Feature: collapse codeowners into directories when possible

[jpmcb]

Suggested solution

Right now, the code owners generation will granularly denote codeowners for each file in the git tree:

path/to/file1 @jpmcb
path/to/file2 @jpmcb
path/to/file3 @jpmcb

In this case, it makes sense to just collapse the directory to:

path/to/ @jpmcb

since all files under that tree are owned by that person. There are probably 1000 and 1 edge cases on this so it'll be critical to think through the data structures and the algorithms for how this would work. Likely, off the top of my head, we'd probably want to take a tree traversal approach where we go deep first over breadth.

[udaypatel1]

Interesting feature!

I'll list an assumption:

1. A directory is considered collapsable if all of its files have the same number and order of attributors

---

Collapsable

path/to/file1 @userA @userC
path/to/file2 @userA @userC
path/to/file3 @userA @userC

Not Collapsable

path/to/file1 @userA
path/to/file2 @userB @userC
path/to/file3 @userB @userA @userC

---

Thinking out loud, I can see 2 paths for potential implementation:

1. Perform the directory collapse process during traversal and CODEOWNERS build time
2. Perform the directory collapse process after traversal and when CODEOWNERS has been built

The latter of the two would be slightly simpler but require the generated CODEOWNERS file to be parsed which may be (algorithmically) less efficient.

Filtering and collapsing directories while generating the CODEOWNERS file would be ideal, but would increase the complexity of designing this feature.

[jpmcb]

Filtering and collapsing directories while generating the CODEOWNERS file would be ideal, but would increase the complexity of designing this feature.

It'd have to be after the file tree with its codeowners has been calculated: this is technically a graph traversal problem and you can't know which nodes to collapse unless you've visited all nodes. This comes with its own challenges sine the internal structure of how the codeowners is calculated isn't really built around this graph / tree idea (and it really should be).