nord-format 0.4.0

//! The section chains that make up a sample instrument's body.
//!
//! Two framings, one shape. The v2 body is an `NWS` chain of 9-byte-header
//! [`Section`]s; the v3/v4 body is an `NSMP` chain of 12-byte-header
//! [`Section4`]s. Both are flat tag/version/length runs that must land exactly
//! on the end of the body.

use crate::error::ParseError;

/// Bytes of section header: 3-char tag, NUL, `u8` version, `u32` length.
pub const HEADER_LEN: usize = 9;

/// Opens the body. Its payload is empty; the first real section follows it.
pub const CONTAINER: &[u8; 3] = b"NWS";

pub const HDR: &[u8; 3] = b"hdr";
pub const CAT: &[u8; 3] = b"cat";
pub const MAP: &[u8; 3] = b"map";
pub const STK: &[u8; 3] = b"stk";
pub const STY: &[u8; 3] = b"sty";

/// One section of a sample instrument body.
///
/// ⚠️ `len` on the wire is **big-endian**, inside a CBIN header that is little-endian
/// throughout. Reading it the same way as the header's `u32`s yields a nonsense length
/// in the hundreds of millions.
///
/// The length counts the payload only — the 9 header bytes are not included.
#[derive(Clone, PartialEq, Eq)]
pub struct Section {
    pub tag: [u8; 3],
    /// Schema version of this section alone; sections revise independently.
    pub version: u8,
    pub payload: Vec<u8>,
}

impl Section {
    /// Length on disk, header included.
    pub fn encoded_len(&self) -> usize {
        HEADER_LEN + self.payload.len()
    }

    pub fn tag_str(&self) -> String {
        String::from_utf8_lossy(&self.tag).into_owned()
    }

    pub fn is(&self, tag: &[u8; 3]) -> bool {
        &self.tag == tag
    }

    pub fn write_to(&self, w: &mut impl std::io::Write) -> Result<(), ParseError> {
        let head = |w: &mut dyn std::io::Write| -> std::io::Result<()> {
            w.write_all(&self.tag)?;
            w.write_all(&[0, self.version])?;
            w.write_all(&(self.payload.len() as u32).to_be_bytes())?;
            w.write_all(&self.payload)
        };
        head(w).map_err(|e| ParseError::AssertFail(format!("writing a section: {e}")))
    }
}

/// A chain whose first section is not the container that opens it — the leading
/// sections are missing, or the bytes are not a chain at all. Raised before the first
/// declared length is trusted; on a corrupt opener that length is arbitrary.
fn wrong_opener(expected: &[u8], found: &[u8]) -> ParseError {
    ParseError::AssertFail(format!(
        "the body does not open with the {} container section; found {}",
        expected.escape_ascii(),
        found.escape_ascii(),
    ))
}

fn missing_opener(expected: &[u8]) -> ParseError {
    ParseError::AssertFail(format!(
        "the body does not open with the {} container section; found end of body",
        expected.escape_ascii(),
    ))
}

/// Walks the chain from the reader's position to its end.
///
/// The chain must open with [`CONTAINER`] and land exactly on the end of the body. Both
/// are strong integrity checks — a wrong length anywhere puts every later section at the
/// wrong offset — so a short or overrunning walk is an error rather than a truncated
/// result.
pub fn read_chain(r: &mut impl std::io::Read) -> Result<Vec<Section>, ParseError> {
    let mut sections = Vec::new();
    let mut pos: u64 = 0;
    loop {
        let head = match read_head(r, pos)? {
            Some(head) => head,
            None if pos == 0 => return Err(missing_opener(CONTAINER)),
            None => return Ok(sections),
        };
        if pos == 0 && &head[..3] != CONTAINER {
            return Err(wrong_opener(CONTAINER, &head[..3]));
        }
        let len = u32::from_be_bytes([head[5], head[6], head[7], head[8]]) as usize;
        let mut payload = vec![0u8; len];
        r.read_exact(&mut payload).map_err(|_| {
            ParseError::AssertFail(format!(
                "section {} at {pos} declares {len} bytes but the body ends first",
                String::from_utf8_lossy(&head[..3]),
            ))
        })?;
        pos += (HEADER_LEN + len) as u64;
        sections.push(Section {
            tag: [head[0], head[1], head[2]],
            version: head[4],
            payload,
        });
    }
}

