Is PY patching a problem (esp. on Windows)

[Dennis Reinhardt]

If I understand correctly , Python looks only at the file timestamp to
determine whether to re-compile a PY file to create a PYC file.  If so, what
prevents the following scenario:

Feb. 2, I release a complete embedded Python hosted program
Feb 10, I issue a zipped PY patch file
Feb 13 (a bad day) the user installs and runs the Feb 2 release.  The Feb 10
patch is applied subsequently.

It looks to me on Feb 13 that the date of the PY file from the Feb 2 release
is Feb 2 and the date of the created PYC is Feb 13.  Now the patch is
applied and the PY file date changes from Feb. 2 to Feb 10.  ... but the
date of the PYC is later and so it does not get changed.  The PYC file is
thus out of date.

On Windows, unzipping a file does not change the file creation date.  This
scenario may not be a problem under Unix because the PY patch file will
carry the timestamp of when it was copied, not when it was created.

I understand the workaround of including a PYC file in the Feb 10 patch file
(doubling the download size).  I am writing here to see if this is a non-bug
because I am overlooking something, if it is a bug, or there is another
solution already in place.  One thing I wonder about and think would be
useful is if there is a way for a program or running option to suppress
generating the PYC file in the first place.
--

Dennis Reinhardt

http://www.dair.com