Text shaping 🚀

- Shapes text with rustybuzz
- Font fallback with family list
- Tofus are shown in the first font

Co-Authored-By: Martin <mhaug@live.de>

3 files changed, 174 insertions, 219 deletions

diff --git a/src/layout/frame.rs b/src/layout/frame.rs
index 6e876151..d3276e99 100644
--- a/src/layout/frame.rs
+++ b/src/layout/frame.rs
@@ -1,7 +1,9 @@
-use super::Shaped;
+use fontdock::FaceId;
+use ttf_parser::GlyphId;
+
 use crate::color::Color;
 use crate::env::ResourceId;
-use crate::geom::{Path, Point, Size};
+use crate::geom::{Length, Path, Point, Size};
 
 /// A finished layout with elements at fixed positions.
 #[derive(Debug, Clone, PartialEq)]
@@ -36,13 +38,67 @@ impl Frame {
 #[derive(Debug, Clone, PartialEq)]
 pub enum Element {
     /// Shaped text.
-    Text(Shaped),
+    Text(ShapedText),
     /// A geometric shape.
     Geometry(Geometry),
     /// A raster image.
     Image(Image),
 }
 
+/// A shaped run of text.
+#[derive(Debug, Clone, PartialEq)]
+pub struct ShapedText {
+    /// The font face the text was shaped with.
+    pub face: FaceId,
+    /// The font size.
+    pub size: Length,
+    /// The width.
+    pub width: Length,
+    /// The extent to the top.
+    pub top: Length,
+    /// The extent to the bottom.
+    pub bottom: Length,
+    /// The glyph fill color / texture.
+    pub color: Fill,
+    /// The shaped glyphs.
+    pub glyphs: Vec<GlyphId>,
+    /// The horizontal offsets of the glyphs. This is indexed parallel to
+    /// `glyphs`. Vertical offsets are not yet supported.
+    pub offsets: Vec<Length>,
+}
+
+impl ShapedText {
+    /// Create a new shape run with `width` zero and empty `glyphs` and `offsets`.
+    pub fn new(
+        face: FaceId,
+        size: Length,
+        top: Length,
+        bottom: Length,
+        color: Fill,
+    ) -> Self {
+        Self {
+            face,
+            size,
+            width: Length::ZERO,
+            top,
+            bottom,
+            glyphs: vec![],
+            offsets: vec![],
+            color,
+        }
+    }
+
+    /// Encode the glyph ids into a big-endian byte buffer.
+    pub fn encode_glyphs_be(&self) -> Vec<u8> {
+        let mut bytes = Vec::with_capacity(2 * self.glyphs.len());
+        for &GlyphId(g) in &self.glyphs {
+            bytes.push((g >> 8) as u8);
+            bytes.push((g & 0xff) as u8);
+        }
+        bytes
+    }
+}
+
 /// A shape with some kind of fill.
 #[derive(Debug, Clone, PartialEq)]
 pub struct Geometry {
diff --git a/src/layout/shaping.rs b/src/layout/shaping.rs
index fd10b41e..f7eece92 100644
--- a/src/layout/shaping.rs
+++ b/src/layout/shaping.rs
@@ -1,205 +1,129 @@
-//! Super-basic text shaping.
-//!
-//! This is really only suited for simple Latin text. It picks the most suitable
-//! font for each individual character. When the direction is right-to-left, the
-//! word is spelled backwards. Vertical shaping is not supported.
-
-use std::fmt::{self, Debug, Display, Formatter};
-
-use fontdock::{FaceId, FontVariant};
-use ttf_parser::{Face, GlyphId};
+use fontdock::FaceId;
+use rustybuzz::UnicodeBuffer;
+use ttf_parser::GlyphId;
 
