mirror of
https://github.com/storytold/vectorcraft.git
synced 2026-11-16 16:42:52 +00:00
Type: a tabbed right-to-left line with negative tracking stops reading after a bounded number of breaks
1042 lines
52 KiB
Rust
1042 lines
52 KiB
Rust
use super::region::Region;
|
||
use super::vertical::stands_upright;
|
||
use super::{
|
||
Affine, BezPath, Breakpoint, Burasagari, EPS, FirstBaseline, FontDb, HyphenationOptions, Kinsoku, KinsokuShoriType, LeadingModel, SGlyph,
|
||
hyphen_glyph, is_cjk, no_line_end, no_line_start, points_for, soft_line_start,
|
||
};
|
||
use crate::justify::is_format_control;
|
||
|
||
/// Vertical metrics of a line: (ascent, descent, leading, cap height, x height).
|
||
#[derive(Clone, Copy, Debug, PartialEq)]
|
||
pub(super) struct Metrics {
|
||
pub(super) asc: f64,
|
||
pub(super) desc: f64,
|
||
pub(super) lead: f64,
|
||
pub(super) cap: f64,
|
||
pub(super) xh: f64,
|
||
/// Height of the top of the ideographic em box above the baseline.
|
||
pub(super) top: f64,
|
||
}
|
||
|
||
impl Metrics {
|
||
pub(super) fn of(g: &SGlyph) -> Self {
|
||
// A warichu glyph's em is the em of the line its block sits in.
|
||
let em = super::vertical::glyph_em(g);
|
||
Self { asc: g.ascent, desc: g.descent, lead: g.leading, cap: g.cap, xh: g.xh, top: (g.face.ideographic_centre() + 0.5) * em }
|
||
}
|
||
pub(super) fn max(g: &[SGlyph]) -> Option<Self> {
|
||
let mut it = g.iter();
|
||
let first = Self::of(it.next()?);
|
||
Some(it.fold(first, |m, g| m.union(Self::of(g))))
|
||
}
|
||
/// The larger of each of the metrics of `self` and `o`: those of a line holding both.
|
||
pub(super) fn union(self, o: Self) -> Self {
|
||
Self {
|
||
asc: self.asc.max(o.asc),
|
||
desc: self.desc.max(o.desc),
|
||
lead: self.lead.max(o.lead),
|
||
cap: self.cap.max(o.cap),
|
||
xh: self.xh.max(o.xh),
|
||
top: self.top.max(o.top),
|
||
}
|
||
}
|
||
/// Distance from the frame top to the first baseline.
|
||
pub(super) fn first_baseline(&self, fb: FirstBaseline, min: f64) -> f64 {
|
||
let v = match fb {
|
||
FirstBaseline::Ascent => self.asc,
|
||
FirstBaseline::CapHeight => self.cap,
|
||
FirstBaseline::XHeight => self.xh,
|
||
FirstBaseline::Leading => self.lead,
|
||
FirstBaseline::Fixed => 0.0,
|
||
};
|
||
v.max(min)
|
||
}
|
||
}
|
||
|
||
/// What a line band was placed below: the baseline of the line above it and the em box top that
|
||
/// line left for the next one (both `None` for a cell's first band), and the space pending between.
|
||
#[derive(Clone, Copy, Default)]
|
||
pub(super) struct Above {
|
||
pub(super) prev: Option<f64>,
|
||
pub(super) next_top: Option<f64>,
|
||
pub(super) pending: f64,
|
||
}
|
||
|
||
/// Where the next line goes: the region being filled and the previous baseline in it.
|
||
#[derive(Clone)]
|
||
pub(super) struct Pen<'r> {
|
||
pub(super) regions: Option<&'r [Region]>,
|
||
pub(super) ri: usize,
|
||
pub(super) prev: Option<f64>,
|
||
pub(super) pending: f64,
|
||
pub(super) fb: FirstBaseline,
|
||
pub(super) fb_min: f64,
|
||
/// Further spans at the current baseline (text wrapping on both sides of an object).
|
||
pub(super) queued: Vec<(f64, f64, f64)>,
|
||
/// What the line band at the current baseline was placed below (see [`Above`]).
|
||
pub(super) above: Above,
|
||
/// Leading measured from em box top to em box top: where the next line's em box top goes.
|
||
pub(super) model: LeadingModel,
|
||
pub(super) next_top: Option<f64>,
|
||
/// The narrowest word no line from here down is wide enough for (see [`Pen::place`]).
|
||
pub(super) no_room_for: f64,
|
||
/// Baseline steps still allowed over all the frame's cells and placements (shared by this pen's
|
||
/// copies): see [`MAX_TOTAL_STEPS`].
|
||
pub(super) steps_left: std::rc::Rc<std::cell::Cell<usize>>,
|
||
/// What is left of the every-line composer's search over the frame's paragraphs (shared by
|
||
/// this pen's copies): see [`super::shaped::Budget`].
|
||
pub(super) search: std::rc::Rc<super::shaped::Budget>,
|
||
}
|
||
|
||
/// The largest number of steps layout takes to move a line down a frame cell to find room for it:
|
||
/// in a cell taller than this many leads, a step is the cell's height divided by this.
|
||
const MAX_STEPS: usize = 100_000;
|
||
|
||
/// The most baseline steps one frame's layout takes in all, over every cell: a cell may take
|
||
/// [`MAX_STEPS`], but rows and columns of tall, narrow-shaped cells would multiply that.
|
||
pub(super) const MAX_TOTAL_STEPS: usize = 10_000_000;
|
||
|
||
impl Pen<'_> {
|
||
/// Baseline and horizontal span for a line with estimated metrics `est` and indents that starts
|
||
/// with a word `word` wide; `None` = the frame is full (overflow).
|
||
/// In a frame that isn't a rectangle, lines too short for the word are left empty and it moves
|
||
/// down to the first one wide enough. Only when no line is does it go on the first line with
|
||
/// room for any text, and get broken there.
|
||
/// The `bool` is true for a further span at the previous line's baseline.
|
||
pub(super) fn place(&mut self, est: Metrics, ind_l: f64, ind_r: f64, word: f64) -> Option<(f64, f64, f64, bool)> {
|
||
if word > 0.0 && word < self.no_room_for {
|
||
let mut pen = self.clone();
|
||
if let Some(placed) = pen.place_at_least(est, ind_l, ind_r, word) {
|
||
*self = pen;
|
||
return Some(placed);
|
||
}
|
||
// Lines further down are no wider for this word: don't search for it, or a wider one, again.
|
||
self.no_room_for = word;
|
||
}
|
||
self.place_at_least(est, ind_l, ind_r, 0.0)
|
||
}
|
||
|
||
/// [`Pen::place`], skipping lines of a non-rectangular frame narrower than `min`.
|
||
fn place_at_least(&mut self, est: Metrics, ind_l: f64, ind_r: f64, min: f64) -> Option<(f64, f64, f64, bool)> {
|
||
let Some(regions) = self.regions else {
|
||
let b = self.prev.map_or(0.0, |b| self.next_baseline(b, self.next_top, self.pending, est));
|
||
return Some((b, f64::NEG_INFINITY, f64::INFINITY, false));
|
||
};
|
||
// A further span was measured across the height of the line set before it on its baseline.
|
||
// This line gets the part of it that is free across its own height: a taller line can
|
||
// reach a wrap object the line before it stayed clear of. A span with no such part is
|
||
// left empty, and so is a part the line's height made narrower than `min`. A span on a
|
||
// baseline nearer the line above the band than this line's own leading sets it is left
|
||
// empty too: a taller line would reach into that line.
|
||
while !self.queued.is_empty() {
|
||
let (b, x0, x1) = self.queued.remove(0);
|
||
let Some(r) = regions.get(self.ri) else { return Some((b, x0, x1, true)) };
|
||
if b + EPS < self.band_baseline(r, est) {
|
||
continue;
|
||
}
|
||
let mut free = r.spans(b - est.asc, b + est.desc).into_iter().filter_map(|(s, e)| {
|
||
let (s, e) = (s.max(x0), e.min(x1));
|
||
let room = e - s - ind_l - ind_r;
|
||
let narrowed = s > x0 + EPS || e < x1 - EPS;
|
||
(room > est.asc.max(1.0) && (!narrowed || room + EPS >= min)).then_some((s, e))
|
||
});
|
||
let Some((s, e)) = free.next() else { continue };
|
||
// An object inside the span splits it: the parts after the first stay queued.
