Better space coalescing logic 🌧

This creates a smaller state machine helper type for softness coalescing, which does not own the resulting nodes. While this creates a bit more duplication in stack and par builder, it makes it a lot easier to integrate additional logic into the paragraph builder.

Furthermore:
- Line breaks are now "hard", that is, not coalesced with each other.
- Text nodes with equal style are now merged allowing for example `f{}i` to form a ligature.

src/exec/context.rs

@@ -65,87 +65,66 @@ impl<'a> ExecContext<'a> {
         mem::replace(&mut self.stack, stack).build()
     }
 
+    /// Push any node into the active paragraph.
+    pub fn push(&mut self, node: impl Into<AnyNode>) {
+        let align = self.state.aligns.cross;
+        self.stack.par.push(ParChild::Any(node.into(), align));
+    }
+
+    /// Push a word space into the active paragraph.
+    pub fn push_word_space(&mut self) {
+        let em = self.state.font.resolve_size();
+        let amount = self.state.par.word_spacing.resolve(em);
+        self.stack.par.push_soft(ParChild::Spacing(amount));
+    }
+
     /// Push text into the active paragraph.
     ///
     /// The text is split into lines at newlines.
     pub fn push_text(&mut self, text: &str) {
         let mut scanner = Scanner::new(text);
-        let mut line = String::new();
-        let push = |this: &mut Self, text| {
-            let props = this.state.font.resolve_props();
-            let node = TextNode { text, props };
-            let align = this.state.aligns.cross;
-            this.stack.par.folder.push(ParChild::Text(node, align))
-        };
+        let mut text = String::new();
 
         while let Some(c) = scanner.eat_merging_crlf() {
             if is_newline(c) {
-                push(self, mem::take(&mut line));
-                self.push_linebreak();
+                self.stack.par.push_text(mem::take(&mut text), &self.state);
+                self.linebreak();
             } else {
-                line.push(c);
+                text.push(c);
             }
         }
 
-        push(self, line);
-    }
-
-    /// Push a word space.
-    pub fn push_word_space(&mut self) {
-        let em = self.state.font.resolve_size();
-        let amount = self.state.par.word_spacing.resolve(em);
-        self.push_spacing(GenAxis::Cross, amount, 1);
-    }
-
-    /// Apply a forced line break.
-    pub fn push_linebreak(&mut self) {
-        let em = self.state.font.resolve_size();
-        let amount = self.state.par.leading.resolve(em);
-        self.push_spacing(GenAxis::Main, amount, 2);
-    }
-
-    /// Apply a forced paragraph break.
-    pub fn push_parbreak(&mut self) {
-        let em = self.state.font.resolve_size();
-        let amount = self.state.par.spacing.resolve(em);
-        self.push_spacing(GenAxis::Main, amount, 1);
+        self.stack.par.push_text(text, &self.state);
     }
 
     /// Push spacing into paragraph or stack depending on `axis`.
-    ///
-    /// The `softness` configures how the spacing interacts with surrounding
-    /// spacing.
-    pub fn push_spacing(&mut self, axis: GenAxis, amount: Length, softness: u8) {
+    pub fn push_spacing(&mut self, axis: GenAxis, amount: Length) {
         match axis {
             GenAxis::Main => {
-                let spacing = StackChild::Spacing(amount);
-                self.stack.finish_par(&self.state);
-                self.stack.folder.push_soft(spacing, softness);
+                self.stack.parbreak(&self.state);
+                self.stack.push_hard(StackChild::Spacing(amount));
             }
             GenAxis::Cross => {
-                let spacing = ParChild::Spacing(amount);
-                self.stack.par.folder.push_soft(spacing, softness);
+                self.stack.par.push_hard(ParChild::Spacing(amount));
             }
         }
     }
 
-    /// Push any node into the active paragraph.
-    pub fn push_into_par(&mut self, node: impl Into<AnyNode>) {
-        let align = self.state.aligns.cross;
-        self.stack.par.folder.push(ParChild::Any(node.into(), align));
+    /// Apply a forced line break.
+    pub fn linebreak(&mut self) {
+        self.stack.par.push_hard(ParChild::Linebreak);
     }
 
