Bug 843215 – Remove block between sys-apps/file and dev-python/python-magic

• Product: Gentoo Linux
• Status: CONFIRMED ---

[Description Joost Ruis 2022-05-07 23:17:06 UTC]

Looking at this commit here: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=2f63a06672fc2d739428ceeb6764d4ac2aad6256

I wonder if the blocker is still needed on both sides. In my tests I did not see any collisions anymore. Would be great if my findings are true. That would it make it a lot more easy for users to install lutris.

Reproducible: Always

[Comment 1 Mike Gilbert 2022-05-08 00:04:46 UTC]

sys-apps/file[python] installs a module called "magic.py" in site-packages.

dev-python/python-magic installs a directory called "magic" in site-packages.

Although they don't directly collide, you'll see very strange behavior if both are installed at the same time.

[Comment 2 Joost Ruis 2022-05-08 00:26:22 UTC]

Can you define "very strange behavior" here. Sounds arbitrary to me.
Can we investigate this further and resolve it?

[Comment 3 Mike Gilbert 2022-05-08 00:35:39 UTC]

It can only be resolved if one of the packages renames it's python module.

[Comment 4 Mike Gilbert 2022-05-08 00:44:48 UTC]

After a quick test, it looks like python loads packages before modules. In other words, "import magic" will load site-packages/magic/__init__.py, and ignore site-packages/magic.py.

This behavior is probably not obvious to your average python programmer.

[Comment 5 Michał Górny 2022-05-08 06:53:58 UTC]

What Mike said.  Permitting both packages to be installed simultaneously would mean that one would be used and the other ignored.

Now imagine that some packages depends on sys-apps/file[magic].  If one installs dev-python/python-magic, it "magically" replaces it and the app suddenly stops working correctly.

[Comment 6 Joost Ruis 2022-05-08 09:21:02 UTC]

Thanks for explaining it. I now understand the QA issue here.
Shouldn't such a problem be reported upstream?

[Comment 7 Michał Górny 2022-05-08 10:48:19 UTC]

(In reply to Joost Ruis from comment #6)
> Thanks for explaining it. I now understand the QA issue here.
> Shouldn't such a problem be reported upstream?

https://github.com/ahupp/python-magic/issues/21

FWICS upstream pretty much assumes his package can replace the one from sys-apps/file, and the only revdeps of sys-apps/file[python] can use either package.

Perhaps the simplest solution would be to remove USE=python from sys-apps/file, or rename it so it is less likely to be enabled accidentally.  @base-system, WDYT?

[Comment 8 Sam James 2022-06-09 02:45:07 UTC]

(In reply to Michał Górny from comment #7)
> (In reply to Joost Ruis from comment #6)
> > Thanks for explaining it. I now understand the QA issue here.
> > Shouldn't such a problem be reported upstream?
> 
> https://github.com/ahupp/python-magic/issues/21
> 
> FWICS upstream pretty much assumes his package can replace the one from
> sys-apps/file, and the only revdeps of sys-apps/file[python] can use either
> package.
> 
> Perhaps the simplest solution would be to remove USE=python from
> sys-apps/file, or rename it so it is less likely to be enabled accidentally.
> @base-system, WDYT?

I guess I'd prefer it masked (and then, less preferably, renamed) in case we want to test it out at some point.

Keywording done.