+use super::{Element, Frame, ShapedText};
 use crate::env::FontLoader;
-use crate::exec::FamilyMap;
-use crate::geom::{Dir, Length, Point, Size};
-use crate::layout::{Element, Fill, Frame};
-
-/// A shaped run of text.
-#[derive(Clone, PartialEq)]
-pub struct Shaped {
-    /// The shaped text.
-    pub text: String,
-    /// The font face the text was shaped with.
-    pub face: FaceId,
-    /// The shaped glyphs.
-    pub glyphs: Vec<GlyphId>,
-    /// The horizontal offsets of the glyphs. This is indexed parallel to
-    /// `glyphs`. Vertical offsets are not yet supported.
-    pub offsets: Vec<Length>,
-    /// The font size.
-    pub font_size: Length,
-    /// The glyph fill color / texture.
-    pub color: Fill,
-}
-
-impl Shaped {
-    /// Create a new shape run with empty `text`, `glyphs` and `offsets`.
-    pub fn new(face: FaceId, font_size: Length, color: Fill) -> Self {
-        Self {
-            text: String::new(),
-            face,
-            glyphs: vec![],
-            offsets: vec![],
-            font_size,
-            color,
-        }
-    }
-
-    /// Encode the glyph ids into a big-endian byte buffer.
-    pub fn encode_glyphs_be(&self) -> Vec<u8> {
-        let mut bytes = Vec::with_capacity(2 * self.glyphs.len());
-        for &GlyphId(g) in &self.glyphs {
-            bytes.push((g >> 8) as u8);
-            bytes.push((g & 0xff) as u8);
-        }
-        bytes
-    }
-}
-
-impl Debug for Shaped {
-    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
-        Debug::fmt(&self.text, f)
-    }
-}
-
-/// Identifies a vertical metric of a font.
-#[derive(Debug, Copy, Clone, Eq, PartialEq, Ord, PartialOrd)]
-pub enum VerticalFontMetric {
-    /// The distance from the baseline to the typographic ascender.
-    ///
-    /// Corresponds to the typographic ascender from the `OS/2` table if present
-    /// and falls back to the ascender from the `hhea` table otherwise.
-    Ascender,
-    /// The approximate height of uppercase letters.
-    CapHeight,
-    /// The approximate height of non-ascending lowercase letters.
-    XHeight,
-    /// The baseline on which the letters rest.
-    Baseline,
-    /// The distance from the baseline to the typographic descender.
-    ///
-    /// Corresponds to the typographic descender from the `OS/2` table if
-    /// present and falls back to the descender from the `hhea` table otherwise.
-    Descender,
-}
+use crate::exec::FontProps;
+use crate::geom::{Length, Point, Size};
 
-impl Display for VerticalFontMetric {
-    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
-        f.pad(match self {
-            Self::Ascender => "ascender",
-            Self::CapHeight => "cap-height",
-            Self::XHeight => "x-height",
-            Self::Baseline => "baseline",
-            Self::Descender => "descender",
-        })
-    }
+/// Shape text into a frame containing shaped [`ShapedText`] runs.
+pub fn shape(text: &str, loader: &mut FontLoader, props: &FontProps) -> Frame {
+    let mut frame = Frame::new(Size::new(Length::ZERO, Length::ZERO));
+    shape_segment(&mut frame, text, props.families.iter(), None, loader, props);
+    frame
 }
 
-/// Shape text into a frame containing [`Shaped`] runs.
-pub fn shape(
+/// Shape text into a frame with font fallback using the `families` iterator.
+fn shape_segment<'a>(
+    frame: &mut Frame,
     text: &str,
-    dir: Dir,
-    families: &FamilyMap,
-    variant: FontVariant,
-    font_size: Length,
-    top_edge: VerticalFontMetric,
-    bottom_edge: VerticalFontMetric,
-    color: Fill,
+    mut families: impl Iterator<Item = &'a str> + Clone,
+    mut first: Option<FaceId>,
     loader: &mut FontLoader,
-) -> Frame {
-    let mut frame = Frame::new(Size::new(Length::ZERO, Length::ZERO));
-    let mut shaped = Shaped::new(FaceId::MAX, font_size, color);
-    let mut width = Length::ZERO;
-    let mut top = Length::ZERO;
-    let mut bottom = Length::ZERO;
-
-    // Create an iterator with conditional direction.
-    let mut forwards = text.chars();
-    let mut backwards = text.chars().rev();
-    let chars: &mut dyn Iterator<Item = char> = if dir.is_positive() {
-        &mut forwards
-    } else {
-        &mut backwards
+    props: &FontProps,
+) {
+    // Select the font family.
+    let (id, fallback) = loop {
+        // Try to load the next available font family.
+        match families.next() {
+            Some(family) => match loader.query(family, props.variant) {
+                Some(id) => break (id, true),
+                None => {}
+            },
+            // We're out of families, so we don't do any more fallback and just
+            // shape the tofus with the first face we originally used.
+            None => match first {
+                Some(id) => break (id, false),
+                None => return,
+            },
+        }
     };
 