-    /// Push any node directly into the stack of paragraphs.
-    ///
-    /// This finishes the active paragraph and starts a new one.
-    pub fn push_into_stack(&mut self, node: impl Into<AnyNode>) {
-        let aligns = self.state.aligns;
-        self.stack.finish_par(&self.state);
-        self.stack.folder.push(StackChild::Any(node.into(), aligns));
+    /// Apply a forced paragraph break.
+    pub fn parbreak(&mut self) {
+        let em = self.state.font.resolve_size();
+        let amount = self.state.par.spacing.resolve(em);
+        self.stack.parbreak(&self.state);
+        self.stack.push_soft(StackChild::Spacing(amount));
     }
 
-    /// Finish the active page.
-    pub fn finish_page(&mut self, keep: bool, hard: bool, source: Span) {
+    /// Apply a forced page break.
+    pub fn pagebreak(&mut self, keep: bool, hard: bool, source: Span) {
         if let Some(builder) = &mut self.page {
             let page = mem::replace(builder, PageBuilder::new(&self.state, hard));
             let stack = mem::replace(&mut self.stack, StackBuilder::new(&self.state));
@@ -158,7 +137,7 @@ impl<'a> ExecContext<'a> {
     /// Finish execution and return the created layout tree.
     pub fn finish(mut self) -> Pass<Tree> {
         assert!(self.page.is_some());
-        self.finish_page(true, false, Span::default());
+        self.pagebreak(true, false, Span::default());
         Pass::new(self.tree, self.diags)
     }
 }
@@ -189,7 +168,8 @@ impl PageBuilder {
 
 struct StackBuilder {
     dirs: Gen<Dir>,
-    folder: SoftFolder<StackChild>,
+    children: Vec<StackChild>,
+    last: Last<StackChild>,
     par: ParBuilder,
 }
 
@@ -197,20 +177,36 @@ impl StackBuilder {
     fn new(state: &State) -> Self {
         Self {
             dirs: Gen::new(Dir::TTB, state.lang.dir),
-            folder: SoftFolder::new(),
+            children: vec![],
+            last: Last::None,
             par: ParBuilder::new(state),
         }
     }
 
-    fn finish_par(&mut self, state: &State) {
+    fn push_soft(&mut self, child: StackChild) {
+        self.last.soft(child);
+    }
+
+    fn push_hard(&mut self, child: StackChild) {
+        self.last.hard();
+        self.children.push(child);
+    }
+
+    fn parbreak(&mut self, state: &State) {
         let par = mem::replace(&mut self.par, ParBuilder::new(state));
-        self.folder.extend(par.build());
+        if let Some(par) = par.build() {
+            self.children.extend(self.last.any());
+            self.children.push(par);
+        }
     }
 
     fn build(self) -> StackNode {
-        let Self { dirs, mut folder, par } = self;
-        folder.extend(par.build());
-        StackNode { dirs, children: folder.finish() }
+        let Self { dirs, mut children, par, mut last } = self;
+        if let Some(par) = par.build() {
+            children.extend(last.any());
+            children.push(par);
+        }
+        StackNode { dirs, children }
     }
 }
 
@@ -218,7 +214,8 @@ struct ParBuilder {
     aligns: Gen<Align>,
     dir: Dir,
     line_spacing: Length,
-    folder: SoftFolder<ParChild>,
+    children: Vec<ParChild>,
+    last: Last<ParChild>,
 }
 
 impl ParBuilder {
@@ -228,13 +225,43 @@ impl ParBuilder {
             aligns: state.aligns,
             dir: state.lang.dir,
             line_spacing: state.par.leading.resolve(em),
-            folder: SoftFolder::new(),
+            children: vec![],
+            last: Last::None,
         }
     }
 
+    fn push(&mut self, child: ParChild) {
+        self.children.extend(self.last.any());
+        self.children.push(child);
+    }
+
+    fn push_text(&mut self, text: String, state: &State) {
+        self.children.extend(self.last.any());
+
+        let align = state.aligns.cross;
+        let props = state.font.resolve_props();
+
+        if let Some(ParChild::Text(prev, prev_align)) = self.children.last_mut() {
+            if *prev_align == align && prev.props == props {
+                prev.text.push_str(&text);
+                return;
+            }
+        }
+
+        self.children.push(ParChild::Text(TextNode { text, props }, align));
+    }
+
+    fn push_soft(&mut self, child: ParChild) {
+        self.last.soft(child);
+    }
+
+    fn push_hard(&mut self, child: ParChild) {
+        self.last.hard();
+        self.children.push(child);
+    }
+
     fn build(self) -> Option<StackChild> {
-        let Self { aligns, dir, line_spacing, folder } = self;
-        let children = folder.finish();
+        let Self { aligns, dir, line_spacing, children, .. } = self;
         (!children.is_empty()).then(|| {
             let node = ParNode { dir, line_spacing, children };
             StackChild::Any(node.into(), aligns)
@@ -242,54 +269,28 @@ impl ParBuilder {
     }
 }
 