|
||
let rest: Vec<_> = free.map(|(s, e)| (b, s, e)).collect();
|
||
self.queued.splice(0..0, rest);
|
||
return Some((b, s, e, true));
|
||
}
|
||
loop {
|
||
let r = regions.get(self.ri)?;
|
||
self.above = Above { prev: self.prev, next_top: self.next_top, pending: self.pending };
|
||
let mut baseline = self.band_baseline(r, est);
|
||
// The baseline moves down the cell a lead at a time, or a `MAX_STEPS`th of the cell's
|
||
// height when that is more, in at most `MAX_STEPS` steps.
|
||
let step = est.lead.max(1.0).max(r.cell.height() / MAX_STEPS as f64);
|
||
for _ in 0..=MAX_STEPS {
|
||
let left = self.steps_left.get();
|
||
if left == 0 {
|
||
// Out of steps: the frame is full as far as this text goes.
|
||
return None;
|
||
}
|
||
self.steps_left.set(left - 1);
|
||
// A baseline a step no longer moves (past a gutter of 1e300, say) won't reach a line.
|
||
if baseline + est.desc > r.bottom() + 0.01 || baseline + step <= baseline || baseline.is_nan() {
|
||
break;
|
||
}
|
||
// A rectangle's lines are all as wide: none below has more room for the word
|
||
// (until wrap shapes can narrow the first baseline: they stay bounded).
|
||
let constant = (r.rect || r.polys.is_empty()) && r.wraps.is_empty();
|
||
let min = if constant { 0.0 } else { min };
|
||
let fits = |&(a, b): &(f64, f64)| {
|
||
let room = b - a - ind_l - ind_r;
|
||
room > est.asc.max(1.0) && room + EPS >= min
|
||
};
|
||
if r.wraps.is_empty() {
|
||
match r.span(baseline - est.asc, baseline + est.desc) {
|
||
Some(s) if fits(&s) => return Some((baseline, s.0, s.1, false)),
|
||
// Too narrow here is too narrow all the way down the cell: next cell.
|
||
Some(_) if constant => break,
|
||
_ => baseline += step,
|
||
}
|
||
} else {
|
||
// Wrap objects can split a line: fill every span, left to right.
|
||
let mut spans: Vec<(f64, f64)> = r.spans(baseline - est.asc, baseline + est.desc).into_iter().filter(fits).collect();
|
||
if spans.is_empty() {
|
||
baseline += step;
|
||
continue;
|
||
}
|
||
let (a, b) = spans.remove(0);
|
||
self.queued = spans.into_iter().map(|(a, b)| (baseline, a, b)).collect();
|
||
return Some((baseline, a, b, false));
|
||
}
|
||
}
|
||
// Next row/column.
|
||
self.ri += 1;
|
||
self.prev = None;
|
||
self.pending = 0.0;
|
||
self.queued.clear();
|
||
}
|
||
}
|
||
pub(super) fn bottom(&self) -> f64 {
|
||
self.regions.and_then(|r| r.get(self.ri)).map_or(f64::INFINITY, |r| r.bottom())
|
||
}
|
||
/// The baseline of the line after the one at `prev`, which left the next em box top at
|
||
/// `next_top`, for a line of metrics `est` with `pending` space between them.
|
||
fn next_baseline(&self, prev: f64, next_top: Option<f64>, pending: f64, est: Metrics) -> f64 {
|
||
match (self.model, next_top) {
|
||
(LeadingModel::EmBoxTop, Some(top)) => top + est.top + pending,
|
||
_ => prev + est.lead + pending,
|
||
}
|
||
}
|
||
/// The highest baseline in cell `r` for a line of metrics `est` in the current line band: as
|
||
/// far below the line above the band ([`Pen::above`]) as its own leading sets it, or at the
|
||
/// cell's first baseline.
|
||
fn band_baseline(&self, r: &Region, est: Metrics) -> f64 {
|
||
let Above { prev, next_top, pending } = self.above;
|
||
match prev {
|
||
None if self.model == LeadingModel::EmBoxTop => r.top() + r.shift + est.top.max(self.fb_min),
|
||
None => r.top() + r.shift + est.first_baseline(self.fb, self.fb_min),
|
||
Some(b) => self.next_baseline(b, next_top, pending, est) + r.gap,
|
||
}
|
||
}
|
||
/// A line of metrics `m` set in a further span at baseline `b` is no nearer the line above
|
||
/// the band than its own leading sets it.
|
||
pub(super) fn clears_above(&self, b: f64, m: Metrics) -> bool {
|
||
self.regions.and_then(|r| r.get(self.ri)).is_none_or(|r| b + EPS >= self.band_baseline(r, m))
|
||
}
|
||
/// A line was set at `baseline` with metrics `m`: where the next one goes from.
|
||
pub(super) fn settled(&mut self, baseline: f64, m: Metrics) {
|
||
self.prev = Some(baseline);
|
||
self.next_top = Some(baseline - m.top + m.lead);
|
||
}
|
||
}
|
||
|
||
/// Advance of a tab at pen position `x`: up to the next stop after it (explicit stops first, then
|
||
/// every ½ inch), and the leader of that stop (default stops have none). Right, centre and decimal
|
||
/// stops align the text drawn after the tab (`rest`, in drawing order, up to the next tab) on the
|
||
/// stop.
|
||
pub(super) fn tab_advance<'t, 'g>(
|
||
tabs: &'t [vectorcraft_doc::TabStop],
|
||
origin: f64,
|
||
x: f64,
|
||
rest: impl IntoIterator<Item = &'g SGlyph>,
|
||
) -> (f64, &'t str) {
|
||
use vectorcraft_doc::TabAlign;
|
||
let rel = x - origin;
|
||
// The text after the tab is read only when a right, center or decimal stop is reached, since
|
||
// only those align it. A tab to a default or left stop reads none of it.
|
||
let mut rest = Some(rest);
|
||
let mut seg: Vec<&SGlyph> = Vec::new();
|
||
let mut width = 0.0;
|
||
let mut explicit = None;
|
||
for t in tabs {
|
||
if t.align != TabAlign::Left
|
||
&& let Some(r) = rest.take()
|
||
{
|
||
seg = r.into_iter().take_while(|g| g.ch != '\t' && g.ch != '\u{2028}').collect();
|
||
width = seg.iter().map(|g| g.adv).sum();
|
||
}
|
||
let at = match t.align {
|
||
TabAlign::Left => t.position,
|
||
TabAlign::Right => t.position - width,
|
||
TabAlign::Center => t.position - width / 2.0,
|
||
TabAlign::Decimal => t.position - seg.iter().take_while(|g| g.ch != t.align_on).map(|g| g.adv).sum::<f64>(),
|
||
};
|
||
if at > rel + 1e-6 {
|
||
explicit = Some((at, t.leader.as_str()));
|
||
break;
|
||
}
|
||
}
|
||
let (at, leader) = explicit.unwrap_or_else(|| {
|
||
let last = tabs.iter().map(|t| t.position).fold(0.0, f64::max).max(0.0);
|
||
let base = rel.max(last);
|
||
(((base / vectorcraft_doc::text::DEFAULT_TAB_INTERVAL).floor() + 1.0) * vectorcraft_doc::text::DEFAULT_TAB_INTERVAL, "")
|
||
});
|
||
((at - rel).max(0.0), leader)
|
||
}
|
||
|
||
/// The glyphs of `sg` drawn after position `k` of `drawn` (indices into `sg`, in drawing order).
