GeezerVison

@serhiy-storchaka, since this is 3.12+, would it be possible to use inspect.signature() instead of relying on a new attribute on the function (supports_root_dir)? Pure Python functions, as well as Cython functions, work fine with inspect.signature().

I do not like relying on inspect. It fails in too many cases. What if the function signature contains **kwargs? inspect looks to me rather as an inspecting and debugging tool.

Setting a magic attribute on a function is always an awkward API. Why not just add a keyword only argument to the register_*_format() function to declare whether the supplied archiver callable supports a root_dir kwarg or not?

Setting a magic attribute on a function is always an awkward API. Why not just add a keyword only argument to the register_*_format() function to declare whether the supplied archiver callable supports a root_dir kwarg or not?

I considered this option. But it would make the caller of shutil.register_archive_format() responsible for passing the necessary option. For example, if py7zr add support for root_dir, the users of py7zr will need to replace shutil.register_archive_format('7zip', pack_7zarchive, description='7zip archive') in their code with shutil.register_archive_format('7zip', pack_7zarchive, description='7zip archive', supports_root_dir=True). Since it will only works on Python 3.12+, the code will be more complicated. Users which will not do this, will still have the working code, but it is not safe.

In future we can deprecate archivers which does not support root_dir. And finally raise an error if shutil.register_archive_format() is called without supports_root_dir=True. Since that this option will became useless but mandatory, and several versions later we could deprecate and finally remove it. Thus, users will be first forced to add this option (first conditionally, then unconditionally), them remove it. And users which skip versions will risk to get incorrectly working program (using an archiver which does not support root_dir with the root_dir argument).

This all will take very long time until we came to new equilibrium, will cause headache to users, and will be errorprone.

With the variant implemented in this PR, the author of the archiver is responsible for setting the supports_root_dir attribute if the archiver supports the root_dir argument. The user's code will not change.

Thank you for your review and suggestions.