ms_codec/

decode.rs

//! Public decoder. Applies SPEC §4 validity rules in order.
//!
//! v0.2.0: also hosts [`decode_with_correction`] — the BCH-error-correcting
//! decode entry point per plan §1 D22 + §2.B.2. Parse → polymod-residue →
//! (if non-zero) call [`crate::bch_decode::decode_regular_errors`] → apply
//! corrections → run the existing [`decode`] path → return
//! `(Tag, Payload, Vec<CorrectionDetail>)`. ms1 is single-chunk per codex32
//! spec, so there is no atomic-multi-chunk variant (cf. md-codec's
//! per-chunk-set version).

use crate::consts::{RESERVED_NOT_EMITTED_V01, TAG_ENTR, VALID_MNEM_STR_LENGTHS, VALID_STR_LENGTHS};
use crate::envelope;
use crate::error::{Error, Result};
use crate::payload::{Payload, PayloadKind};
use crate::tag::Tag;
use codex32::Codex32String;

/// Union of all emittable string lengths (entr ∪ mnem). Used as the
/// pre-dispatch gate in `decode` before kind-specific binding.
fn is_known_length(len: usize) -> bool {
    VALID_STR_LENGTHS.contains(&len) || VALID_MNEM_STR_LENGTHS.contains(&len)
}

/// Return the kind-appropriate allowed-length set for error reporting.
fn allowed_for_kind(kind: PayloadKind) -> &'static [usize] {
    match kind {
        PayloadKind::Entr => VALID_STR_LENGTHS,
        PayloadKind::Mnem => VALID_MNEM_STR_LENGTHS,
    }
}

/// Decode an ms1 string into `(Tag, Payload)`.
///
/// Rejects per SPEC §4 rules 1-10 (extended for v0.2 mnem):
///
/// - Rule 1: upstream codex32 parse failure (Codex32 variant).
/// - Rules 2-4, 8: wire-invariant violations (delegated to envelope::discriminate).
/// - Rules 5-7: tag-table membership rules (here).
/// - Rule 9: total string length not in the union {entr lengths} ∪ {mnem lengths}
///   (here, before parse); then bound to the discriminated kind post-dispatch.
/// - Rule 10: payload byte length mismatch for the tag (here, via Payload::validate()).
pub fn decode(s: &str) -> Result<(Tag, Payload)> {
    // §4 rule 9 (pre-dispatch): total string length must be in the union set.
    if !is_known_length(s.len()) {
        return Err(Error::UnexpectedStringLength {
            got: s.len(),
            allowed: VALID_STR_LENGTHS, // report the entr set as the primary allowed set
        });
    }

    // §4 rule 1: delegate parse + checksum to rust-codex32.
    let c = Codex32String::from_string(s.to_string())?;

    // §4 rules 2, 3, 4, 8 + tag-alphabet rule 5: envelope (returns typed Payload).
    let (tag, payload) = envelope::discriminate(&c)?;

    // §4 rule 9 (post-dispatch, bind to kind): length must be in the kind-appropriate set.
    let kind_allowed = allowed_for_kind(payload.kind());
    if !kind_allowed.contains(&s.len()) {
        return Err(Error::UnexpectedStringLength {
            got: s.len(),
            allowed: kind_allowed,
        });
    }

    // §4 rule 7: reserved-not-emitted tags.
    if RESERVED_NOT_EMITTED_V01.contains(tag.as_bytes()) {
        return Err(Error::ReservedTagNotEmittedInV01 {
            got: *tag.as_bytes(),
        });
    }