-/// This is used to remove leading and trailing word/line/paragraph spacing
-/// as well as collapse sequences of spacings into just one.
-struct SoftFolder<N> {
-    nodes: Vec<N>,
-    last: Last<N>,
-}
-
+/// Finite state machine for spacing coalescing.
 enum Last<N> {
     None,
-    Hard,
-    Soft(N, u8),
+    Any,
+    Soft(N),
 }
 
-impl<N> SoftFolder<N> {
-    fn new() -> Self {
-        Self { nodes: vec![], last: Last::Hard }
-    }
-
-    fn push(&mut self, node: N) {
-        let last = mem::replace(&mut self.last, Last::None);
-        if let Last::Soft(soft, _) = last {
-            self.nodes.push(soft);
-        }
-        self.nodes.push(node);
-    }
-
-    fn push_soft(&mut self, node: N, softness: u8) {
-        if softness == 0 {
-            self.last = Last::Hard;
-            self.nodes.push(node);
-        } else {
-            match self.last {
-                Last::Hard => {}
-                Last::Soft(_, other) if softness >= other => {}
-                _ => self.last = Last::Soft(node, softness),
-            }
+impl<N> Last<N> {
+    fn any(&mut self) -> Option<N> {
+        match mem::replace(self, Self::Any) {
+            Self::Soft(soft) => Some(soft),
+            _ => None,
         }
     }
 
-    fn finish(self) -> Vec<N> {
-        self.nodes
-    }
-}
-
-impl<N> Extend<N> for SoftFolder<N> {
-    fn extend<T: IntoIterator<Item = N>>(&mut self, iter: T) {
-        for elem in iter {
-            self.push(elem);
+    fn soft(&mut self, soft: N) {
+        if let Self::Any = self {
+            *self = Self::Soft(soft);
         }
     }
+
+    fn hard(&mut self) {
+        *self = Self::None;
+    }
 }

src/layout/par.rs

@@ -23,6 +23,8 @@ pub enum ParChild {
     Text(TextNode, Align),
     /// Any child node and how to align it in its line.
     Any(AnyNode, Align),
+    /// A forced linebreak.
+    Linebreak,
 }
 
 /// A consecutive, styled run of text.
@@ -55,6 +57,7 @@ impl Layout for ParNode {
                         layouter.push_frame(frame, align);
                     }
                 }
+                ParChild::Linebreak => layouter.finish_line(),
             }
         }
         layouter.finish()

src/library/align.rs

@@ -52,7 +52,7 @@ pub fn align(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
         if let Some(vertical) = vertical {
             ctx.state.aligns.main = vertical.to_align(Dir::TTB);
             if ctx.state.aligns.main != snapshot.aligns.main {
-                ctx.push_linebreak();
+                ctx.parbreak();
             }
         }
 

src/library/image.rs

@@ -23,7 +23,7 @@ pub fn image(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
             let loaded = ctx.env.resources.load(&path.v, ImageResource::parse);
             if let Some((res, img)) = loaded {
                 let dimensions = img.buf.dimensions();
-                ctx.push_into_par(ImageNode { res, dimensions, width, height });
+                ctx.push(ImageNode { res, dimensions, width, height });
             } else {
                 ctx.diag(error!(path.span, "failed to load image"));
             }

src/library/lang.rs

@@ -32,7 +32,7 @@ pub fn lang(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
             }
         }
 
-        ctx.push_parbreak();
+        ctx.parbreak();
     })
 }
 

src/library/markup.rs

@@ -14,7 +14,7 @@ use crate::syntax::{HeadingNode, RawNode};
 /// A template that inserts a line break.
 pub fn linebreak(_: &mut EvalContext, _: &mut FuncArgs) -> Value {
     Value::template(Node::LINEBREAK, move |ctx| {
-        ctx.push_linebreak();
+        ctx.linebreak();
     })
 }
 
