Files
craft-libs/crates/doc-binary/src/fkp.rs
T
Dylan M. Taylor d516b44691 Add craft-doc-binary: WordCraft's .doc reader on craft-wordmodel and craft-cfb
Add craft-doc-binary: WordCraft's .doc reader on craft-wordmodel and craft-cfb

* craft-doc-binary: WordCraft's crates/docbin as it is at wordcraft 07f97737

The source, unchanged, so the next commit shows only what the move changes.
It has no manifest yet, so it is not built.

* craft-doc-binary: on craft-wordmodel, craft-cfb and craft-codepage

- The compound file is read with craft-cfb (bounded directory, streams refused
  above 1 GiB before anything is allocated), in place of the cfb crate and
  WordCraft's cfb_guard precheck; an oversized stream is DocbinError::Limit.
- 8-bit text uses craft-codepage in place of wordcraft_doc::encoding.
- wordcraft-doc is craft-wordmodel; wordcraft-geom's helpers are its geom.

The tests keep building their files with the cfb crate (a dev-dependency), so
craft-cfb's reader is checked against another implementation's writer. The
repeating-DIFAT regression is craft-cfb's own; the guard asserts in the
valid-files test go with the guard.
2026-11-11 22:43:11 -05:00

148 lines
7.1 KiB
Rust
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
//! Character and paragraph formatting lookup: the bin tables (`PlcBteChpx`/`PlcBtePapx`) in
//! the Table stream point at 512-byte FKP pages in the WordDocument stream, which map FC
//! ranges to CHPX (grpprls) and PAPX (istd + grpprl) ([MS-DOC] §2.4.6.1/2.4.6.2).
use crate::DocbinError;
/// Largest bin table (entries) we accept.
const MAX_BINS: usize = 100_000;
/// A Chpx grpprl and the FCs `[lo, hi)` for which it is the one [`Bins::chpx`] returns.
#[derive(Clone, Copy)]
pub(crate) struct ChpxRun<'w> {
pub(crate) grpprl: &'w [u8],
lo: u32,
hi: u32,
}
impl ChpxRun<'_> {
pub(crate) fn contains(&self, fc: u32) -> bool {
self.lo <= fc && fc < self.hi
}
}
/// A parsed `PlcBte*`: FC breakpoints and the FKP page each range lives in.
pub(crate) struct Bins {
fcs: Vec<u32>,
/// `fcs` ascends, so the bin of an FC holds every FC up to the next breakpoint (a corrupt
/// table that does not gets no cached ranges, [`ChpxRun`]).
sorted: bool,
/// Page number (× 512 = byte offset in the WordDocument stream).
pns: Vec<u32>,
}
impl Bins {
/// Parse the Plc at `table[fc..fc+lcb]` (aFc[n+1] u32s + aPn[n] u32s).
pub(crate) fn parse(table: &[u8], fc: u32, lcb: u32) -> Result<Bins, DocbinError> {
let fc = fc as usize;
let lcb = lcb as usize;
if lcb < 4 || !(lcb - 4).is_multiple_of(8) {
return Ok(Bins { fcs: Vec::new(), sorted: true, pns: Vec::new() });
}
let end = fc.checked_add(lcb).ok_or_else(|| DocbinError::Malformed("bin table overflow".into()))?;
let plc = table.get(fc..end).ok_or_else(|| DocbinError::Malformed("bin table outside the Table stream".into()))?;
let n = (lcb - 4) / 8;
if n > MAX_BINS {
return Err(DocbinError::Limit(format!("bin table has {n} entries")));
}
let at = |i: usize| -> Option<u32> { plc.get(i * 4..i * 4 + 4).map(|b| u32::from_le_bytes([b[0], b[1], b[2], b[3]])) };
let mut fcs = Vec::with_capacity(n + 1);
for i in 0..=n {
fcs.push(at(i).ok_or_else(|| DocbinError::Malformed("truncated bin aFc".into()))?);
}
let mut pns = Vec::with_capacity(n);
for i in 0..n {
// PnBte* keeps the page number in its low bits; the rest must be zero.
pns.push(at(n + 1 + i).ok_or_else(|| DocbinError::Malformed("truncated bin aPn".into()))? & 0x003F_FFFF);
}
let sorted = fcs.windows(2).all(|w| w.first() <= w.get(1));
Ok(Bins { fcs, sorted, pns })
}
/// The page containing `fc`, as a slice of the WordDocument stream, and the FC range
/// `[lo, hi)` of its bin (`hi` is `u32::MAX` for the last; an empty range when the table
/// is not ascending, so nothing is cached from it).
fn page<'w>(&self, word: &'w [u8], fc: u32) -> Option<(&'w [u8], u32, u32)> {