|
||
pub(super) fn drawn_after_tab<'g>(sg: &'g [SGlyph], drawn: &'g [usize], k: usize) -> impl Iterator<Item = &'g SGlyph> {
|
||
drawn.iter().skip(k + 1).filter_map(move |&j| sg.get(j))
|
||
}
|
||
|
||
/// The width of the glyphs of `sg` at `drawn` (indices into `sg`, in the order the line draws
|
||
/// them) set on a line whose content starts `x_start` from the tab origin: tabs of the text take
|
||
/// what [`tab_advance`] gives them from where the glyphs drawn before them end, aligning the glyphs
|
||
/// drawn after them, not the width of their glyph. The tab of a list marker (length 0) keeps its
|
||
/// own width. With `end_trim` (a glyph index and an amount), that glyph is set that much narrower
|
||
/// where it is drawn (a line's end trim), before the tabs drawn after it take their stops.
|
||
pub(super) fn drawn_width(sg: &[SGlyph], drawn: &[usize], tabs: &[vectorcraft_doc::TabStop], x_start: f64, end_trim: Option<(usize, f64)>) -> f64 {
|
||
let mut x = x_start;
|
||
for (k, j, g) in drawn.iter().enumerate().filter_map(|(k, &j)| sg.get(j).map(|g| (k, j, g))) {
|
||
x += if g.ch == '\t' && g.len > 0 {
|
||
tab_advance(tabs, 0.0, x, drawn_after_tab(sg, drawn, k)).0
|
||
} else {
|
||
g.adv - end_trim.filter(|&(at, _)| at == j).map_or(0.0, |(_, trim)| trim)
|
||
};
|
||
}
|
||
x - x_start
|
||
}
|
||
|
||
/// Most breaks, or hyphenation points, [`break_line`] measures as drawn one by one from the latest
|
||
/// back before bisecting the earlier ones.
|
||
const MAX_DRAWN_CHECKS: usize = 256;
|
||
|
||
/// Most breaks, or hyphenation points, [`break_line`] measures as drawn one by one when a glyph of
|
||
/// the line has a negative advance: a later break can then be drawn narrower than an earlier one,
|
||
/// and bisecting can miss the latest break that fits. It is also the most legal breaks such a line
|
||
/// reads past where it first overflows in the order of its text.
|
||
const MAX_DRAWN_CHECKS_STEPPING_BACK: usize = 1024;
|
||
|
||
/// Most copies of a leader drawn in one tab: a tiny type size and a far stop must not make
|
||
/// millions of outlines.
|
||
const MAX_LEADER_COPIES: f64 = 4096.0;
|
||
|
||
/// The outline of a tab leader: `leader`'s characters, in the font and size of tab `g`, repeated
|
||
/// across the tab's gap `from..to` on baseline `y`. Copies sit on a grid of the leader's width
|
||
/// from `origin` (where tab stops are measured from), so the leaders of lines above one another
|
||
/// line up; only whole copies are drawn. Characters the font lacks draw nothing.
|
||
pub(super) fn leader_outline(db: &FontDb, g: &SGlyph, leader: &str, origin: f64, from: f64, to: f64, y: f64) -> BezPath {
|
||
let mut out = BezPath::new();
|
||
let glyphs: Vec<(u32, f64)> = leader
|
||
.chars()
|
||
.filter(|c| !c.is_control())
|
||
.map(|c| {
|
||
let gid = g.face.glyph_for(c);
|
||
(gid, g.face.advance(gid) * g.sx)
|
||
})
|
||
.collect();
|
||
let width: f64 = glyphs.iter().map(|(_, a)| a).sum();
|
||
if !(width > EPS && width.is_finite() && from.is_finite() && to.is_finite()) {
|
||
return out;
|
||
}
|
||
let first = ((from - origin) / width - EPS).ceil();
|
||
let copies = (((to - origin) / width + EPS).floor() - first).clamp(0.0, MAX_LEADER_COPIES) as usize;
|
||
let local = Affine::translate((0.0, -g.bshift)) * Affine::rotate(-g.rotation.to_radians()) * Affine::scale_non_uniform(g.sx, g.sy);
|
||
for k in 0..copies {
|
||
let mut pen = origin + (first + k as f64) * width;
|
||
for &(gid, adv) in &glyphs {
|
||
if gid != 0 {
|
||
let m = Affine::translate((pen, y)) * local;
|
||
out.extend(db.outline(&g.face, gid).elements().iter().map(|el| m * *el));
|
||
}
|
||
pen += adv;
|
||
}
|
||
}
|
||
out
|
||
}
|
||
|
||
/// Width of the word starting at glyph `i`: its advances up to the first place a line may break
|
||
/// (spaces left out, the hyphen of a soft hyphen in), the least room a line needs not to break
|
||
/// inside it. As in [`break_line`], kinsoku keeps a character that may not end a line with the one
|
||
/// after it, a No Break run (its spaces inside it included) is one word up to a forced line break,
|
||
/// and a cluster is never split. A soft hyphen just before a forced line break adds no hyphen. A list marker at `i` (glyphs of length 0) counts whole, with the word after it. A word
|
||
/// that ends where a line may end is measured as the line is set: without the spaces at its end,
|
||
/// and with its last glyph as narrow as `line_end` sets it; one that ends at a soft hyphen or past
|
||
/// the measured glyphs is not.
|
||
pub(super) fn word_width(g: &[SGlyph], i: usize, line_end: &[f64], kinsoku: Kinsoku) -> f64 {
|
||
/// Glyphs measured at most: a word wrapped over many lines is measured again from each line's
|
||
/// start, and one this long is broken inside whatever cell it meets.
|
||
const MAX_GLYPHS: usize = 1024;
|
||
let rest = g.get(i..).unwrap_or_default();
|
||
let marker = rest.iter().take_while(|gl| gl.len == 0).count();
|
||
let mut w: f64 = rest.iter().take(marker).map(|gl| gl.adv).sum();
|
||
let start = i + marker;
|
||
let mut end = start;
|
||
// The spaces at the end of what is measured so far (inside a No Break run), and the end of its
|
||
// last glyph that is not a space.
|
||
let (mut trailing, mut ink) = (0.0, start);
|
||
let set = |w: f64, trailing: f64, ink: usize| w - trailing - line_end_credit(g, line_end, ink);
|
||
while let Some(gl) = g.get(end).filter(|_| end - start < MAX_GLYPHS) {
|
||
// A forced line break ends the line, inside a No Break run too, as in [`break_line`].
|
||
if gl.ch == '\u{2028}' {
|
||
return set(w, trailing, ink);
|
||
}
|
||
if gl.is_space() && (end == start || !splits_no_break(g, end)) {
|
||
return set(w, trailing, ink);
|
||
}
|
||
let forced_next = g.get(end + 1).is_some_and(|n| n.ch == '\u{2028}');
|
||
if gl.is_soft_hyphen() && !forced_next && kinsoku_allows(g, end, kinsoku) && !splits_no_break(g, end + 1) {
|
||
return w + hyphen_glyph(gl).adv;
|
||
}
|
||
w += gl.adv;
|
||
end += 1;
|
||
if gl.is_space() {
|
||
trailing += gl.adv;
|
||
} else {
|
||
(trailing, ink) = (0.0, end);
|
||
}
|
||
let cluster_ends = g.get(end).is_none_or(|next| next.byte != gl.byte);
|
||
if gl.break_after() && cluster_ends && kinsoku_allows(g, end - 1, kinsoku) && !splits_no_break(g, end) {
|
||
return set(w, trailing, ink);
|
||
}
|
||
}
|
||
if end >= g.len() { set(w, trailing, ink) } else { w }
|
||
}
|
||
|
||
/// Greedy break: returns (end glyph index, hyphenated) for a line starting at `i` of `width`.
|
||
/// With burasagari, a comma or full stop that doesn't fit ends the line, hanging outside it.
