Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[BEAM-1251] Fix Python 3 compatibility bug in pickler.py #7953

Merged
merged 1 commit into from
Feb 28, 2019
Merged

[BEAM-1251] Fix Python 3 compatibility bug in pickler.py #7953

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions sdks/python/apache_beam/internal/pickler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -164,6 +164,17 @@ def new_save_module_dict(pickler, obj):
obj_id = id(obj)
if not known_module_dicts or '__file__' in obj or '__package__' in obj:
if obj_id not in known_module_dicts:
# Trigger loading of lazily loaded modules (such as pytest vendored
# modules).
# This first pass over sys.modules needs to iterate on a copy of
# sys.modules since lazy loading modifies the dictionary, hence the use
# of list().
for m in list(sys.modules.values()):
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we understand why does sys.modules.values() change during iteration? Is it a side-effect of pytest?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Probably, though I don't understand why it sys.modules changes in this loop (assuming no background threads). I believe that accessing m.__dict__ triggers a lazy loading of the module.

I tested adding an additional for-loop before the original. The first loop iterates on list(sys.modules.value()), while the second on sys.modules.value(). The first loop conditionally accesses m.dict to see which modules change sys.modules. If the second loop succeeds, the touched modules are culprits.

I've isolated this module as a culprit:
py._vendored_packages.apipkg.ApiModule

Maybe related:
pytest has a module import hook for rewriting assertions (no need for self.assertEqual(), etc.):
https://github.com/pytest-dev/pytest/blob/7dcd9bf5add337686ec6f2ee81b24e8424319dba/src/_pytest/assertion/rewrite.py#L277-L301

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

https://github.com/pytest-dev/py/tree/master/py/_vendored_packages
https://github.com/pytest-dev/py/blob/master/py/__init__.py

So an alternative solution would be to add an additional loop before the original one:

_ = [m.__dict__ for m in list(sys.modules.values())]

for m in sys.modules.values():
  ...

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@robertwb

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@charlesccychen I believe my above suggestion is the more correct one, allowing lazily-loaded modules to initialize before putting their contents in known_module_dicts.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @udim--that makes sense. Are you going to update this change?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@charlesccychen updated, ready to merge once all tests pass

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is list() essential in line 168? If so please add a comment so that somebody doesn't try to clean it up, otherwise we can remove it.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Right, added more commentary

try:
_ = m.__dict__
except AttributeError:
pass

for m in sys.modules.values():
try:
if (m
Expand Down