    // §4 rule 6: tag must be in the v0.2 accept set (currently {entr}).
    // cycle-15 Lane M (slug #2): MOVE the decoded bytes straight into the
    // public `Payload` rather than cloning out of a throwaway `Zeroizing`
    // envelope. The prior code wrapped `data` in a Zeroizing envelope and then
    // deref-cloned it into the live `Payload`, which only scrubbed the
    // already-moved-from buffer while allocating an EXTRA un-scrubbed heap copy
    // — net theater. The move is strictly fewer copies and byte-identical wire
    // behavior (`Payload::Entr(Vec<u8>)` shape is unchanged — bare-by-design per
    // the deferred public-API slug; callers wrap at their use site, see payload.rs).
    let payload = match *tag.as_bytes() {
        x if x == TAG_ENTR => {
            match payload {
                Payload::Entr(data) => {
                    let p = Payload::Entr(data);
                    // §4 rule 10: validate payload length.
                    p.validate()?;
                    p
                }
                Payload::Mnem { language, entropy } => {
                    let p = Payload::Mnem { language, entropy };
                    // §4 rule 10: validate (language range + entropy length).
                    p.validate()?;
                    p
                }
            }
        }
        _ => {
            return Err(Error::UnknownTag {
                got: *tag.as_bytes(),
            });
        }
    };

    Ok((tag, payload))
}

// ---------------------------------------------------------------------------
// v0.2.0: BCH-error-correcting decode (plan §1 D22 + §2.B.2).
// ---------------------------------------------------------------------------

/// Per-correction report emitted by [`decode_with_correction`]. One entry
/// per repaired character. `position` is 0-indexed into the codex32
/// data-part (i.e. the characters following the `ms1` HRP + separator);
/// `was` is the original (corrupted) char from the input; `now` is the
/// corrected char.
///
/// ms1 is single-chunk per codex32 spec, so there is no `chunk_index`
/// field (cf. md-codec's `CorrectionDetail`).
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct CorrectionDetail {
    /// 0-indexed position of the corrected character within the codex32
    /// data-part (post-HRP-and-separator).
    pub position: usize,
    /// The original (corrupted) character at this position.
    pub was: char,
    /// The corrected character at this position.
    pub now: char,
}

/// Local codex32 alphabet (BIP 173 lowercase). Each char = one 5-bit
/// symbol. Mirrors md-codec's `chunk.rs` local copy — kept private here so
/// this module doesn't widen the codex32 public surface.
const CODEX32_ALPHABET: &[u8; 32] = b"qpzry9x8gf2tvdw0s3jn54khce6mua7l";

/// BIP 173 HRP for ms1 strings (HRP + separator).
const HRP_PREFIX: &str = "ms1";

/// Parse an ms1 string into its 5-bit data-part symbol vector. Returns
/// the data-with-checksum symbols (i.e. all symbols after `ms1`). The
/// returned symbol count includes the 13-symbol BCH checksum tail.
///
/// Returns [`Error::WrongHrp`] if the string does not start with `ms1`,
/// or [`Error::Codex32`] (via a `codex32::Error::InvalidChar`) if any
/// data-part character is not in the codex32 alphabet.
fn parse_ms1_symbols(s: &str) -> Result<Vec<u8>> {
    let lower = s.to_ascii_lowercase();
    if !lower.starts_with(HRP_PREFIX) {
        // Report the observed HRP (everything before the last '1' separator)
        // so the error is actionable. '1' is ASCII, so `rfind('1')` always
        // returns a char boundary — slicing there is safe regardless of any
        // multi-byte content elsewhere. When there is NO separator, the whole
        // (malformed) string is the observed HRP; never slice at `len-1`,
        // which can land inside a multi-byte char and panic (found by
        // stress-Cycle-C fuzzing on a no-`'1'` lossy-UTF8 input).
        //
        // SECRET-LEAK BOUND (ms-codec-error-display-echoes-input, 0.4.4): a
        // data-char→`'1'` mutation can stretch the "observed HRP" into a long
        // secret prefix. Cap the stored `got` to the first 4 CHARS (not bytes —
        // multibyte chars like "ñ"/"é"/"😀" would re-introduce the v0.4.3 panic
        // on a byte slice). 4 < the 8-char leak window and still carries the
        // "you typed mk1/lnbc not ms1" diagnostic. Construction-time bound so
        // downstream re-echoers (ms-cli, toolkit) inherit it for free.
        let observed = match lower.rfind('1') {
            Some(i) => &lower[..i],
            None => &lower,
        };
        let got = observed.chars().take(4).collect::<String>();
        return Err(Error::WrongHrp { got });
    }
    let rest = &lower[HRP_PREFIX.len()..];
    let mut symbols: Vec<u8> = Vec::with_capacity(rest.len());
    // Non-alphabet characters can't appear in a valid v0.1 string. We
    // can't fabricate a `codex32::Error` value here (the upstream crate
    // doesn't expose a constructor for `InvalidChar`), so we use
    // `UnexpectedStringLength` as a stand-in: the existing `decode` path
    // would have rejected the string for the same reason on a different
    // axis. Toolkit-side helper at B.7 absorbs into `UnparseableInput`
    // per plan §2.B.4 D29 error-mapping table.
    for c in rest.chars() {
        let lc = c as u8;
        let sym = CODEX32_ALPHABET
            .iter()
            .position(|&b| b == lc)
            .ok_or(Error::UnexpectedStringLength {
                got: s.len(),
                allowed: VALID_STR_LENGTHS,
            })? as u8;
        symbols.push(sym);
    }
    Ok(symbols)
}