|
||
/// `hyphenate` holds the paragraph's Hyphenation options and the Hyphenation Zone in force when
|
||
/// the line may hyphenate a word; a line whose text before the overflowing word already ends
|
||
/// within the zone breaks before that word. A tab is as wide as it is set: up to its next stop
|
||
/// (`tabs`), with the line's content starting `x_start` from the tab origin. Where kinsoku alone
|
||
/// keeps the line from breaking before the character that overflows, `kinsoku_type` decides
|
||
/// whether the line pushes characters in ([`push_in`]) or out; `hanging` is true when Optical
|
||
/// Margin Alignment or Roman Hanging Punctuation is on, and `vertical` in vertical type.
|
||
/// `line_end[k]` is how much narrower a line ending with glyph `k` is set (Mojikumi); a cluster
|
||
/// that fits only that narrower stays on the line when the line may end after it (empty: none).
|
||
/// `fits`, for a line drawn in another order than its text's (right to left), is whether the line
|
||
/// from `i` ending before glyph `end` (with a hyphen when `hyphenated`) fits `width` as it is drawn:
|
||
/// once the line holds a tab, whose width depends on what is drawn before it, the line reads on
|
||
/// while the least it can be drawn at fits `width`, then ends at the latest legal break where it
|
||
/// fits as drawn (past [`MAX_DRAWN_CHECKS`] breaks back from the latest, or
|
||
/// [`MAX_DRAWN_CHECKS_STEPPING_BACK`] when a glyph of the line has a negative advance, the earlier
|
||
/// ones without a hyphen are bisected); hyphenation points are measured the same way. Once such a
|
||
/// line holds a glyph with a negative advance and has been wider than `width` in the order of its
|
||
/// text, it reads at most [`MAX_DRAWN_CHECKS_STEPPING_BACK`] legal breaks further, and a forced
|
||
/// line break right after the last of them, and ends at the latest of the breaks read that fits as
|
||
/// drawn.
|
||
#[allow(clippy::too_many_arguments)]
|
||
pub(super) fn break_line(
|
||
text: &str,
|
||
g: &[SGlyph],
|
||
i: usize,
|
||
width: f64,
|
||
tabs: &[vectorcraft_doc::TabStop],
|
||
x_start: f64,
|
||
hyphenate: Option<(&HyphenationOptions, f64)>,
|
||
burasagari: Burasagari,
|
||
kinsoku: Kinsoku,
|
||
kinsoku_type: KinsokuShoriType,
|
||
hanging: bool,
|
||
vertical: bool,
|
||
line_end: &[f64],
|
||
fits: Option<&dyn Fn(usize, bool) -> bool>,
|
||
) -> (usize, bool) {
|
||
// The line ending at `chosen`, once it holds a tab, when it fits as drawn; otherwise the latest
|
||
// of the legal `breaks` before it that fits as drawn, measuring at most `checks` of them from
|
||
// the latest back and bisecting the earlier ones without a hyphen; the first of them when none
|
||
// of those measured fits.
|
||
let settle = |tabbed: bool, checks: usize, chosen: (usize, bool), breaks: &[(usize, bool)]| -> (usize, bool) {
|
||
let Some(fits) = fits.filter(|_| tabbed) else { return chosen };
|
||
if fits(chosen.0, chosen.1) {
|
||
return chosen;
|
||
}
|
||
let earlier = breaks.get(breaks.partition_point(|b| b.0 <= i)..breaks.partition_point(|b| b.0 < chosen.0)).unwrap_or_default();
|
||
if let Some(&b) = earlier.iter().rev().take(checks).find(|b| fits(b.0, b.1)) {
|
||
return b;
|
||
}
|
||
// Without a hyphen, a line that ends later is drawn no narrower, so those breaks bisect.
|
||
let rest = earlier.get(..earlier.len().saturating_sub(checks)).unwrap_or_default();
|
||
let plain: Vec<&(usize, bool)> = rest.iter().filter(|b| !b.1).collect();
|
||
let fitting = plain.partition_point(|b| fits(b.0, b.1));
|
||
fitting.checked_sub(1).and_then(|k| plain.get(k).copied()).or(earlier.first()).copied().unwrap_or(chosen)
|
||
};
|
||
// How many breaks a line measures as drawn one by one.
|
||
let checks = |back: f64| if back > 0.0 { MAX_DRAWN_CHECKS_STEPPING_BACK } else { MAX_DRAWN_CHECKS };
|
||
let mut breaks: Vec<(usize, bool)> = Vec::new();
|
||
let mut tabbed = false;
|
||
if !width.is_finite() {
|
||
if let Some(pos) = g[i..].iter().position(|gl| gl.ch == '\u{2028}') {
|
||
return (i + pos + 1, false);
|
||
}
|
||
return (g.len(), false);
|
||
}
|
||
let mut x = 0.0;
|
||
// The width of the line's glyphs other than its tabs, how far those with a negative advance step
|
||
// back in all, and how many tabs it holds.
|
||
let mut ink = 0.0;
|
||
let mut back = 0.0;
|
||
let mut tabs_read = 0usize;
|
||
// The `n`th of the stops a tab can go to, in order: the explicit stops, then every default
|
||
// stop after the last of them (0 when `n` is 0).
|
||
let mut explicit: Vec<f64> = if fits.is_some() { tabs.iter().map(|t| t.position).filter(|p| p.is_finite()).collect() } else { Vec::new() };
|
||
explicit.sort_by(f64::total_cmp);
|
||
explicit.dedup();
|
||
let last_explicit = tabs.iter().map(|t| t.position).fold(0.0, f64::max).max(0.0);
|
||
let nth_stop = |n: usize| -> f64 {
|
||
use vectorcraft_doc::text::DEFAULT_TAB_INTERVAL;
|
||
match n.checked_sub(1) {
|
||
None => 0.0,
|
||
Some(k) => explicit
|
||
.get(k)
|
||
.copied()
|
||
.unwrap_or_else(|| ((last_explicit / DEFAULT_TAB_INTERVAL).floor() + (n - explicit.len()) as f64) * DEFAULT_TAB_INTERVAL),
|
||
}
|
||
};
|
||
let mut last_break: Option<(usize, bool)> = None;
|
||
// Whether the line, once it holds a tab and is measured as drawn, has been wider in the order of
|
||
// its text than `width`, and how many legal breaks it has read since then with a glyph stepping
|
||
// back: past MAX_DRAWN_CHECKS_STEPPING_BACK of them the line stops reading.
|
||
let mut overflowed = false;
|
||
let mut breaks_past_overflow = 0usize;
|
||
// A list marker (length 0) is a cluster of its own; the line holds it and at least the first
|
||
// cluster after it.
|
||
let first = i + g.get(i..).unwrap_or_default().iter().take_while(|h| h.len == 0).count();
|
||
let mut j = i;
|
||
while j < g.len() {
|
||
let gl = &g[j];
|
||
let mut cluster_end = j + 1;
|
||
while g.get(cluster_end).is_some_and(|next| next.byte == gl.byte && (next.len == 0) == (gl.len == 0)) {
|
||
cluster_end += 1;
|
||
}
|
||
// No Break: the marked characters stay on one line, including across a space inside the run;
|
||
// Bunri-Kinshi keeps an inseparable pair, or a chain of them, on one line too.
|
||
if gl.ch != '\u{2028}' {
|
||
while let Some(n) = g.get(cluster_end) {
|
||
if !splits_no_break(g, cluster_end) || n.ch == '\u{2028}' {
|
||
break;
|
||
}
|
||
let b = n.byte;
|
||
cluster_end += 1;
|
||
while g.get(cluster_end).is_some_and(|next| next.byte == b) {
|
||
cluster_end += 1;
|
||
}
|
||
}
|
||
}
|
||
// A list marker's glyphs, its tab included, keep their own widths.
|
||
let is_tab = gl.ch == '\t' && gl.len > 0;
|
||
let cluster_width: f64 = if is_tab {
|
||
tabbed = true;
|
||
tabs_read += 1;
|
||
tab_advance(tabs, 0.0, x_start + x, g.get(cluster_end..).unwrap_or_default()).0
|
||
} else {
|
||
g[j..cluster_end].iter().map(|glyph| glyph.adv).sum()
|
||
};
|
||
let cluster_ink = if is_tab { 0.0 } else { cluster_width };
|
||
let cluster_back: f64 = if is_tab { 0.0 } else { g[j..cluster_end].iter().map(|glyph| (-glyph.adv).max(0.0)).sum() };
|
||
if gl.ch == '\u{2028}' {
|
||
return settle(tabbed, checks(back), (cluster_end, false), &breaks);
|
||
}
|
||
// Measured as the line is set when it ends with this cluster: only the cluster's last glyph
|
||
// is narrowed, and only where the line may end.
