This sounds like a project requiring a considerable amount of plumbing to get the info from where it's available to where it's needed. For example, importlib reads the PYC file, checks the header, and then passes the rest of the file to marshal.loads(), which creates the (nested) code objects. Similarly, when reading the PY file, the compile() builtin is called to create the code objects without access to metadata other than filename.

I also question whether the software updates that fall prey to this issue are being done the right way -- maybe the server should be stopped before moving the new files in place. :-)