/// The next 9-byte section header, `None` on a clean end of the chain. Bytes that
/// run out mid-header are a truncation, not an end.
fn read_head(r: &mut impl std::io::Read, at: u64) -> Result<Option<[u8; 9]>, ParseError> {
    let mut head = [0u8; HEADER_LEN];
    let mut got = 0;
    while got < HEADER_LEN {
        match r.read(&mut head[got..]) {
            Ok(0) if got == 0 => return Ok(None),
            Ok(0) => {
                return Err(ParseError::AssertFail(format!(
                    "truncated section header at {at}"
                )))
            }
            Ok(n) => got += n,
            Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
            Err(e) => return Err(ParseError::AssertFail(format!("reading a section: {e}"))),
        }
    }
    Ok(Some(head))
}

/// Bytes of a v3/v4 section header: 4-byte tag, `u32` version, `u32` length —
/// the u32s big-endian like the v2 chain's length.
pub const HEADER4_LEN: usize = 12;

/// Opens a v3/v4 body. Its payload is 4 bytes — `0002000c` on every v3
/// specimen, `00020005` on every v4 — constant per generation and unrelated to
/// the stroke count. Meaning open; preserved verbatim.
pub const CONTAINER4: &[u8; 4] = b"NSMP";

/// ⚠️ Three-letter tags are NUL-padded on the *left* in this chain (`\0hdr`),
/// the opposite of the CBIN header's own three-letter tags (`nsp\0`).
pub const HDR4: &[u8; 4] = b"\0hdr";
pub const CAT4: &[u8; 4] = b"\0cat";
pub const MAP4: &[u8; 4] = b"\0map";
pub const STK4: &[u8; 4] = b"\0stk";
pub const STY4: &[u8; 4] = b"\0sty";
/// Trails every v3/v4 specimen, after `sty`.
pub const META4: &[u8; 4] = b"meta";

/// One section of a v3/v4 sample instrument body.
///
/// The `4` is the tag width. Same walk rules as [`Section`]: the length counts
/// the payload only, and the chain must land exactly on the end of the body.
#[derive(Clone, PartialEq, Eq)]
pub struct Section4 {
    pub tag: [u8; 4],
    /// Schema version of this section alone; sections revise independently.
    /// The `map` version — 12/14 on v3 specimens, 21 on v4 — is what selects a
    /// zone-record layout, not the file's content version.
    pub version: u32,
    pub payload: Vec<u8>,
}

impl Section4 {
    /// Length on disk, header included.
    pub fn encoded_len(&self) -> usize {
        HEADER4_LEN + self.payload.len()
    }

    /// The tag without its padding NUL.
    pub fn tag_str(&self) -> String {
        String::from_utf8_lossy(&self.tag)
            .trim_start_matches('\0')
            .to_owned()
    }

    pub fn is(&self, tag: &[u8; 4]) -> bool {
        &self.tag == tag
    }

    pub fn write_to(&self, w: &mut impl std::io::Write) -> Result<(), ParseError> {
        let head = |w: &mut dyn std::io::Write| -> std::io::Result<()> {
            w.write_all(&self.tag)?;
            w.write_all(&self.version.to_be_bytes())?;
            w.write_all(&(self.payload.len() as u32).to_be_bytes())?;
            w.write_all(&self.payload)
        };
        head(w).map_err(|e| ParseError::AssertFail(format!("writing a section: {e}")))
    }
}

/// Walks a v3/v4 chain from the reader's position to its end, under the same
/// open-with-[`CONTAINER4`] and land-exactly rules as [`read_chain`].
pub fn read_chain4(r: &mut impl std::io::Read) -> Result<Vec<Section4>, ParseError> {
    let mut sections = Vec::new();
    let mut pos: u64 = 0;
    loop {
        let mut head = [0u8; HEADER4_LEN];
        if !read_exact_or_end(r, &mut head, pos)? {
            return if pos == 0 {
                Err(missing_opener(CONTAINER4))
            } else {
                Ok(sections)
            };
        }
        if pos == 0 && &head[..4] != CONTAINER4 {
            return Err(wrong_opener(CONTAINER4, &head[..4]));
        }
        let len = u32::from_be_bytes([head[8], head[9], head[10], head[11]]) as usize;
        let mut payload = vec![0u8; len];
        r.read_exact(&mut payload).map_err(|_| {
            ParseError::AssertFail(format!(
                "section {} at {pos} declares {len} bytes but the body ends first",
                String::from_utf8_lossy(&head[..4]),
            ))
        })?;
        pos += (HEADER4_LEN + len) as u64;
        sections.push(Section4 {
            tag: [head[0], head[1], head[2], head[3]],
            version: u32::from_be_bytes([head[4], head[5], head[6], head[7]]),
            payload,
        });
    }
}