|
||
let end_credit =
|
||
Some(line_end_credit(g, line_end, cluster_end)).filter(|&c| c > 0.0 && may_end_after(g, cluster_end, kinsoku)).unwrap_or(0.0);
|
||
// Once the line holds a tab and is measured as drawn, it reads on while the least it can be
|
||
// drawn at fits, and where it ends is settled then. That least is its glyphs other than the
|
||
// tabs and, while no glyph steps back, the stop its tabs reach: each tab then goes to a later
|
||
// stop than the one before and the text aligned on it ends at or past that stop, so the
|
||
// line's tabs, all before this cluster, reach at least the stop as many places along as
|
||
// there are tabs.
|
||
let stops_reached = if back + cluster_back > 0.0 { f64::NEG_INFINITY } else { nth_stop(tabs_read) - x_start };
|
||
let measured = if tabbed && fits.is_some() { (ink + cluster_ink).max(stops_reached) } else { x + cluster_width };
|
||
overflowed |= tabbed && fits.is_some() && j > first && !gl.is_space() && x + cluster_width - end_credit > width + EPS;
|
||
if j > first && !gl.is_space() && measured - end_credit > width + EPS {
|
||
// It ends the line with the spaces after it.
|
||
if burasagari != Burasagari::None && hangs(gl) && kinsoku_allows(g, cluster_end - 1, kinsoku) {
|
||
let mut end = cluster_end;
|
||
while g.get(end).is_some_and(SGlyph::is_space) {
|
||
end += 1;
|
||
}
|
||
return (end, false);
|
||
}
|
||
let pushed_out = last_break.map(|b| b.0);
|
||
if let Some(end) = push_in(g, i, j, cluster_end, x + cluster_width, width, pushed_out, kinsoku, kinsoku_type, hanging, vertical, line_end)
|
||
{
|
||
return (end, false);
|
||
}
|
||
break;
|
||
}
|
||
x += cluster_width;
|
||
ink += cluster_ink;
|
||
back += cluster_back;
|
||
let last = &g[cluster_end - 1];
|
||
if last.break_after() && last.len > 0 && kinsoku_allows(g, cluster_end - 1, kinsoku) && !splits_no_break(g, cluster_end) {
|
||
// Once the line holds a tab, whether a break fits as drawn is settled when the line ends.
|
||
let hyphenated = last.is_soft_hyphen();
|
||
if !hyphenated || (tabbed && fits.is_some()) || x + hyphen_glyph(last).adv <= width + EPS {
|
||
last_break = Some((cluster_end, hyphenated));
|
||
if fits.is_some() {
|
||
breaks.push((cluster_end, hyphenated));
|
||
}
|
||
// A line with a glyph stepping back stops reading MAX_DRAWN_CHECKS_STEPPING_BACK
|
||
// breaks past where it first overflowed in the order of its text, and ends at the
|
||
// latest break read that fits as drawn: each of those breaks is measured one by one.
|
||
// A soft hyphen ending the text draws no hyphen there, as at the end of the text below.
|
||
// A break right before a forced line break reads on to it: the line ends there.
|
||
breaks_past_overflow += usize::from(overflowed && back > 0.0);
|
||
if breaks_past_overflow >= MAX_DRAWN_CHECKS_STEPPING_BACK && g.get(cluster_end).is_none_or(|n| n.ch != '\u{2028}') {
|
||
return settle(tabbed, checks(back), (cluster_end, hyphenated && cluster_end < g.len()), &breaks);
|
||
}
|
||
}
|
||
}
|
||
j = cluster_end;
|
||
}
|
||
if j >= g.len() {
|
||
return settle(tabbed, checks(back), (g.len(), false), &breaks);
|
||
}
|
||
// Hyphenate the word that overflows. A list marker starting the line is not part of a word.
|
||
let word_start = last_break.map_or(first, |b| b.0);
|
||
let in_zone = |zone: f64| {
|
||
let mut te = word_start;
|
||
while te > first && g[te - 1].is_space() {
|
||
te -= 1;
|
||
}
|
||
te > first && zone > 0.0 && width - g[i..te].iter().map(|g| g.adv).sum::<f64>() <= zone + EPS
|
||
};
|
||
if let Some((opts, zone)) = hyphenate
|
||
&& g[j].is_letter()
|
||
&& !in_zone(zone)
|
||
{
|
||
let mut we = j;
|
||
while we < g.len() && !g[we].is_space() && !g[we].break_after() {
|
||
we += 1;
|
||
}
|
||
let x_ws: f64 = g[i..word_start].iter().map(|g| g.adv).sum();
|
||
let pts: Vec<usize> = hyphen_breaks(text, g, word_start, we, opts)
|
||
.into_iter()
|
||
.filter(|&k| k > 0 && kinsoku_allows(g, k - 1, kinsoku) && !splits_no_break(g, k))
|
||
.collect();
|
||
// The points the line may end at, the latest first. Measured as drawn, at most
|
||
// `checks(back)` of them are measured one by one, and the earlier ones are bisected:
|
||
// each ends with a hyphen, so a later one is drawn no narrower.
|
||
let points: Vec<usize> = pts.iter().rev().copied().filter(|&k| k <= j && k > i).collect();
|
||
let drawn = fits.filter(|_| tabbed);
|
||
let one_by_one = checks(back);
|
||
for &k in points.iter().take(if drawn.is_some() { one_by_one } else { usize::MAX }) {
|
||
let w = x_ws + g[word_start..k].iter().map(|g| g.adv).sum::<f64>() + hyphen_glyph(&g[k - 1]).adv;
|
||
if drawn.map_or(w <= width + EPS, |fits| fits(k, true)) {
|
||
return (k, true);
|
||
}
|
||
}
|
||
if let Some(fits) = drawn {
|
||
let earlier: Vec<usize> = points.iter().skip(one_by_one).rev().copied().collect();
|
||
let fitting = earlier.partition_point(|&k| fits(k, true));
|
||
if let Some(&k) = fitting.checked_sub(1).and_then(|n| earlier.get(n)) {
|
||
return (k, true);
|
||
}
|
||
}
|
||
}
|
||
// Break candidates and greedy overflow are cluster boundaries; hyphenation also returns only
|
||
// cluster starts, so the resulting end always keeps each source cluster together.
|
||
match last_break {
|
||
Some(b) if b.0 > i => settle(tabbed, checks(back), b, &breaks),
|
||
_ => (j, false),
|
||
}
|
||
}
|
||
|
||
/// Can glyph `g` hang outside the line (burasagari)? An East Asian comma or full stop, full width
|
||
/// (、。,.) or half width (、。); not a closing bracket, nor Latin punctuation (Latin text keeps
|
||
/// its line breaks and composer). Never an inline graphic (U+FFFC), nor a character of a
|
||
/// tate-chu-yoko or warichu block.
|
||
pub(super) fn hangs(g: &SGlyph) -> bool {
|
||
g.inline.is_none() && matches!(g.ch, '、' | '。' | ',' | '.' | '、' | '。') && g.tcy.is_none() && g.wari.is_none()
|
||
}
|
||
|
||
/// Optical Margin Alignment: the share of glyph `g`'s advance that hangs outside the margin when
|
||
/// `g` starts or ends a line. Quotation marks, hyphens, full stops and commas hang whole, and
|
||
/// asterisks, tildes, ellipses, en and em dashes, colons and semicolons hang by half (the amounts
|
||
/// published for Roman Hanging Punctuation). A, V, W, Y, v, w and y hang by a tenth of their
|
||
/// advance, VectorCraft's own amount for their slanted sides. Other characters, East Asian
|
||
/// punctuation (which burasagari hangs), inline graphics and the characters of a warichu block
|
||
/// don't hang.