/// Re-encode a 5-bit data-part symbol vector as a complete ms1 string.
fn encode_ms1_string(data_with_checksum: &[u8]) -> String {
    let mut out = String::with_capacity(HRP_PREFIX.len() + data_with_checksum.len());
    out.push_str(HRP_PREFIX);
    for &v in data_with_checksum {
        out.push(CODEX32_ALPHABET[(v & 0x1F) as usize] as char);
    }
    out
}

/// BCH-error-correcting decode for a single ms1 string.
///
/// Per plan §1 Q1 lock — full-decode semantics: this is the single entry
/// point that callers needing both "did anything get repaired?" AND "the
/// fully-decoded `(Tag, Payload)`" should use.
///
/// Algorithm:
/// 1. Parse the input as ms1 (`ms1` HRP + codex32 data-part) into a
///    5-bit symbol vector.
/// 2. Compute the BCH polymod residue
///    (`hrp_expand("ms") || data_with_checksum`) XOR'd against
///    [`crate::bch::MS_REGULAR_CONST`].
/// 3. Residue `== 0` ⇒ clean codeword; pass through to the existing
///    [`decode`] entry point unchanged.
/// 4. Residue `!= 0` ⇒ invoke
///    [`crate::bch_decode::decode_regular_errors`]. If `None`, return
///    `Err(Error::TooManyErrors { bound: 8 })` per plan §2.B.4 D29
///    error-mapping table.
/// 5. Apply corrections to the symbol vector, re-verify via polymod (a
///    defensive catch for pathological 5+-error patterns that fool BM
///    into returning a degree-≤4 locator with 4 valid roots), and record
///    one [`CorrectionDetail`] per repaired character.
/// 6. Re-encode the corrected symbol vector as an ms1 string and forward
///    it to the existing [`decode`] entry point.
///
/// Per Q1 lock + D29 error-mapping table, any §4-rule error from the
/// full decode (orphan variants like `ThresholdNotZero`,
/// `ReservedTagNotEmittedInV01`, etc.) surfaces directly; toolkit-side
/// `repair_via_ms_codec` (B.7) absorbs these into
/// `RepairError::PostCorrectionDecodeFailed`.
///
/// Returns `(Tag, Payload, Vec<CorrectionDetail>)` on success. The
/// correction-detail vector is in ascending `position` order; an empty
/// vector means the input was already a valid codeword.
pub fn decode_with_correction(s: &str) -> Result<(Tag, Payload, Vec<CorrectionDetail>)> {
    // Parse data-part symbols. Length checks live in `decode` proper
    // (rule 9 is enforced there after we've potentially corrected, since
    // BCH correction does not change the string length).
    let symbols = parse_ms1_symbols(s)?;

