fix: ensure that the parent of a SourceRoot cannot be itself

[roife]

fix #17378.

In FileSetConfig.map, different roots might be mapped to the same root_id due to deduplication in ProjectFolders::new:

// Example from rustup
/Users/roife/code/rustup/target/debug/build/rustup-863a063426b56c51/out
/Users/roife/code/rustup

In source_root_parent_map, r-a might encounter paths where their SourceRootId (i.e. root_id) is identical, yet one the them is the parent of the another. This situation can cause the root_id to be its own parent, potentially leading to an infinite loop.

This PR resolves such cases by adding a check.

[roife]

Could a circular reference occur here?

For example:

rootA: ["/ROOT/a", "/ROOT/a/b/c/d"]
rootB: ["/ROOT/a/b"]

This might cause rootA to be the parent of rootB while rootB is the parent of rootA.🤯

Or perhaps for paths with a parent-child relationship within the same SourceRoot, we only consider the root path (which means r-a should exit the inner loop when root_id == root2_id).

[rami3l]

@roife I can confirm that your test build on 73399 works for me. Nice work!

[Veykril]

We already depend on petgraph so we could just use a proper graph impl for this instead that then deals with accidental cycle handling if they can actually occur. Though this entire thing is very hacky in the first place...

Thanks a lot for investigating!
@bors r+

[bors]

📌 Commit 733995c has been approved by Veykril

It is now in the queue for this repository.

[Veykril]

@lnicola we might wanna do a second release after this as today's release becomes mostly unusable for a bunch of people (and then there is also the random startup panics that I also (presumably) fixed this morning). If you have the time that is this week :)

[bors]

⌛ Testing commit 733995c with merge c86d623...

[bors]

☀️ Test successful - checks-actions
Approved by: Veykril
Pushing c86d623 to master...