@@ -24,7 +24,7 @@ pub fn linebreak(_: &mut EvalContext, _: &mut FuncArgs) -> Value {
 /// A template that inserts a paragraph break.
 pub fn parbreak(_: &mut EvalContext, _: &mut FuncArgs) -> Value {
     Value::template(Node::PARBREAK, move |ctx| {
-        ctx.push_parbreak();
+        ctx.parbreak();
     })
 }
 
@@ -118,7 +118,7 @@ pub fn heading(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
         body.exec(ctx);
         ctx.state = snapshot;
 
-        ctx.push_parbreak();
+        ctx.parbreak();
     })
 }
 
@@ -155,7 +155,7 @@ pub fn raw(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
 
     Value::template(Node::RAW, move |ctx| {
         if block {
-            ctx.push_parbreak();
+            ctx.parbreak();
         }
 
         let snapshot = ctx.state.clone();
@@ -164,7 +164,7 @@ pub fn raw(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
         ctx.state = snapshot;
 
         if block {
-            ctx.push_parbreak();
+            ctx.parbreak();
         }
     })
 }

src/library/pad.rs

@@ -32,6 +32,6 @@ pub fn pad(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
 
     Value::template("pad", move |ctx| {
         let child = ctx.exec_group(&body).into();
-        ctx.push_into_par(PadNode { padding, child });
+        ctx.push(PadNode { padding, child });
     })
 }

src/library/page.rs

@@ -83,13 +83,13 @@ pub fn page(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
             std::mem::swap(&mut page.size.width, &mut page.size.height);
         }
 
-        ctx.finish_page(false, true, span);
+        ctx.pagebreak(false, true, span);
 
         if let Some(body) = &body {
             // TODO: Restrict body to a single page?
             body.exec(ctx);
             ctx.state = snapshot;
-            ctx.finish_page(true, false, span);
+            ctx.pagebreak(true, false, span);
         }
     })
 }
@@ -101,6 +101,6 @@ pub fn page(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
 pub fn pagebreak(_: &mut EvalContext, args: &mut FuncArgs) -> Value {
     let span = args.span;
     Value::template("pagebreak", move |ctx| {
-        ctx.finish_page(true, true, span);
+        ctx.pagebreak(true, true, span);
     })
 }

src/library/par.rs

@@ -27,6 +27,6 @@ pub fn par(ctx: &mut EvalContext, args: &mut FuncArgs) -> Value {
             ctx.state.par.word_spacing = word_spacing;
         }
 
-        ctx.push_parbreak();
+        ctx.parbreak();
     })
 }

src/library/shapes.rs

@@ -63,13 +63,13 @@ fn rect_impl(
         let node = FixedNode { width, height, aspect, child };
 
         if let Some(color) = fill {
-            ctx.push_into_par(BackgroundNode {
+            ctx.push(BackgroundNode {
                 shape: BackgroundShape::Rect,
                 fill: Fill::Color(color),
                 child: node.into(),
             });
         } else {
-            ctx.push_into_par(node);
+            ctx.push(node);
         }
     })
 }
@@ -146,13 +146,13 @@ fn ellipse_impl(
         };
 
         if let Some(color) = fill {
-            ctx.push_into_par(BackgroundNode {
+            ctx.push(BackgroundNode {
                 shape: BackgroundShape::Ellipse,
                 fill: Fill::Color(color),
                 child: node.into(),
             });
         } else {
-            ctx.push_into_par(node);
+            ctx.push(node);
         }
     })
 }

src/library/spacing.rs

@@ -32,7 +32,7 @@ fn spacing_impl(
     Value::template(name, move |ctx| {
         if let Some(linear) = spacing {
             let amount = linear.resolve(ctx.state.font.resolve_size());
-            ctx.push_spacing(axis, amount, 0);
+            ctx.push_spacing(axis, amount);
         }
     })
 }

tests/typ/control/for.typ

@@ -14,8 +14,8 @@
 
 // Dictionary is not traversed in insertion order.
 // Should output `age: 1, name: Typst,`.
-#for k, v in (name: "Typst", age: 2) [
-    {k}: {v}, \
+#for k, v in (Name: "Typst", Age: 2) [
+    {k}: {v}.
 ]
 