    // Polymod residue against ms1's target constant.
    let mut input = crate::bch::hrp_expand("ms");
    input.extend_from_slice(&symbols);
    let residue = crate::bch::polymod_run(&input) ^ crate::bch::MS_REGULAR_CONST;

    if residue == 0 {
        // Already a valid codeword; pass through to the existing decoder.
        let (tag, payload) = decode(s)?;
        return Ok((tag, payload, Vec::new()));
    }

    // Attempt BCH correction.
    let (positions, magnitudes) = crate::bch_decode::decode_regular_errors(residue, symbols.len())
        .ok_or(Error::TooManyErrors { bound: 8 })?;

    // Apply corrections; record (was, now) chars per position.
    let mut corrected = symbols.clone();
    let mut details: Vec<CorrectionDetail> = Vec::with_capacity(positions.len());
    for (&pos, &mag) in positions.iter().zip(&magnitudes) {
        if pos >= corrected.len() {
            // Defensive: chien_search bounded pos to [0, L); but a
            // pathological 5+-error pattern could in principle skirt
            // that.
            return Err(Error::TooManyErrors { bound: 8 });
        }
        let was_byte = corrected[pos];
        let now_byte = was_byte ^ mag;
        let was = CODEX32_ALPHABET[(was_byte & 0x1F) as usize] as char;
        let now = CODEX32_ALPHABET[(now_byte & 0x1F) as usize] as char;
        details.push(CorrectionDetail {
            position: pos,
            was,
            now,
        });
        corrected[pos] = now_byte;
    }

    // Defensive re-verify (catches pathological 5+-error patterns that
    // happen to produce a degree-≤4 locator with 4 valid roots).
    let mut verify_input = crate::bch::hrp_expand("ms");
    verify_input.extend_from_slice(&corrected);
    let verify_residue =
        crate::bch::polymod_run(&verify_input) ^ crate::bch::MS_REGULAR_CONST;
    if verify_residue != 0 {
        return Err(Error::TooManyErrors { bound: 8 });
    }

    // Hand the corrected string to the existing decoder. Any §4-rule
    // error surfaces directly per Q1 lock; toolkit helper at B.7 absorbs.
    let corrected_str = encode_ms1_string(&corrected);
    let (tag, payload) = decode(&corrected_str)?;
    Ok((tag, payload, details))
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::encode;

    #[test]
    fn round_trip_entr_all_lengths() {
        for len in [16usize, 20, 24, 28, 32] {
            let entropy = (0..len as u8)
                .map(|i| i.wrapping_mul(7))
                .collect::<Vec<_>>();
            let p = Payload::Entr(entropy.clone());
            let s = encode::encode(Tag::ENTR, &p).unwrap();
            let (tag, recovered) = decode(&s).unwrap();
            assert_eq!(tag, Tag::ENTR);
            assert_eq!(recovered, p);
        }
    }

    #[test]
    fn decode_rejects_unexpected_length() {
        // 52 chars is outside both the entr set [50,56,62,69,75]
        // and the mnem set [51,58,64,70,77].
        let s = "ms10entrsxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
        assert_eq!(s.len(), 52, "test string must be 52 chars");
        assert!(matches!(
            decode(s),
            Err(Error::UnexpectedStringLength { .. })
        ));
    }

    #[test]
    fn decode_routes_share_to_is_share_not_single_string() {
        // A distributed share of an entr-16 secret is a 50-char string (same
        // length as a v0.1 entr-16 single — disambiguated by the threshold char,
        // not length). It passes the length gate, parses, then discriminate must
        // route it → IsShareNotSingleString (NOT ThresholdNotZero).
        use crate::shares::{encode_shares, Threshold};
        let p = Payload::Entr(vec![0xAAu8; 16]);
        let shares = encode_shares(Tag::ENTR, Threshold::new(2).unwrap(), 3, &p).unwrap();
        let s = &shares[0];
        assert_eq!(s.len(), 50, "threshold=2 entr-16 share must be 50 chars");
        match decode(s) {
            Err(Error::IsShareNotSingleString { threshold, .. }) => {
                assert_eq!(threshold, '2');
            }
            other => panic!("expected IsShareNotSingleString, got {other:?}"),
        }
    }