|
||
pub(super) fn optical_share(g: &SGlyph) -> f64 {
|
||
if g.inline.is_some() || g.tcy.is_some() || g.wari.is_some() {
|
||
return 0.0;
|
||
}
|
||
match g.ch {
|
||
'-' | '\u{2010}' | '\u{2011}' | '.' | ',' => 1.0,
|
||
c if is_quote(c) => 1.0,
|
||
'*' | '~' | '…' | '–' | '—' | ':' | ';' => 0.5,
|
||
'A' | 'V' | 'W' | 'Y' | 'v' | 'w' | 'y' => 0.1,
|
||
_ => 0.0,
|
||
}
|
||
}
|
||
|
||
/// Roman Hanging Punctuation: the share of glyph `g`'s advance that hangs outside the margin when
|
||
/// `g` starts or ends a line: [`optical_share`]'s punctuation, and no letters.
|
||
pub(super) fn roman_share(g: &SGlyph) -> f64 {
|
||
if g.ch.is_alphabetic() { 0.0 } else { optical_share(g) }
|
||
}
|
||
|
||
/// A punctuation mark that hangs ([`optical_share`]), not a letter.
|
||
pub(super) fn optical_punct(g: &SGlyph) -> bool {
|
||
!g.ch.is_alphabetic() && optical_share(g) > 0.0
|
||
}
|
||
|
||
/// A single or double quotation mark: straight, curly, low-9 or reversed-9.
|
||
fn is_quote(c: char) -> bool {
|
||
matches!(c, '\'' | '"' | '‘' | '’' | '‚' | '‛' | '“' | '”' | '„' | '‟')
|
||
}
|
||
|
||
/// Optical Margin Alignment and Roman Hanging Punctuation: how far a line hangs outside its left
|
||
/// and right margins, from the glyph at each end of `line` (its glyphs in visual order, each with
|
||
/// the advance it is set with) and its `share` ([`optical_share`] or [`roman_share`]), and how many glyphs hang at each end → (left, right, left glyphs,
|
||
/// right glyphs). A punctuation mark followed by a quotation mark hangs together with it: `."`
|
||
/// ending a line and `'"` starting one hang both their glyphs. A line of one glyph hangs at
|
||
/// neither end.
|
||
pub(super) fn optical_hangs(line: &[(&SGlyph, f64)], share: fn(&SGlyph) -> f64) -> (f64, f64, usize, usize) {
|
||
let n = line.len();
|
||
if n < 2 {
|
||
return (0.0, 0.0, 0, 0);
|
||
}
|
||
let hang = |k: usize| line.get(k).map_or(0.0, |&(g, adv)| share(g) * adv.max(0.0));
|
||
let punct = |k: usize| line.get(k).is_some_and(|&(g, _)| optical_punct(g));
|
||
let quote = |k: usize| line.get(k).is_some_and(|&(g, _)| is_quote(g.ch) && share(g) > 0.0);
|
||
let mut left = hang(0);
|
||
// The glyphs that hang at the left; the right end takes none of them.
|
||
let mut taken = usize::from(left > 0.0);
|
||
if left > 0.0 && n > 2 && punct(0) && quote(1) {
|
||
left += hang(1);
|
||
taken = 2;
|
||
}
|
||
let mut right = hang(n - 1);
|
||
let mut right_glyphs = usize::from(right > 0.0);
|
||
if right > 0.0 && n > 2 && n - 2 >= taken && punct(n - 2) && quote(n - 1) {
|
||
right += hang(n - 2);
|
||
right_glyphs = 2;
|
||
}
|
||
(left, right, taken, right_glyphs)
|
||
}
|
||
|
||
/// Works as [`optical_hangs`] does for a right-to-left line, whose `line` is also in visual order.
|
||
/// The line is read from right to left: `."` ending it hangs both its glyphs at the left, `'"` starting it hangs both at
|
||
/// the right, and the right end takes its glyphs first.
|
||
pub(super) fn optical_hangs_rtl(line: &[(&SGlyph, f64)], share: fn(&SGlyph) -> f64) -> (f64, f64, usize, usize) {
|
||
let reversed: Vec<(&SGlyph, f64)> = line.iter().rev().copied().collect();
|
||
let (right, left, right_glyphs, left_glyphs) = optical_hangs(&reversed, share);
|
||
(left, right, left_glyphs, right_glyphs)
|
||
}
|
||
|
||
/// How much narrower a line whose last character is glyph `end - 1` is set: `line_end` of that
|
||
/// glyph ([`break_line`]), at most its advance, and 0 when it is not finite.
|
||
fn line_end_credit(g: &[SGlyph], line_end: &[f64], end: usize) -> f64 {
|
||
let Some(k) = end.checked_sub(1) else { return 0.0 };
|
||
match (line_end.get(k), g.get(k)) {
|
||
(Some(&credit), Some(gl)) if credit.is_finite() && credit > 0.0 => credit.min(gl.adv.max(0.0)),
|
||
_ => 0.0,
|
||
}
|
||
}
|
||
|
||
/// May [`break_line`] end a line after glyph `end - 1` (and the spaces after it)? At the paragraph's
|
||
/// end or a forced line break, and where it records a break without a hyphen.
|
||
fn may_end_after(g: &[SGlyph], end: usize, kinsoku: Kinsoku) -> bool {
|
||
let mut e = end;
|
||
while g.get(e).is_some_and(SGlyph::is_space) {
|
||
e += 1;
|
||
}
|
||
if e >= g.len() || g[e].ch == '\u{2028}' {
|
||
return true;
|
||
}
|
||
let Some(last) = e.checked_sub(1).and_then(|k| g.get(k)) else { return false };
|
||
last.break_after() && last.len > 0 && !last.is_soft_hyphen() && kinsoku_allows(g, e - 1, kinsoku) && !splits_no_break(g, e)
|
||
}
|
||
|
||
/// Would a line ending at glyph index `end` split a No Break run, or an inseparable pair that
|
||
/// Bunri-Kinshi keeps together ([`SGlyph::keeps_next`])?
|
||
fn splits_no_break(g: &[SGlyph], end: usize) -> bool {
|
||
end > 0 && g.get(end - 1).is_some_and(|a| (a.no_break && g.get(end).is_some_and(|b| b.no_break)) || (a.keeps_next && g.get(end).is_some()))
|
||
}
|
||
|
||
/// Does kinsoku allow a line break after glyph `j`? Not after an opening bracket, nor before a
|
||
/// closing one, a comma, a full stop, a small kana… (the character is pushed to the next line
|
||
/// with the one before it).
|
||
fn kinsoku_allows(g: &[SGlyph], j: usize, kinsoku: Kinsoku) -> bool {
|
||
g.get(j).is_some_and(|gl| kinsoku_between(gl.ch, g.get(j + 1).map(|n| n.ch), kinsoku))
|
||
}
|
||
|
||
/// Kinsoku for a break between `before` and `after` (none: the end of the paragraph): Hard keeps
|
||
/// every character of the set off the line start and end, Soft lets 々, ー and small kana start a
|
||
/// line, None allows the break.
|
||
fn kinsoku_between(before: char, after: Option<char>, kinsoku: Kinsoku) -> bool {
|
||
match kinsoku {
|
||
Kinsoku::None => true,
|
||
Kinsoku::Hard => !no_line_end(before) && after.is_none_or(|c| !no_line_start(c)),
|
||
Kinsoku::Soft => !no_line_end(before) && after.is_none_or(|c| !no_line_start(c) || soft_line_start(c)),
|
||
}
|
||
}
|
||
|
||
/// The share of an em by which a push-in ([`push_in`]) may close each gap next to a CJK character
|
||
/// on the line, measured on the character before the gap.
|
||
pub(super) const PUSH_IN_SQUEEZE: f64 = 0.1;
|
||
|
||
/// The most glyphs after the overflowing cluster a push-in ([`push_in`]) takes onto the line; a
|
||
/// longer run of characters kinsoku keeps together pushes out.