/// Fills `buf` from `r`, `Ok(false)` on a clean end before the first byte.
/// Bytes that run out mid-buffer are a truncation, not an end.
fn read_exact_or_end(
    r: &mut impl std::io::Read,
    buf: &mut [u8],
    at: u64,
) -> Result<bool, ParseError> {
    let mut got = 0;
    while got < buf.len() {
        match r.read(&mut buf[got..]) {
            Ok(0) if got == 0 => return Ok(false),
            Ok(0) => {
                return Err(ParseError::AssertFail(format!(
                    "truncated section header at {at}"
                )))
            }
            Ok(n) => got += n,
            Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
            Err(e) => return Err(ParseError::AssertFail(format!("reading a section: {e}"))),
        }
    }
    Ok(true)
}

/// Finds the single v3/v4 section with `tag`.
///
/// ⚠️ The same repeat trap as [`find`]: `stk` repeats, one per stroke.
pub fn find4<'a>(sections: &'a [Section4], tag: &[u8; 4]) -> Option<&'a Section4> {
    sections.iter().find(|s| s.is(tag))
}

pub fn find_mut4<'a>(sections: &'a mut [Section4], tag: &[u8; 4]) -> Option<&'a mut Section4> {
    sections.iter_mut().find(|s| s.is(tag))
}

impl std::fmt::Debug for Section4 {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("Section4")
            .field("tag", &self.tag_str())
            .field("version", &self.version)
            .field("len", &self.payload.len())
            .finish()
    }
}

/// Finds the single section with `tag`.
///
/// ⚠️ Only for tags that appear at most once. `stk` repeats — one per zone — so it must
/// be collected in order instead; a lookup by tag silently keeps one stroke and drops the
/// rest, and every single-zone file hides that completely.
pub fn find<'a>(sections: &'a [Section], tag: &[u8; 3]) -> Option<&'a Section> {
    sections.iter().find(|s| s.is(tag))
}

pub fn find_mut<'a>(sections: &'a mut [Section], tag: &[u8; 3]) -> Option<&'a mut Section> {
    sections.iter_mut().find(|s| s.is(tag))
}

impl std::fmt::Debug for Section {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("Section")
            .field("tag", &self.tag_str())
            .field("version", &self.version)
            .field("len", &self.payload.len())
            .finish()
    }
}

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

    fn section(tag: &[u8; 3], version: u8, payload: &[u8]) -> Vec<u8> {
        let mut v = tag.to_vec();
        v.push(0);
        v.push(version);
        v.extend_from_slice(&(payload.len() as u32).to_be_bytes());
        v.extend_from_slice(payload);
        v
    }

    #[test]
    fn chain_round_trips() {
        let mut bytes = section(CONTAINER, 11, &[]);
        bytes.extend(section(HDR, 9, &[1, 2, 3]));
        bytes.extend(section(STK, 9, &[4; 20]));

        let chain = read_chain(&mut bytes.as_slice()).unwrap();
        assert_eq!(chain.len(), 3);
        assert_eq!(chain[0].payload.len(), 0);
        assert_eq!(chain[1].payload, vec![1, 2, 3]);
        assert_eq!(chain[2].version, 9);

        let mut out = Vec::new();
        for s in &chain {
            s.write_to(&mut out).unwrap();
        }
        assert_eq!(out, bytes);
    }

    /// The empty `NWS` section every v2 body opens with.
    fn opener() -> Vec<u8> {
        section(CONTAINER, 11, &[])
    }

    #[test]
    fn length_is_big_endian() {
        // 0x00000102 = 258 read big-endian; little-endian would be 0x02010000.
        let mut bytes = opener();
        bytes.extend(section(HDR, 1, &[0; 258]));
        let chain = read_chain(&mut bytes.as_slice()).unwrap();
        assert_eq!(chain[1].payload.len(), 258);
    }

