chore: Replace fxhash with ahash

[haecker-felix]

The fxhash crate is no longer maintained. See https://rustsec.org/advisories/RUSTSEC-2025-0057.html

The rustc-hash is pretty much a drop-in replacement for fxhash. It has the same type names, but uses a different hashing algorithm under the hood:

The original hash algorithm provided by this crate was one taken from Firefox, hence the hasher it provides is called FxHasher. This name is kept for backwards compatibility, but the underlying hash has since been replaced. The current design for the hasher is a polynomial hash finished with a single bit rotation, together with a wyhash-inspired compression function for strings/slices, both designed by Orson Peters.

[baszalmstra]

If we are switching the hash algorithm we could also use ahash instead!

[haecker-felix]

Now using the standard library’s HashMap and HashSet with ahash::RandomState.

Unlike FxHashMap and FxHashSet, the standard library versions no longer guarantee deterministic ordering of elements. I had to update some tests to account for this.

Not sure whether there may be other unintended side effects that the tests haven’t caught...

[haecker-felix]

Another thing I had to change because of the no longer deterministic ordering of the HashSet.

[baszalmstra]

I prefer we use IndexMap for these types to make sure they are deterministic instead of reordering which is more expensive.

[haecker-felix]

FAIL [   0.040s] rattler_conda_types repo_data::topological_sort::tests::test_topological_sort::case_6
  stdout ───

    running 1 test
    test repo_data::topological_sort::tests::test_topological_sort::case_6 ... FAILED

    failures:

    failures:
        repo_data::topological_sort::tests::test_topological_sort::case_6

    test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 324 filtered out; finished in 0.03s
    
  stderr ───

    thread 'repo_data::topological_sort::tests::test_topological_sort::case_6' panicked at crates/rattler_conda_types/src/repo_data/topological_sort.rs:292:21:
    attempting to install panel (package 336 of 339) but dependency holoviews is not yet installed
    note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

@baszalmstra Even with IndexMap / IndexSet that particular test keeps failing if I don't sort the package names manually.

Could it be possible, that there's a bug which didn't get triggered with FxHashMap as it had a different random ordering - but what happened to be accidentally the correct order for the test to be successful?

[baszalmstra]

I prefer we use IndexMap for these types to make sure they are deterministic instead of reordering which is more expensive.

[Hofer-Julian]

@baszalmstra I think it would be good if you could quickly glance at Felix' comment

@haecker-felix could you please solve the merge conflicts?

[haecker-felix]

Rebased, merge conflicts resolved, and tests are passing now (windows test is an unrelated network error, should be fine when running again).

Looks like my issue in #1674 (comment) has been resolved by another change through rebasing.

[Hofer-Julian]

Thank you, Felix ✨

[baszalmstra]

Ah Im sorry I didnt didnt find the time to properly comment here. I was still noodling what was bothering me and why we now sometimes need to resort the entries.

The current PR changes the functionality of the library in unexpected way, as fields are no longer sorted before serialization. I think we should add back the serialize_with code to ensure that when writing the repodata files they are properly sorted.

[baszalmstra]

I think we should fix that before we release because it might break repodata indexing.