|
||
const PUSH_IN_RUN: usize = 16;
|
||
|
||
/// Kinsoku Shori Type: the end of the line from `i` that overflows `width` at the cluster
|
||
/// `j..cluster_end` (the line is `x` wide through it) when the line pushes in, or None when it
|
||
/// pushes out. A line pushes in only where kinsoku rules out the break before `j`: a character
|
||
/// that may not end a line comes before it (an opening bracket or quotation mark, Latin ones
|
||
/// included), or one that may not start a line comes at `j`. It then takes the characters up to
|
||
/// the next break kinsoku allows (not at a soft hyphen), and the spaces after it, when closing
|
||
/// its gaps next to CJK characters, each by up to [`PUSH_IN_SQUEEZE`] of an em of the character
|
||
/// before it, makes room for them. Push In First tries this before it pushes out. Push Out First
|
||
/// tries it only when pushing out, to the line's last allowed break `pushed_out`, would leave the
|
||
/// line short by more than an em of the character before `j`. Push Out Only never pushes in, and
|
||
/// neither does a line holding a tab or right-to-left text, nor one where the characters taken in
|
||
/// reach a space after a character that may not end a line. With `hanging` (Optical Margin
|
||
/// Alignment or Roman Hanging Punctuation), the gaps that hanging punctuation keeps open are not
|
||
/// counted. `vertical`: the line is a column of vertical type ([`squeeze_em`]). Both lines are
|
||
/// measured with their last character set as narrow as `line_end` sets it ([`break_line`]).
|
||
#[allow(clippy::too_many_arguments)]
|
||
fn push_in(
|
||
g: &[SGlyph],
|
||
i: usize,
|
||
j: usize,
|
||
cluster_end: usize,
|
||
x: f64,
|
||
width: f64,
|
||
pushed_out: Option<usize>,
|
||
kinsoku: Kinsoku,
|
||
kinsoku_type: KinsokuShoriType,
|
||
hanging: bool,
|
||
vertical: bool,
|
||
line_end: &[f64],
|
||
) -> Option<usize> {
|
||
if kinsoku == Kinsoku::None || kinsoku_type == KinsokuShoriType::PushOutOnly {
|
||
return None;
|
||
}
|
||
let before = g.get(j.checked_sub(1)?)?;
|
||
if before.len == 0 || g.get(j)?.no_break || splits_no_break(g, j) || kinsoku_allows(g, j - 1, kinsoku) {
|
||
return None;
|
||
}
|
||
let allowed = |k: usize| {
|
||
k >= g.len()
|
||
|| (g[k].byte != g[k - 1].byte
|
||
&& kinsoku_allows(g, k - 1, kinsoku)
|
||
&& !splits_no_break(g, k)
|
||
&& (g[k].is_space() || (g[k - 1].break_after() && !g[k - 1].is_soft_hyphen())))
|
||
};
|
||
let (mut end, mut x) = (cluster_end, x);
|
||
while !allowed(end) {
|
||
let gl = &g[end];
|
||
// A space here follows a character kinsoku keeps off a line's end: the line pushes out.
|
||
if gl.is_space() || gl.ch == '\t' || gl.ch == '\u{2028}' || gl.no_break || end - cluster_end >= PUSH_IN_RUN {
|
||
return None;
|
||
}
|
||
x += gl.adv;
|
||
end += 1;
|
||
}
|
||
// A tab's width depends on where it falls; a line with one (other than a list marker's)
|
||
// pushes out.
|
||
if g[i..end].iter().any(|h| h.ch == '\t' && h.len > 0) {
|
||
return None;
|
||
}
|
||
// With `hanging`, punctuation keeps the gap after a mark that may hang at the line's start,
|
||
// and the gap before one that may hang at its end ([`optical_hangs`]): those gaps are not
|
||
// counted.
|
||
let mut last = end;
|
||
while last > i && g[last - 1].is_space() {
|
||
last -= 1;
|
||
}
|
||
// A line holding right-to-left text is reordered before it is drawn, and justification closes
|
||
// the gaps between its glyphs in that order: it pushes out.
|
||
if g[i..last].iter().any(|h| h.len > 0 && h.level.is_rtl()) {
|
||
return None;
|
||
}
|
||
let punct = |k: usize| g.get(k).is_some_and(optical_punct);
|
||
let quote = |k: usize| g.get(k).is_some_and(|h| is_quote(h.ch));
|
||
let may_hang = |a: usize| {
|
||
hanging
|
||
&& ((a == i && punct(i))
|
||
|| (a == i + 1 && punct(i) && quote(i + 1))
|
||
|| (a + 2 == last && punct(last - 1))
|
||
|| (a + 3 == last && punct(last - 2) && quote(last - 1)))
|
||
};
|
||
let room: f64 = (i..end - 1).filter(|&a| closes(g, a, end) && !may_hang(a)).map(|a| PUSH_IN_SQUEEZE * squeeze_em(&g[a], vertical)).sum();
|
||
if x - line_end_credit(g, line_end, last) - width > room + EPS {
|
||
return None;
|
||
}
|
||
if kinsoku_type == KinsokuShoriType::PushOutFirst {
|
||
let mut kept = pushed_out.filter(|&b| b > i).unwrap_or(i);
|
||
while kept > i && g[kept - 1].is_space() {
|
||
kept -= 1;
|
||
}
|
||
let set = if kept > i { line_end_credit(g, line_end, kept) } else { 0.0 };
|
||
let short = width - (g[i..kept].iter().map(|h| h.adv).sum::<f64>() - set);
|
||
if short <= squeeze_em(before, vertical) + EPS {
|
||
return None;
|
||
}
|
||
}
|
||
while g.get(end).is_some_and(SGlyph::is_space) {
|
||
end += 1;
|
||
}
|
||
if g.get(end).is_some_and(|h| h.ch == '\u{2028}') {
|
||
end += 1;
|
||
}
|
||
Some(end)
|
||
}
|
||
|
||
/// May a push-in close the gap after glyph `a`? A gap between two characters, at least one of them
|
||
/// CJK, outside a tate-chu-yoko or warichu block, a list marker and a shaped cluster, and not next to a space,
|
||
/// a tab or an inline graphic before it, nor inside an inseparable pair that Bunri-Kinshi keeps
|
||
/// together. Zero-width format controls take no gap: the gap is the one between the visible
|
||
/// characters on either side, before `end`, as the layout justifies it.
|
||
fn closes(g: &[SGlyph], a: usize, end: usize) -> bool {
|
||
let Some(p) = g.get(a).filter(|p| !is_format_control(p.ch)) else { return false };
|
||
let Some(n) = g.get(a + 1..end).and_then(|rest| rest.iter().find(|n| !is_format_control(n.ch))) else { return false };
|
||
p.len > 0
|
||
&& !p.keeps_next
|
||
&& p.byte != n.byte
|
||
&& p.inline.is_none()
|
||
&& !p.is_space()
|
||
&& !n.is_space()
|
||
&& p.ch != '\t'
|
||
&& !n.continues_block()
|
||
&& (is_cjk(p.ch) || is_cjk(n.ch))
|
||
}
|
||
|
||
/// The em a push-in measures in, along the line: `g`'s em at its Horizontal Scale, as Mojikumi's
|
||
/// spacing is, or at its Vertical Scale when it stands upright in vertical type and the font's
|
||
/// vertical metrics set its length down the column ([`super::vertical::upright_cell`]).