    #[test]
    fn decode_v01_single_strings_still_ok() {
        // v0.1 entr single + v0.2 mnem single both decode unchanged.
        let entr = encode::encode(Tag::ENTR, &Payload::Entr(vec![0x11u8; 16])).unwrap();
        assert!(decode(&entr).is_ok(), "v0.1 entr single must still decode");
        let mnem = encode::encode(
            Tag::ENTR,
            &Payload::Mnem { language: 1, entropy: vec![0x22u8; 16] },
        )
        .unwrap();
        assert!(decode(&mnem).is_ok(), "mnem single must still decode");
    }

    #[test]
    fn decode_rejects_short_seed_string_with_reserved_tag() {
        // Hand-build a 50-char string with id="seed" — 16-B entropy worth.
        // The string-length check passes; tag-rule 7 fails.
        let mut data = vec![0x00u8];
        data.extend_from_slice(&[0xAAu8; 16]);
        let c = Codex32String::from_seed("ms", 0, "seed", codex32::Fe::S, &data).unwrap();
        let s = c.to_string();
        assert_eq!(s.len(), 50, "expected str.len 50 for 16-B + prefix");
        assert!(matches!(
            decode(&s),
            Err(Error::ReservedTagNotEmittedInV01 { .. })
        ));
    }

    // Regression: `decode_with_correction` must NOT panic on a non-`ms1`
    // input with no `'1'` separator. Found by stress-Cycle-C fuzzing
    // (`ms1_decode`): `parse_ms1_symbols` sliced `lower[..len-1]`, which lands
    // inside a multi-byte char when there is no separator → char-boundary
    // panic. The minimized reproducer is a single `0xaa` byte, which
    // `String::from_utf8_lossy` turns into the 3-byte U+FFFD.
    #[test]
    fn decode_with_correction_no_separator_multibyte_does_not_panic() {
        // Each input has no `'1'`, and `len-1` lands inside a multi-byte
        // char at a different offset (1-, 2-, 3-, 4-byte chars + a long run).
        let cases = [
            String::from_utf8_lossy(&[0xaa]).into_owned(), // U+FFFD, 3 bytes — the fuzz reproducer
            "é".to_string(),                               // 2-byte
            "añ".to_string(),                              // ascii + 2-byte
            "€".to_string(),                               // 3-byte
            "😀".to_string(),                              // 4-byte
            "é".repeat(25),                                // 50-byte multi-byte run
            "İ".to_string(),                               // dotted-capital-I (case-fold edge)
        ];
        for s in &cases {
            // Must return cleanly, never panic. No `'1'` ⇒ WrongHrp, with the
            // observed HRP CAPPED to the first 4 chars (the 0.4.4
            // secret-leak bound; char-counted so multibyte cases don't panic).
            match decode_with_correction(s) {
                Err(Error::WrongHrp { got }) => {
                    assert_eq!(
                        got,
                        s.chars().take(4).collect::<String>().to_ascii_lowercase(),
                        "got is the first 4 chars of the no-separator input (capped)"
                    );
                }
                other => panic!("expected WrongHrp for {s:?}, got {other:?}"),
            }
        }
    }

    // Preservation: an input WITH a `'1'` but a wrong HRP still reports the
    // pre-separator part as `got` (byte-identical to pre-fix behavior).
    #[test]
    fn decode_with_correction_wrong_hrp_with_separator_unchanged() {
        match decode_with_correction("xy1qqq") {
            Err(Error::WrongHrp { got }) => assert_eq!(got, "xy"),
            other => panic!("expected WrongHrp {{ got: \"xy\" }}, got {other:?}"),
        }
        // A `'1'` deep in a multi-byte string still slices at the (ASCII) '1'
        // boundary, never inside the preceding char.
        match decode_with_correction("ñ1zzz") {
            Err(Error::WrongHrp { got }) => assert_eq!(got, "ñ"),
            other => panic!("expected WrongHrp {{ got: \"ñ\" }}, got {other:?}"),
        }
    }
}