-    for c in chars {
-        for family in families.iter() {
-            if let Some(id) = loader.query(family, variant) {
-                let face = loader.face(id).get();
-                let (glyph, glyph_width) = match lookup_glyph(face, c) {
-                    Some(v) => v,
-                    None => continue,
-                };
-
-                let units_per_em = f64::from(face.units_per_em().unwrap_or(1000));
-                let convert = |units| units / units_per_em * font_size;
+    // Register that this is the first available font.
+    let face = loader.face(id);
+    if first.is_none() {
+        first = Some(id);
+    }
 
-                // Flush the buffer and reset the metrics if we use a new font face.
-                if shaped.face != id {
-                    place(&mut frame, shaped, width, top, bottom);
+    // Find out some metrics and prepare the shaped text container.
+    let ttf = face.ttf();
+    let units_per_em = f64::from(ttf.units_per_em().unwrap_or(1000));
+    let convert = |units| f64::from(units) / units_per_em * props.size;
+    let top = convert(i32::from(props.top_edge.lookup(ttf)));
+    let bottom = convert(i32::from(props.bottom_edge.lookup(ttf)));
+    let mut shaped = ShapedText::new(id, props.size, top, bottom, props.color);
+
+    // Fill the buffer with our text.
+    let mut buffer = UnicodeBuffer::new();
+    buffer.push_str(text);
+    buffer.guess_segment_properties();
+
+    // Find out the text direction.
+    // TODO: Replace this once we do BiDi.
+    let rtl = matches!(buffer.direction(), rustybuzz::Direction::RightToLeft);
+
+    // Shape!
+    let glyphs = rustybuzz::shape(face.buzz(), &[], buffer);
+    let info = glyphs.glyph_infos();
+    let pos = glyphs.glyph_positions();
+    let mut iter = info.iter().zip(pos).peekable();
+
+    while let Some((info, pos)) = iter.next() {
+        // Do font fallback if the glyph is a tofu.
+        if info.codepoint == 0 && fallback {
+            // Flush what we have so far.
+            if !shaped.glyphs.is_empty() {
+                place(frame, shaped);
+                shaped = ShapedText::new(id, props.size, top, bottom, props.color);
+            }
 
-                    shaped = Shaped::new(id, font_size, color);
-                    width = Length::ZERO;
-                    top = convert(f64::from(lookup_metric(face, top_edge)));
-                    bottom = convert(f64::from(lookup_metric(face, bottom_edge)));
+            // Determine the start and end cluster index of the tofu sequence.
+            let mut start = info.cluster as usize;
+            let mut end = info.cluster as usize;
+            while let Some((info, _)) = iter.peek() {
+                if info.codepoint != 0 {
+                    break;
                 }
-
-                shaped.text.push(c);
-                shaped.glyphs.push(glyph);
-                shaped.offsets.push(width);
-                width += convert(f64::from(glyph_width));
-                break;
+                end = info.cluster as usize;
+                iter.next();
             }
-        }
-    }
 
-    place(&mut frame, shaped, width, top, bottom);
-
-    frame
-}
+            // Because Harfbuzz outputs glyphs in visual order, the start
+            // cluster actually corresponds to the last codepoint in
+            // right-to-left text.
+            if rtl {
+                assert!(end <= start);
+                std::mem::swap(&mut start, &mut end);
+            }
 
-/// Place shaped text into a frame.
-fn place(frame: &mut Frame, shaped: Shaped, width: Length, top: Length, bottom: Length) {
-    if !shaped.text.is_empty() {
-        frame.push(Point::new(frame.size.width, top), Element::Text(shaped));
-        frame.size.width += width;
-        frame.size.height = frame.size.height.max(top - bottom);
+            // The end cluster index points right before the last character that
+            // mapped to the tofu sequence. So we have to offset the end by one
+            // char.
+            let offset = text[end ..].chars().next().unwrap().len_utf8();
+            let range = start .. end + offset;
+
+            // Recursively shape the tofu sequence with the next family.
+            shape_segment(frame, &text[range], families.clone(), first, loader, props);
+        } else {
+            // Add the glyph to the shaped output.
+            // TODO: Don't ignore y_advance and y_offset.
+            let glyph = GlyphId(info.codepoint as u16);
+            shaped.glyphs.push(glyph);
+            shaped.offsets.push(shaped.width + convert(pos.x_offset));
+            shaped.width += convert(pos.x_advance);
+        }
     }
-}
 
