orbit-oss/flask
4
4
Fork 2
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Refactored flask.ext process to not swallow exceptions on weird Pythons.

Browse source
This commit is contained in:
Armin Ronacher 2011-09-27 13:33:23 +02:00
parent f01b654ac4
commit 95c4dcb4d5
4 changed files with 211 additions and 109 deletions
Download patch file Download diff file Expand all files Collapse all files

92
flask/ext/__init__.py
View file

@ -19,91 +19,11 @@
"""
class _ExtensionImporter(object):
"""This importer redirects imports from this submodule to other locations.
This makes it possible to transition from the old flaskext.name to the
newer flask_name without people having a hard time.
"""
_module_choices = ['flask_%s', 'flaskext.%s']
def __init__(self):
from sys import meta_path
self.prefix = __name__ + '.'
self.prefix_cutoff = __name__.count('.') + 1
# since people might reload the flask.ext module (by accident or
# intentionally) we have to make sure to not add more than one
# import hook. We can't check class types here either since a new
# class will be created on reload. As a result of that we check
# the name of the class and remove stale instances.
def _name(x):
cls = type(x)
return cls.__module__ + '.' + cls.__name__
this = _name(self)
meta_path[:] = [x for x in meta_path if _name(x) != this] + [self]
def find_module(self, fullname, path=None):
if fullname.startswith(self.prefix):
return self
def load_module(self, fullname):
from sys import modules, exc_info
if fullname in modules:
return modules[fullname]
modname = fullname.split('.', self.prefix_cutoff)[self.prefix_cutoff]
for path in self._module_choices:
realname = path % modname
try:
__import__(realname)
except ImportError:
exc_type, exc_value, tb = exc_info()
# since we only establish the entry in sys.modules at the
# very this seems to be redundant, but if recursive imports
# happen we will call into the move import a second time.
# On the second invocation we still don't have an entry for
# fullname in sys.modules, but we will end up with the same
# fake module name and that import will succeed since this
# one already has a temporary entry in the modules dict.
# Since this one "succeeded" temporarily that second
# invocation now will have created a fullname entry in
# sys.modules which we have to kill.
modules.pop(fullname, None)
if self.is_important_traceback(realname, tb):
raise exc_type, exc_value, tb
continue
module = modules[fullname] = modules[realname]
if '.' not in modname:
setattr(modules[__name__], modname, module)
return module
raise ImportError('No module named %s' % fullname)
def is_important_traceback(self, important_module, tb):
"""Walks a traceback's frames and checks if any of the frames
originated in the given important module. If that is the case
then we were able to import the module itself but apparently
something went wrong when the module was imported. (Eg: import
of an import failed).
"""
# Why can we access f_globals' __name__ here and the value is
# not None? I honestly don't know but here is my thinking.
# The module owns a reference to globals and the frame has one.
# Each function only keeps a reference to the globals not do the
# module which normally causes the problem that when the module
# shuts down all globals are set to None. Now however when the
# import system fails Python takes the short way out and does not
# actually properly shut down the module by Noneing the values
# but by just removing the entry from sys.modules. This means
# that the regular reference based cleanup kicks in.
#
# The good thing: At worst we will swallow an exception we should
# not and the error message will be messed up. However I think
# this should be sufficiently reliable.
while tb is not None:
if tb.tb_frame.f_globals.get('__name__') == important_module:
return True
tb = tb.tb_next
return False
def setup():
from ..exthook import ExtensionImporter
importer = ExtensionImporter(['flask_%s', 'flaskext.%s'], __name__)
importer.install()
_ExtensionImporter()
del _ExtensionImporter
setup()
del setup