[benr-ml]

Problem

The test_sk_zeroization_on_drop tests (and test_zeroization_on_drop in tbls) take raw pointers into a key inside a block, let the key drop at the end of the block, and then dereference the pointers. Reading through a pointer after its target goes out of scope is undefined behaviour. The tests pass only because the stack slot happens not to be reused, and Miri or AddressSanitizer (use-after-scope) would flag them. This blocks adding those tools to CI.

Fix

• New test helpers in fastcrypto/src/tests/test_helpers.rs:
  • bytes_after_drop(value, ranges) moves the value into a ManuallyDrop, runs its destructor in place with ManuallyDrop::drop, and returns the requested (offset, len) byte ranges. The storage is never freed, so the reads are well defined. It also asserts that every range lies inside the value.
  • field_offset(base, field) computes those offsets while the value is still alive.
• Rewrote the eight affected tests to use them, keeping the same assertions:
  • ed25519
  • secp256k1 and secp256k1 recoverable
  • secp256r1 and secp256r1 recoverable
  • bls12381
  • aes
  • tbls ecies_v1, which inlines the same ManuallyDrop pattern because it is a separate crate

Testing

• cargo test -p fastcrypto -p fastcrypto-tbls zeroization passes, with and without --features copy_key.
• cargo xclippy -D warnings is clean.
• Not yet run under Miri; most of these keys wrap C libraries, which Miri cannot execute.

🤖 Generated with Claude Code