refactor(crypto): move inline merkle-backend tests to tests/

[diegokingston]

Relocate #[cfg(test)] mod tests blocks out of crypto/crypto source files into crypto/crypto/src/tests/.

• backends/field_element.rs tests -> tests/field_element_tests.rs
• backends/field_element_vector.rs tests -> tests/field_element_vector_tests.rs
• merkle.rs disk_spill_serde_tests -> appended to tests/merkle_tests.rs
• MerkleTree::mmap_backing bumped to pub(crate) so the relocated disk-spill serde test can reach it

[claude]

Medium — unnecessary visibility widening

mmap_backing is promoted to pub(crate) solely so the relocated test can write restored.mmap_backing.is_none(). This leaks a disk-spill implementation detail to every crate-internal caller, not just test code.

Preferred alternatives (in order of preference):

1. Add a #[cfg(test)] accessor on MerkleTree so the visibility stays private outside tests:

Suggested change

	pub(crate) mmap_backing: Option<MmapNodeBacking>,
	mmap_backing: Option<MmapNodeBacking>,

#[cfg(test)]
impl<B: IsMerkleTreeBackend> MerkleTree<B> {
    pub(crate) fn has_mmap_backing(&self) -> bool {
        self.mmap_backing.is_some()
    }
}

2. Test a weaker but equivalent invariant that doesn't require field access, e.g. check that restored.nodes is non-empty instead.

[claude]

Low — redundant test in cfg predicate

This file is only compiled because lib.rs declares #[cfg(test)] pub mod tests;, so the test condition here is already implied. The inner module will only ever exist during a test build.

Suggested change

	#[cfg(all(test, feature = "serde", feature = "disk-spill"))]
	#[cfg(all(feature = "serde", feature = "disk-spill"))]

[github-actions]

Codex Code Review

No issues found in the PR diff.

I could not complete test verification in this sandbox: cargo test was blocked by rustup/cargo trying to write under read-only /home/runner/.rustup / /home/runner/.cargo, and offline mode still required the workspace git dependency checkout.

[github-actions]

Codex Code Review

Findings

• Low Security / Invariant Risk: crypto/crypto/src/merkle_tree/merkle.rs widens mmap_backing from private to pub(crate) only so the relocated test can inspect it. This field participates in the unsafe mmap-backed node access invariant; after spill_nodes_to_disk, nodes is empty and correctness depends on mmap_backing not being mutated externally. Prefer keeping the test inside merkle.rs, or add a narrow #[cfg(test)] accessor like has_mmap_backing() instead of exposing the field crate-wide.

No Critical/High/Medium issues found in the PR diff.

I attempted to run tests, but cargo test was blocked because rustup tried to write under /home/runner/.rustup/tmp, which is read-only in this environment.

[claude]

Low — wildcard import pulls in more than needed

Only MerkleTree is used from this module; the glob makes the dependency surface opaque.

Suggested change

	use crate::merkle_tree::merkle::*;
	use crate::merkle_tree::merkle::MerkleTree;

[claude]

Review: refactor(crypto): move inline merkle-backend tests to tests/

Clean refactor — the test logic is identical to what was removed from the source files, and the module structure is correct (the tests parent is already behind #[cfg(test)] in lib.rs, so all the pub mod declarations in tests/mod.rs are safe).

Three minor issues flagged inline:

Severity	File	Issue
Medium	merkle.rs:60	mmap_backing bumped to pub(crate) just to serve one test assertion — leaks an internal field to all crate code when disk-spill is active. A #[cfg(test)] accessor keeps the field private.
Low	merkle_tests.rs:140	#[cfg(all(test, ...)) — test is redundant; the whole tests module is already #[cfg(test)].
Low	merkle_tests.rs:143	use crate::merkle_tree::merkle::* — only MerkleTree is used; prefer the explicit import.

The pub(crate) concern is the only one worth addressing before merge; the other two are cosmetic.

[claude]

Low — unnecessary alloc::vec::Vec import

This was needed in the original field_element.rs where it was compiled in no_std configurations. This tests/ module is #[cfg(test)]-gated and always compiled with std, so Vec is already in the prelude. The sibling field_element_vector_tests.rs doesn't import it.

Suggested change

use alloc::vec::Vec;

[claude]

Low — dead test predicate in cfg guard

This file lives inside a module that is already declared #[cfg(test)] in lib.rs, so the test predicate here is always true and does nothing. It was correct in merkle.rs (where it was copy-pasted from) because that file is compiled unconditionally, but here it's misleading.

Suggested change

	#[cfg(all(test, feature = "serde", feature = "disk-spill"))]
	#[cfg(all(feature = "serde", feature = "disk-spill"))]