|
||
pub(super) fn squeeze_em(g: &SGlyph, vertical: bool) -> f64 {
|
||
let down_the_column = vertical && stands_upright(g) && g.face.vertical_glyph(g.gid).is_some();
|
||
g.face.units_per_em() * if down_the_column { g.sy } else { g.sx }
|
||
}
|
||
|
||
#[cfg(test)]
|
||
#[test]
|
||
fn kinsoku_keeps_closing_marks_off_line_starts_and_opening_ones_off_line_ends() {
|
||
for (a, b) in [('字', '。'), ('弧', '」'), ('」', '、'), ('ャ', 'ー'), ('カ', 'ッ'), ('「', 'か'), ('(', '雅')] {
|
||
assert!(!kinsoku_between(a, Some(b), Kinsoku::Hard), "{a}{b}");
|
||
}
|
||
for (a, b) in [('。', '雅'), ('」', 'は'), ('字', '「'), ('の', '演')] {
|
||
assert!(kinsoku_between(a, Some(b), Kinsoku::Hard), "{a}{b}");
|
||
}
|
||
assert!(kinsoku_between('。', None, Kinsoku::Hard));
|
||
}
|
||
|
||
/// Soft kinsoku lets 々, ー and small kana (JLREQ cl-09's 々, cl-10, cl-11) start a line and keeps
|
||
/// the rest of the set (half-width forms aren't in those classes); None allows any break.
|
||
#[cfg(test)]
|
||
#[test]
|
||
fn soft_kinsoku_lets_small_kana_and_the_prolonged_sound_mark_start_a_line() {
|
||
for (a, b) in [('ャ', 'ー'), ('カ', 'ッ'), ('時', '々'), ('ト', 'ㇰ'), ('か', 'ゃ')] {
|
||
assert!(!kinsoku_between(a, Some(b), Kinsoku::Hard), "{a}{b}");
|
||
assert!(kinsoku_between(a, Some(b), Kinsoku::Soft), "{a}{b}");
|
||
}
|
||
for (a, b) in [('字', '。'), ('弧', '」'), ('「', 'か'), ('カ', 'ッ'), ('カ', 'ー'), ('字', 'ゝ')] {
|
||
assert!(!kinsoku_between(a, Some(b), Kinsoku::Soft), "{a}{b}");
|
||
assert!(kinsoku_between(a, Some(b), Kinsoku::None), "{a}{b}");
|
||
}
|
||
}
|
||
|
||
/// Glyph indices inside `g[ws..we]` (a word) where a hyphenated break may go under `opts`, in the
|
||
/// language of the word's first letter; never next to a Hangul letter, which takes no hyphen, and
|
||
/// nowhere in a Korean word that Korean Line Break Anywhere off keeps whole.
|
||
fn hyphen_breaks(text: &str, g: &[SGlyph], ws: usize, we: usize, opts: &HyphenationOptions) -> Vec<usize> {
|
||
if g.get(ws..we).is_some_and(|word| word.iter().any(|h| h.keeps_word)) {
|
||
return Vec::new();
|
||
}
|
||
let mut points = word_hyphen_breaks(text, g, ws, we, opts);
|
||
let hangul = |i: usize| g.get(i).is_some_and(|h| h.inline.is_none() && crate::shape::is_hangul(h.ch));
|
||
points.retain(|&k| !hangul(k.wrapping_sub(1)) && !hangul(k));
|
||
points
|
||
}
|
||
|
||
/// [`hyphen_breaks`] before the points next to Hangul are left out.
|
||
fn word_hyphen_breaks(text: &str, g: &[SGlyph], ws: usize, we: usize, opts: &HyphenationOptions) -> Vec<usize> {
|
||
// Strip leading/trailing punctuation (quotes, commas).
|
||
let (mut a, mut b) = (ws, we);
|
||
while a < b && !g[a].is_letter() {
|
||
a += 1;
|
||
}
|
||
while b > a && !g[b - 1].is_letter() {
|
||
b -= 1;
|
||
}
|
||
if b <= a || g[a..b].iter().any(|g| !g.is_letter()) {
|
||
return vec![];
|
||
}
|
||
let (s, e) = (g[a].byte, g[b - 1].byte + g[b - 1].len);
|
||
let Some(word) = text.get(s..e) else { return vec![] };
|
||
let offs: Vec<usize> = word.char_indices().map(|(o, _)| s + o).collect();
|
||
// The first glyph of each cluster start (not inside a ligature), by its byte: one pass, not a
|
||
// search for each point.
|
||
let mut starts = std::collections::HashMap::new();
|
||
for k in a + 1..b {
|
||
if g[k].byte != g[k - 1].byte {
|
||
starts.entry(g[k].byte).or_insert(k);
|
||
}
|
||
}
|
||
points_for(word, opts, g[a].hyphen_language).into_iter().filter_map(|ci| starts.get(offs.get(ci)?).copied()).collect()
|
||
}
|
||
|
||
/// Break candidates for the every-line composer: hyphenation points under `hyphenate` (the
|
||
/// paragraph's Hyphenation options) when it is given. No break comes before a space: a line that
|
||
/// may end after a dash, a Thai word or a space ends after the spaces that follow it, as in
|
||
/// [`break_line`].
|
||
pub(super) fn candidates(text: &str, g: &[SGlyph], hyphenate: Option<&HyphenationOptions>, kinsoku: Kinsoku) -> Vec<Breakpoint> {
|
||
let mut v = vec![];
|
||
let mut ws = 0;
|
||
for (j, gl) in g.iter().enumerate() {
|
||
// A list marker (the glyphs of length 0 here) ends before the first word.
|
||
let marker_end = gl.len == 0 && g.get(j + 1).is_some_and(|n| n.len > 0);
|
||
let end_of_word = gl.is_space() || gl.break_after() || marker_end;
|
||
if end_of_word {
|
||
if let Some(opts) = hyphenate
|
||
&& j > ws
|
||
{
|
||
for k in
|
||
hyphen_breaks(text, g, ws, j, opts).into_iter().filter(|&k| k > 0 && kinsoku_allows(g, k - 1, kinsoku) && !splits_no_break(g, k))
|
||
{
|
||
v.push(Breakpoint { end: k, hyphen: hyphen_glyph(&g[k - 1]).adv });
|
||
}
|
||
}
|
||
let hy = if gl.is_soft_hyphen() { hyphen_glyph(gl).adv } else { 0.0 };
|
||
if j + 1 < g.len() && g[j + 1].byte != gl.byte && !g[j + 1].is_space() && kinsoku_allows(g, j, kinsoku) && !splits_no_break(g, j + 1) {
|
||
v.push(Breakpoint { end: j + 1, hyphen: hy });
|
||
}
|
||
ws = j + 1;
|
||
}
|
||
}
|
||
if let Some(opts) = hyphenate
|
||
&& g.len() > ws
|
||
{
|
||
for k in
|
||
hyphen_breaks(text, g, ws, g.len(), opts).into_iter().filter(|&k| k > 0 && kinsoku_allows(g, k - 1, kinsoku) && !splits_no_break(g, k))
|
||
{
|
||
v.push(Breakpoint { end: k, hyphen: hyphen_glyph(&g[k - 1]).adv });
|
||
}
|
||
}
|
||
v.sort_by_key(|b| b.end);
|
||
v.dedup_by_key(|b| b.end);
|
||
v
|
||
}
|
||
|
||
#[cfg(test)]
|
||
#[test]
|
||
fn a_pen_out_of_baseline_steps_finds_no_room() {
|
||
let frame = kurbo::Shape::to_path(&kurbo::Rect::new(0.0, 0.0, 400.0, 400.0), 0.1);
|
||
let regions = Region::cells(&frame, &super::LayoutOptions::default(), &[]);
|
||
let est = Metrics { asc: 10.0, desc: 3.0, lead: 14.0, cap: 7.0, xh: 5.0, top: 10.0 };
|
||
let pen = |steps| Pen {
|
||
regions: Some(®ions),
|
||
ri: 0,
|
||
prev: None,
|
||
pending: 0.0,
|
||
fb: FirstBaseline::Ascent,
|
||
fb_min: 0.0,
|
||
queued: vec![],
|
||
above: Default::default(),
|
||
model: LeadingModel::default(),
|
||
next_top: None,
|
||
no_room_for: f64::INFINITY,
|
||
steps_left: std::rc::Rc::new(std::cell::Cell::new(steps)),
|
||
search: Default::default(),
|
||
};
|
||
assert!(pen(MAX_TOTAL_STEPS).place(est, 0.0, 0.0, 0.0).is_some());
|
||
assert!(pen(0).place(est, 0.0, 0.0, 0.0).is_none());
|
||
}
|