 // String.

tests/typ/expand.typ

@@ -7,7 +7,7 @@
 // Top-level paragraph fills page, boxed paragraph only when width is fixed.
 L #right[R] \
 #rect(width: 50pt)[L #right[R]] \
-#rect[L #right[R]] \
+#rect[L #right[R]]
 
 // Pad inherits expansion behaviour.
 #pad[PL #right[PR]] \

tests/typ/expr/call.typ

@@ -60,7 +60,7 @@
 #args[a] \
 #args(a) \
 #args(a, [b]) \
-#args(a)[b] \
+#args(a)[b]
 
 // Template can be argument or body depending on whitespace.
 #if "template" == type[b] [Sure ]

tests/typ/library/font.typ

@@ -53,7 +53,7 @@ Emoji: 🐪, 🌋, 🏞
 #font(sans-serif: "PT Sans")
 #font(sans-serif)[Sans-serif.] \
 #font(monospace)[Monospace.] \
-#font(monospace, monospace: ("Nope", "Latin Modern Math"))[Math.] \
+#font(monospace, monospace: ("Nope", "Latin Modern Math"))[Math.]
 
 ---
 // Ref: false

tests/typ/markup/linebreak.typ

@@ -10,17 +10,11 @@ Line \ Break
 // Directly before word does not work.
 No \Break
 
----
-// Leading line break.
-\ Leading
+\ Before
 
-// Trailing before paragraph break.
-Trailing 1 \
+Multiple \ \ \
 
-Trailing 2
-
-// Trailing before end of document.
-Trailing 3 \
+Times
 
 ---
 #let linebreak() = [
@@ -28,4 +22,4 @@ Trailing 3 \
     #circle(radius: 2pt, fill: #000) \
 ]
 
-A \ B \ C \
+A \ B \ C

tests/typ/markup/nbsp.typ

@@ -1,5 +1,4 @@
 // Test the non breaking space.
 
 ---
-// Parsed correctly, but not actually doing anything at the moment.
-The non-breaking~space does not work.
+The non-breaking~space does work.

tests/typ/repr.typ

@@ -9,7 +9,7 @@
 
 {name} \
 {ke-bab} \
-{α} \
+{α}
 
 // Error: 2-3 unknown variable
 {_}
@@ -18,7 +18,7 @@
 // Literal values.
 {none} (empty) \
 {true} \
-{false} \
+{false}
 
 ---
 // Numerical values.
@@ -33,16 +33,16 @@
 {2.5rad} \
 {45deg} \
 // Not in monospace via repr.
-#repr(45deg) \
+#repr(45deg)
 
 ---
 // Colors.
-{#f7a20500} \
+{#f7a20500}
 
 ---
 // Strings and escaping.
 {"hi"} \
-{"a\n[]\"\u{1F680}string"} \
+{"a\n[]\"\u{1F680}string"}
 
 ---
 // Templates.
@@ -54,4 +54,4 @@
 
 {rect} \
 {f} \
-{() => none} \
+{() => none}

tests/typ/spacing.typ

@@ -7,7 +7,7 @@
 A#let;B \
 A#let x = 1;B  #test(x, 1) \
 A #let x = 2;B #test(x, 2) \
-A#let x = 3; B #test(x, 3) \
+A#let x = 3; B #test(x, 3)
 
 ---
 // Spacing around if-else.
@@ -17,7 +17,7 @@ A#if true [B] C \
 A #if true{"B"}C \
 A #if true{"B"} C \
 A#if false [] #else [B]C \
-A#if true [B] #else [] C \
+A#if true [B] #else [] C
 
 ---
 // Spacing around while loop.
@@ -25,11 +25,11 @@ A#if true [B] #else [] C \
 #let c = true; A#while c [{c = false}B]C \
 #let c = true; A#while c [{c = false}B] C \
 #let c = true; A #while c { c = false; "B" }C \
-#let c = true; A #while c { c = false; "B" } C \
+#let c = true; A #while c { c = false; "B" } C
 
 ---
 // Spacing around for loop.
 
 A#for _ in (none,) [B]C  \
 A#for _ in (none,) [B] C \
-A #for _ in (none,) {"B"}C \
+A #for _ in (none,) {"B"}C