-/// Look up the glyph for `c` and returns its index alongside its advance width.
-fn lookup_glyph(face: &Face, c: char) -> Option<(GlyphId, u16)> {
-    let glyph = face.glyph_index(c)?;
-    let width = face.glyph_hor_advance(glyph)?;
-    Some((glyph, width))
-}
-
-/// Look up a vertical metric.
-fn lookup_metric(face: &Face, metric: VerticalFontMetric) -> i16 {
-    match metric {
-        VerticalFontMetric::Ascender => lookup_ascender(face),
-        VerticalFontMetric::CapHeight => face
-            .capital_height()
-            .filter(|&h| h > 0)
-            .unwrap_or_else(|| lookup_ascender(face)),
-        VerticalFontMetric::XHeight => face
-            .x_height()
-            .filter(|&h| h > 0)
-            .unwrap_or_else(|| lookup_ascender(face)),
-        VerticalFontMetric::Baseline => 0,
-        VerticalFontMetric::Descender => lookup_descender(face),
+    if !shaped.glyphs.is_empty() {
+        place(frame, shaped)
     }
 }
 
-/// The ascender of the face.
-fn lookup_ascender(face: &Face) -> i16 {
-    // We prefer the typographic ascender over the Windows ascender because
-    // it can be overly large if the font has large glyphs.
-    face.typographic_ascender().unwrap_or_else(|| face.ascender())
-}
-
-/// The descender of the face.
-fn lookup_descender(face: &Face) -> i16 {
-    // See `lookup_ascender` for reason.
-    face.typographic_descender().unwrap_or_else(|| face.descender())
+/// Place shaped text into a frame.
+fn place(frame: &mut Frame, shaped: ShapedText) {
+    let offset = frame.size.width;
+    frame.size.width += shaped.width;
+    frame.size.height = frame.size.height.max(shaped.top - shaped.bottom);
+    frame.push(Point::new(offset, shaped.top), Element::Text(shaped));
 }
diff --git a/src/layout/text.rs b/src/layout/text.rs
index 2239afac..39866907 100644
--- a/src/layout/text.rs
+++ b/src/layout/text.rs
@@ -1,50 +1,25 @@
 use std::fmt::{self, Debug, Formatter};
-use std::rc::Rc;
-
-use fontdock::FontVariant;
 
 use super::*;
-use crate::exec::FamilyMap;
+use crate::exec::FontProps;
 
 /// A consecutive, styled run of text.
 #[derive(Clone, PartialEq)]
 pub struct TextNode {
-    /// The text.
-    pub text: String,
     /// The text direction.
     pub dir: Dir,
     /// How to align this text node in its parent.
     pub aligns: LayoutAligns,
-    /// The list of font families for shaping.
-    pub families: Rc<FamilyMap>,
-    /// The font variant,
-    pub variant: FontVariant,
-    /// The font size.
-    pub font_size: Length,
-    /// The top end of the text bounding box.
-    pub top_edge: VerticalFontMetric,
-    /// The bottom end of the text bounding box.
-    pub bottom_edge: VerticalFontMetric,
-    /// The glyph fill.
-    pub color: Fill,
+    /// The text.
+    pub text: String,
+    /// Properties used for font selection and layout.
+    pub props: FontProps,
 }
 
 impl Layout for TextNode {
     fn layout(&self, ctx: &mut LayoutContext, _: &Areas) -> Fragment {
-        Fragment::Frame(
-            shape(
-                &self.text,
-                self.dir,
-                &self.families,
-                self.variant,
-                self.font_size,
-                self.top_edge,
-                self.bottom_edge,
-                self.color,
-                &mut ctx.env.fonts,
-            ),
-            self.aligns,
-        )
+        let frame = shape(&self.text, &mut ctx.env.fonts, &self.props);
+        Fragment::Frame(frame, self.aligns)
     }
 }