Please add the concept of branching in the Virtual Environment Manager

[showkeyjar]

Checklist

• I added a descriptive title
• I searched open requests and couldn't find a duplicate

What is the idea?

Make the virtual environment no longer a flat structure, but a tree structure containing root and branch nodes.

Why is this needed?

I create 10 of environments, it all contain some big packages, for example pytorch, tensorflow etc.
Repeatedly installing huge packages wastes time and space.
Unfortunately, if there is a slight conflict between certain small packages, a new environment must be created.

What should happen?

If conda has branch functions, I can create a root env, install pytorch

then create level1 level2 env1, env2 from root, and then, install those small different packages in env1, env2.

Additional Context

Eagerly anticipating the launch of this feature

[jaimergp]

Unfortunately this is a bit trickier than it looks like. Most conda packages with compiled objects will have been built to link to dynamic libraries in a relative path (usually something like ../lib/mylib.so). If we branch environments like that, there's a chance the required libraries won't be in place to be found. We would need to start adding symlinks in all required places and... well that would end up looking like a flat environment again :)

Another clarification, most of the files are not copied but hardlinked. So the disk usage overhead is actually tiny. If you are concerned about the time spent while hardlinking, then a more interesting approach would be to implement copy-on-write linking.

[showkeyjar]

thanks for your attention.

first, dynamic libraries problem:
There are two types of files that need to be distinguished here: old packages in root env and new packages in branchs envs.
old packages no need to recompiled, new package compiled in branchs envs separately.
so I think it's not problem.

second, tree structure of envs:
I understand your description, but conda only has one root env(base),
for example, if I want create two different root env: pytorch and tensorflow
and then, I create p1,p2 from pytorch env, create t1, t2 from tensorflow env,
it still hardlinked files?
I test those operation on my windows computer, but disk still increase huge.