Harmonize shutil.copymode() on Windows and POSIX

[serhiy-storchaka]

The behavior of shutil.copymode() on Windows was changed by adding os.lchmod() in #59616. Now if follow_symlinks is false and both src and dst are symlinks, it copies the symlink mode (only the read-only bit). Previously it did nothing.

But there is still a difference between Windows and POSIX. If follow_symlinks is true and dst is a symlink, it sets the permission of the symlink on Windows (because os.chmod() still does not follow symlinks by default on Windows) and the permission of the target on POSIX.

shutil.copystat() has different behavior, because it passes the follow_symlinks argument explicitly.

I propose to harmonize shutil.copymode() behavior on Windows with POSIX and with shutil.copystat().

Linked PRs

• gh-113188: Fix shutil.copymode() on Windows #113189
• [3.12] gh-113188: Fix shutil.copymode() and shutil.copystat() on Windows #113285
• [3.11] gh-113188: Fix shutil.copymode() and shutil.copystat() on Windows (GH-113285) #113426

[serhiy-storchaka]

I am not sure whether this change should be backported.

[zooba]

I am not sure whether this change should be backported.

I left a comment on the PR pointing out the logic is too complicated for me to see exactly what changed. However, I'd suggest that as this function doesn't really do much of interest on Windows (read only property is the only mode we support, and it's not a security attribute), chances are the only people using it are using it for POSIX and just glad it works elsewhere.

If they'll be glad that their non-readonly (rare) symlink to a (rare) readonly file now correctly copies the readonly attribute, then I'd say backport. (I maybe misunderstood the scenario, but I think the logic applies no matter, so probably backport after explaining the logic clearly in the docs.)

[serhiy-storchaka]

#113285 is a solution for 3.12.

[zooba]

Is there a specific reason we should fix copymode/copystat but not chmod in 3.12? Or just leave them all for 3.13 only?

[serhiy-storchaka]

My main motivation is to avoid adding complicated versionchanged directives and What's New entry. They are not needed for bugfix.

Backporting changes in chmod does not help much without backporting the addition of os.lchmod, because some code tests its existency. And this is further in a "new feature" territory. If do not backport os.lchmod, the code will be of similar complexity to #113285.

[zooba]

Fair enough. Backporting the fix for the high level function seems okay to me now.