At current default-branch commit 8ad5ab7eddb85138a06824cc517ef6f32c3b7d93, two form/query decoders validate percent pairs with std.fmt.parseInt(u8, pair, 16):
On Zig 0.17.0, this accepts %+A as newline and %-0 as NUL. Each escape must contain exactly two hexadecimal digits. This is a decoder correctness finding; no downstream exploit is claimed.
Zat v0.7.0-alpha.10 provides strict URI decoding, but it preserves raw +. Stream must retain its form behavior: raw + becomes space, %2B remains plus, and decoding happens once. Do not convert plus after percent decoding. Coordinate shared ownership with the Zat follow-up; a small per-digit fix need not wait for a larger API decision.
Acceptance:
- Reject signed/nonhex/truncated escapes in both helpers while preserving their error mappings and form semantics.
- Cover
+,%2B,%252B, mixed-case hex, NUL and malformed suffixes through filter/status-page callers. - Exercise allocation failures and cleanup, including malformed-input exits.
- Run adversarial review, replayable fuzz/property checks and focused before/after performance tests before shipping, then smoke the affected query paths.
Background: shared decoder review, following Brooke's ZDS contribution.