Low-severity coverage gaps from adversarial review:
- GCM short tags have no known-answer coverage on the wasmtime target: the vendored
aes_gcm_test.json is all tagSize 128 (upstream offers no short-tag file); short tags are covered by one round-trip probe plus WPT's tagLength sweep, which reaches the composed provider and jco but never the wasmtime host. The "Deliberately deferred" golden-artifact rationale ("the vector cases already pin the wire format on every target") does not hold for tag-size < default. Residual risk is small (truncation lives in shared rust/core).
- The composed target's buffering-limit error path is never driven end-to-end: buffer.rs has unit tests, jco and wasmtime have direct admission tests, but no gate runs a composed instance into its store memory limit to observe the documented
error.other.
- Drain-on-error under chunked delivery: the cipher kind's invalid-parameter paths run whole-schedule only, and AES-CTR (no rejection vectors) is never touched by the 1-in-20 straddle rejection sample; AES-CBC's sample covers the kind's decrypt-failure family, so this is a corner.
Related: #84.
Low-severity coverage gaps from adversarial review:
aes_gcm_test.jsonis all tagSize 128 (upstream offers no short-tag file); short tags are covered by one round-trip probe plus WPT's tagLength sweep, which reaches the composed provider and jco but never the wasmtime host. The "Deliberately deferred" golden-artifact rationale ("the vector cases already pin the wire format on every target") does not hold for tag-size < default. Residual risk is small (truncation lives in shared rust/core).error.other.Related: #84.