Module encoding/percent

Percent-encoding (RFC 3986) — the URI component codec (plans/archive/STD_API_AUDIT.md §7 P0 item 1).

percent.encode keeps the RFC 3986 UNRESERVED set (A-Z a-z 0-9 - _ . ~) and encodes every other BYTE as %XX (uppercase hex), so multibyte UTF-8 is encoded byte-by-byte — the component form (encodeURIComponent / Rust's percent-encoding NON_ALPHANUMERIC minus the unreserved marks).

percent.decode reverses it: %XX → byte, validated as UTF-8 on the way back into a String (D1 style 2: a pure fallible transform returning Result(_, PercentError)); percent.decode_bytes returns the raw bytes for callers decoding non-text data.

+ is NOT treated as a space: that is the application/x-www-form-urlencoded FORM dialect, not RFC 3986 — a form variant can be added additively.

Stability

unstable — percent_encode's escape set is HARDCODED to the unreserved set, and that is the open question. It makes this function encodeURIComponent and nothing else: handed a whole path it escapes the / to %2F, handed a query string it escapes the & and =, so the only correct use is one component at a time. Rust's percent-encoding crate takes an AsciiSet for exactly this reason and ships NON_ALPHANUMERIC, CONTROLS, PATH_SEGMENT, QUERY and USERINFO variants of it. Growing that here is additive if it arrives as a second function (percent_encode_with(input, set)) and breaking if percent_encode gains a parameter — and which of those it is has not been decided.

Two smaller open items. The FORM dialect (+ for space) is still absent, noted above as additive, and std/url's query_pairs currently hand-rolls it by replacing + before calling here — so the encoder half has no home at all. And the names stutter (D2): percent_encode repeats its module, where the decided direction for the whole encoding group is percent.encode on the imported module value (plans/STD_API_STABILIZATION.md §4, STILL OPEN).

What IS settled: the decoder's split into percent_decode (validating UTF-8, returning a String) and percent_decode_bytes (raw, for non-text data), and PercentError carrying the byte index of the fault rather than only its kind. Both have been through several releases unchanged and are covered by tests/encoding/percent.test.yo.