let i = self.fcs.partition_point(|&f| f <= fc).checked_sub(1)?;
let pn = *self.pns.get(i)?;
let at = pn.checked_mul(512)? as usize;
let page = word.get(at..at.checked_add(512)?)?;
if !self.sorted {
return Some((page, 1, 0));
}
let lo = self.fcs.get(i).copied().unwrap_or(0);
let hi = self.fcs.get(i + 1).copied().unwrap_or(u32::MAX);
Some((page, lo, hi))
}
/// The grpprl of the Chpx covering `fc` ([MS-DOC] ChpxFkp: rgfc u32s, rgb offsets ×2,
/// crun in the last byte).
pub(crate) fn chpx<'w>(&self, word: &'w [u8], fc: u32) -> &'w [u8] {
self.chpx_run(word, fc).grpprl
}
/// [`Bins::chpx`] with the FC range over which the same grpprl is the answer, so a walk
/// over consecutive characters looks it up once per run, not once per character.
pub(crate) fn chpx_run<'w>(&self, word: &'w [u8], fc: u32) -> ChpxRun<'w> {
let none = ChpxRun { grpprl: &[], lo: 1, hi: 0 };
let Some((page, page_lo, page_hi)) = self.page(word, fc) else { return none };
let Some(&crun) = page.get(511) else { return none };
let crun = crun.min(0x65) as usize;
let rgfc = |i: usize| -> Option<u32> { page.get(i * 4..i * 4 + 4).map(|b| u32::from_le_bytes([b[0], b[1], b[2], b[3]])) };
// rgb array sits right after rgfc (crun+1 u32s); each byte is a half-offset into the
// region between rgb's end and crun.
let rgb_at = (crun + 1) * 4;
// The last run starting at or before `fc`. It stays the answer for every FC from its own
// start up to the smallest start after it (also in a page whose starts do not ascend).
let j = (0..crun).rev().find(|&j| rgfc(j).is_some_and(|f| f <= fc));
let first_after = j.map_or(0, |j| j + 1);
let hi = (first_after..crun).filter_map(rgfc).min().map_or(page_hi, |m| m.min(page_hi));
let lo = j.and_then(rgfc).map_or(page_lo, |f| f.max(page_lo));
let run = |grpprl: &'w [u8]| ChpxRun { grpprl, lo, hi };
let Some(j) = j else { return run(&[]) };
let Some(&off) = page.get(rgb_at + j) else { return run(&[]) };
if off == 0 {
return run(&[]);
}
let at = off as usize * 2;
let Some(&cb) = page.get(at) else { return run(&[]) };
run(page.get(at + 1..at + 1 + cb as usize).unwrap_or(&[]))
}
/// The (istd, grpprl) of the Papx of the paragraph whose mark is just before `fc`
/// ([MS-DOC] PapxFkp: rgfc u32s — each the FC *following* a paragraph mark —, rgbx BX
/// entries (1 offset byte + 12 PHE bytes), crun in the last byte; the PAPX itself is a
/// PapxInFkp with the cb/cb' size quirk). Query with the FC of the character after the mark.
pub(crate) fn papx<'w>(&self, word: &'w [u8], fc: u32) -> (u16, &'w [u8]) {
let Some((page, _, _)) = self.page(word, fc) else { return (0, &[]) };
let Some(&crun) = page.get(511) else { return (0, &[]) };
// BX is 13 bytes; cap crun at what fits before byte 511.
let max_run = (511usize.saturating_sub((crun as usize + 1) * 4)) / 13;
let crun = (crun as usize).min(max_run);
let rgfc = |i: usize| -> Option<u32> { page.get(i * 4..i * 4 + 4).map(|b| u32::from_le_bytes([b[0], b[1], b[2], b[3]])) };
let rgbx_at = (crun + 1) * 4;
let j = (0..crun).rev().find(|&j| rgfc(j).is_some_and(|f| f <= fc));
let Some(j) = j else { return (0, &[]) };
let Some(&off) = page.get(rgbx_at + j * 13) else { return (0, &[]) };
if off == 0 {
return (0, &[]);
}
let at = off as usize * 2;
let u16at = |o: usize| -> Option<u16> { page.get(o..o + 2).map(|b| u16::from_le_bytes([b[0], b[1]])) };
let Some(cb) = page.get(at).copied() else { return (0, &[]) };
// PapxInFkp: if cb is nonzero the GrpPrlAndIstd (istd u16 + grpprl) is 2×cb-1 bytes;
// if cb is zero, the next byte cb' (1 byte) gives 2×cb' bytes of it.
let (data_at, data_len) = if cb != 0 {
(at + 1, 2 * cb as usize - 1)
} else {
let cb2 = page.get(at + 1).copied().unwrap_or(0) as usize;
(at + 2, 2 * cb2)
};
if data_len < 2 {
return (0, &[]);
}
let istd = u16at(data_at).unwrap_or(0);
(istd, page.get(data_at + 2..data_at + data_len).unwrap_or(&[]))
}
}