    #[test]
    fn overrunning_length_is_an_error() {
        let mut hdr = section(HDR, 1, &[7; 4]);
        hdr[8] = 200; // claim 200 payload bytes where 4 exist
        let mut bytes = opener();
        bytes.extend(hdr);
        assert!(read_chain(&mut bytes.as_slice()).is_err());
    }

    #[test]
    fn trailing_bytes_are_an_error() {
        let mut bytes = opener();
        bytes.extend(section(HDR, 1, &[7; 4]));
        bytes.extend_from_slice(&[0, 0, 0]); // not enough for another header
        assert!(read_chain(&mut bytes.as_slice()).is_err());
    }

    #[test]
    fn a_chain_not_opening_with_its_container_names_the_expected_tag() {
        let bytes = section(HDR, 1, &[7; 4]);
        assert_eq!(
            read_chain(&mut bytes.as_slice()).unwrap_err().to_string(),
            "the body does not open with the NWS container section; found hdr"
        );

        let bytes = section4(HDR4, 1, &[7; 4]);
        assert_eq!(
            read_chain4(&mut bytes.as_slice()).unwrap_err().to_string(),
            "the body does not open with the NSMP container section; found \\x00hdr"
        );
    }

    #[test]
    fn an_empty_body_reports_its_missing_container() {
        assert_eq!(
            read_chain(&mut [].as_slice()).unwrap_err().to_string(),
            "the body does not open with the NWS container section; found end of body"
        );
        assert_eq!(
            read_chain4(&mut [].as_slice()).unwrap_err().to_string(),
            "the body does not open with the NSMP container section; found end of body"
        );
    }

    #[test]
    fn a_corrupt_opener_is_reported_before_its_length() {
        // A garbage tag's "length" is arbitrary: big-endian 0xfeffff3e here.
        let bytes = [0x00, 0x00, 0x00, 0x00, 0xff, 0xfe, 0xff, 0xff, 0x3e];
        assert_eq!(
            read_chain(&mut bytes.as_slice()).unwrap_err().to_string(),
            "the body does not open with the NWS container section; found \\x00\\x00\\x00"
        );
    }

    fn section4(tag: &[u8; 4], version: u32, payload: &[u8]) -> Vec<u8> {
        let mut v = tag.to_vec();
        v.extend_from_slice(&version.to_be_bytes());
        v.extend_from_slice(&(payload.len() as u32).to_be_bytes());
        v.extend_from_slice(payload);
        v
    }

    #[test]
    fn chain4_round_trips() {
        let mut bytes = section4(CONTAINER4, 30, &[0, 2, 0, 0x0c]);
        bytes.extend(section4(HDR4, 10, &[1; 112]));
        bytes.extend(section4(STK4, 11, &[4; 20]));

        let chain = read_chain4(&mut bytes.as_slice()).unwrap();
        assert_eq!(chain.len(), 3);
        assert_eq!(chain[0].payload.len(), 4);
        assert_eq!(chain[1].version, 10);
        assert_eq!(chain[2].tag_str(), "stk");

        let mut out = Vec::new();
        for s in &chain {
            s.write_to(&mut out).unwrap();
        }
        assert_eq!(out, bytes);
    }

    #[test]
    fn chain4_overrun_and_truncation_are_errors() {
        let opener4 = || section4(CONTAINER4, 30, &[0, 2, 0, 0x0c]);

        let mut hdr = section4(HDR4, 1, &[7; 4]);
        hdr[11] = 200; // claim 200 payload bytes where 4 exist
        let mut bytes = opener4();
        bytes.extend(hdr);
        assert!(read_chain4(&mut bytes.as_slice()).is_err());

        let mut bytes = opener4();
        bytes.extend(section4(HDR4, 1, &[7; 4]));
        bytes.extend_from_slice(&[0; 5]); // not enough for another header
        assert!(read_chain4(&mut bytes.as_slice()).is_err());
    }

    #[test]
    fn repeated_tags_are_all_kept() {
        let mut bytes = opener();
        bytes.extend(section(STK, 9, &[1]));
        bytes.extend(section(STK, 9, &[2]));
        bytes.extend(section(STK, 9, &[3]));
        let chain = read_chain(&mut bytes.as_slice()).unwrap();
        assert_eq!(chain.len(), 4);
        assert_eq!(
            chain[1..].iter().map(|s| s.payload[0]).collect::<Vec<_>>(),
            vec![1, 2, 3]
        );
    }
}