New submission from David Wilson <[email protected]>:
The subprocess package since 880d42a3b24 / September 2018 has begun using this
idiom:
try:
import _foo
except ModuleNotFoundError:
bar
However, ModuleNotFoundError may not be thrown by older import hook
implementations, since that subclass was only introduced in Python 3.6 -- and
so the test above will fail. PEP-302 continues to document ImportError as the
appropriate exception that should be raised.
https://mitogen.readthedocs.io/en/stable/ is one such import hook that lazily
loads packages over the network when they aren't available locally. Current
Python subprocess master breaks with Mitogen because when it discovers the
master cannot satisfy the import, it throws ImportError.
The new exception subtype was introduced in https://bugs.python.org/issue15767
, however very little in the way of rationale was included, and so it's unclear
to me what need the new subtype is addressing, whether this is a problem with
the subprocess module or the subtype as a whole, or indeed whether any of this
should be considered a bug.
It seems clear that some kind of regression is in the process of occurring
during a minor release, and it also seems clear the new subtype will
potentially spawn a whole variety of similar new regressions.
I will be updating Mitogen to throw the new subtype if it is available, but I
felt it was important to report the regression to see what others think.
----------
components: Library (Lib)
messages: 331774
nosy: dw
priority: normal
severity: normal
status: open
title: subprocess module breaks backwards compatibility with older import hooks
versions: Python 3.8
_______________________________________
Python tracker <[email protected]>
<https://bugs.python.org/issue35486>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
