Zat v0.7.0-alpha.10 ships strict URI decoding through zat.percent.decodeAlloc and zat.percent.decodedEql. Its literal-+ behavior is correct. The downstream audit found form/query decoders that cannot adopt it directly because they map raw + to space.
Decide whether the shared form-decoding primitive belongs in Zat, in zttp, or in a small caller adapter before extending the API. Concrete consumers:
- Stream filter and status-page decoders use form semantics and parse percent pairs with
std.fmt.parseInt(u8, pair, 16). - ZDS HTTP decoder has the same semantics and needs caller-provided output storage.
- Prefect server also has a form decoder, but deliberately tolerates malformed escapes; it is not automatically a consumer of a strict API.
parseInt accepts signed pairs: %+A becomes byte 0x0a and %-0 becomes NUL. These are not valid percent escapes. Preserve raw + → space, %2B → literal plus, and exactly one decoding pass. Replacing plus after URI decoding would incorrectly turn %2B into space.
Acceptance:
- Record the ownership/API decision using the actual caller requirements; avoid adding unused modes or wrappers.
- If shared code is added, reuse strict per-digit validation and keep URI behavior unchanged.
- Cover signs, truncated/nonhex escapes,
%2B,%252B, NUL, allocation failures and buffer boundaries where applicable. - Run
zig zen, adversarial review, replayable fuzzing with mutation calibration, and paired performance measurements before shipping; verify real consumers.
This follows Brooke's ZDS fix and the shared-decoder review and measurements. The released URI fix is complete; this issue tracks the remaining form semantics and ownership decision.
Downstream tracking: Stream #6, ZDS #34, Prefect server #35. The URI-only adopter is tracked separately in Typeahead #51.