breaks code such aszc.zodbdgc.This PR makes
noloadreturn to its 2.6 roots, but keeps the bug fix from upstream: intrinsic objects are loaded and populated, but subclasses of them are ignored.Furthermore, because
noloadwas only implemented in C for Python 2, it couldn't be used from PyPy, so it also addsnoloadto the Python implementation for Python 2.The Python developers seem to consider 'noload' an undocumented extension, and it seems to be removed in Python 3 anyway, so to me it seems that diverging on the behaviour is probably okay.
I tested under Python 2.6, 2.7, 3.2 and 3.3, along with PyPy 2.5.1. (The tests are broken under Python 3.4, I'll try to submit another PR that adds support for 3.4 later.)