Merge branch 'main' into main

Cargo.lock

@@ -413,8 +413,7 @@ dependencies = [
 [[package]]
 name = "codex"
 version = "0.1.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "724d27a0ee38b700e5e164350e79aba601a0db673ac47fce1cb74c3e38864036"
+source = "git+https://github.com/typst/codex?rev=56eb217#56eb2172fc0670f4c1c8b79a63d11f9354e5babe"
 
 [[package]]
 name = "color-print"
@@ -2921,6 +2920,7 @@ name = "typst-docs"
 version = "0.13.1"
 dependencies = [
  "clap",
+ "codex",
  "ecow",
  "heck",
  "pulldown-cmark",
@@ -3039,6 +3039,7 @@ dependencies = [
  "icu_provider_blob",
  "icu_segmenter",
  "kurbo",
+ "memchr",
  "rustybuzz",
  "smallvec",
  "ttf-parser",
@@ -3112,6 +3113,7 @@ dependencies = [
  "unicode-segmentation",
  "unscanny",
  "usvg",
+ "utf8_iter",
  "wasmi",
  "xmlwriter",
 ]
@@ -3200,6 +3202,7 @@ dependencies = [
 name = "typst-syntax"
 version = "0.13.1"
 dependencies = [
+ "comemo",
  "ecow",
  "serde",
  "toml",

Cargo.toml

@@ -47,7 +47,7 @@ clap = { version = "4.4", features = ["derive", "env", "wrap_help"] }
 clap_complete = "4.2.1"
 clap_mangen = "0.2.10"
 codespan-reporting = "0.11"
-codex = "0.1.1"
+codex = { git = "https://github.com/typst/codex", rev = "56eb217" }
 color-print = "0.3.6"
 comemo = "0.4"
 csv = "1"
@@ -135,6 +135,7 @@ unicode-segmentation = "1"
 unscanny = "0.1"
 ureq = { version = "2", default-features = false, features = ["native-tls", "gzip", "json"] }
 usvg = { version = "0.45", default-features = false, features = ["text"] }
+utf8_iter = "1.0.4"
 walkdir = "2"
 wasmi = "0.40.0"
 web-sys = "0.3"

crates/typst-cli/src/compile.rs

@@ -16,7 +16,7 @@ use typst::diag::{
 use typst::foundations::{Datetime, Smart};
 use typst::html::HtmlDocument;
 use typst::layout::{Frame, Page, PageRanges, PagedDocument};
-use typst::syntax::{FileId, Source, Span};
+use typst::syntax::{FileId, Lines, Span};
 use typst::WorldExt;
 use typst_pdf::{PdfOptions, PdfStandards, Timestamp};
 
@@ -696,7 +696,7 @@ fn label(world: &SystemWorld, span: Span) -> Option<Label<FileId>> {
 impl<'a> codespan_reporting::files::Files<'a> for SystemWorld {
     type FileId = FileId;
     type Name = String;
-    type Source = Source;
+    type Source = Lines<String>;
 
     fn name(&'a self, id: FileId) -> CodespanResult<Self::Name> {
         let vpath = id.vpath();

crates/typst-cli/src/query.rs

@@ -5,9 +5,9 @@ use typst::diag::{bail, HintedStrResult, StrResult, Warned};
 use typst::engine::Sink;
 use typst::foundations::{Content, IntoValue, LocatableSelector, Scope};
 use typst::layout::PagedDocument;
-use typst::syntax::Span;
+use typst::syntax::{Span, SyntaxMode};
 use typst::World;
-use typst_eval::{eval_string, EvalMode};
+use typst_eval::eval_string;
 
 use crate::args::{QueryCommand, SerializationFormat};
 use crate::compile::print_diagnostics;
@@ -63,7 +63,7 @@ fn retrieve(
         Sink::new().track_mut(),
         &command.selector,
         Span::detached(),
-        EvalMode::Code,
+        SyntaxMode::Code,
         Scope::default(),
     )
     .map_err(|errors| {

crates/typst-cli/src/timings.rs

@@ -85,6 +85,6 @@ fn resolve_span(world: &SystemWorld, span: Span) -> Option<(String, u32)> {
     let id = span.id()?;
     let source = world.source(id).ok()?;
     let range = source.range(span)?;
-    let line = source.byte_to_line(range.start)?;
+    let line = source.lines().byte_to_line(range.start)?;
     Some((format!("{id:?}"), line as u32 + 1))
 }

crates/typst-cli/src/watch.rs

@@ -10,11 +10,12 @@ use codespan_reporting::term::{self, termcolor};
 use ecow::eco_format;
 use notify::{Event, RecommendedWatcher, RecursiveMode, Watcher as _};
 use same_file::is_same_file;
-use typst::diag::{bail, StrResult};
+use typst::diag::{bail, warning, StrResult};
+use typst::syntax::Span;
 use typst::utils::format_duration;
 
 use crate::args::{Input, Output, WatchCommand};
-use crate::compile::{compile_once, CompileConfig};
+use crate::compile::{compile_once, print_diagnostics, CompileConfig};
 use crate::timings::Timer;
 use crate::world::{SystemWorld, WorldCreationError};
 use crate::{print_error, terminal};
@@ -55,6 +56,11 @@ pub fn watch(timer: &mut Timer, command: &WatchCommand) -> StrResult<()> {
     // Perform initial compilation.
     timer.record(&mut world, |world| compile_once(world, &mut config))??;
 
+    // Print warning when trying to watch stdin.
+    if matches!(&config.input, Input::Stdin) {
+        warn_watching_std(&world, &config)?;
+    }
+
     // Recompile whenever something relevant happens.
     loop {
         // Watch all dependencies of the most recent compilation.
@@ -332,3 +338,15 @@ impl Status {
         }
     }
 }
+
+/// Emits a warning when trying to watch stdin.
+fn warn_watching_std(world: &SystemWorld, config: &CompileConfig) -> StrResult<()> {
+    let warning = warning!(
+        Span::detached(),
+        "cannot watch changes for stdin";
+        hint: "to recompile on changes, watch a regular file instead";
+        hint: "to compile once and exit, please use `typst compile` instead"
+    );
+    print_diagnostics(world, &[], &[warning], config.diagnostic_format)
+        .map_err(|err| eco_format!("failed to print diagnostics ({err})"))
+}

crates/typst-cli/src/world.rs

@@ -9,7 +9,7 @@ use ecow::{eco_format, EcoString};
 use parking_lot::Mutex;
 use typst::diag::{FileError, FileResult};
 use typst::foundations::{Bytes, Datetime, Dict, IntoValue};
-use typst::syntax::{FileId, Source, VirtualPath};
+use typst::syntax::{FileId, Lines, Source, VirtualPath};
 use typst::text::{Font, FontBook};
 use typst::utils::LazyHash;
 use typst::{Library, World};
@@ -181,10 +181,20 @@ impl SystemWorld {
         }
     }
 
-    /// Lookup a source file by id.
+    /// Lookup line metadata for a file by id.
     #[track_caller]
-    pub fn lookup(&self, id: FileId) -> Source {
-        self.source(id).expect("file id does not point to any source file")
+    pub fn lookup(&self, id: FileId) -> Lines<String> {
+        self.slot(id, |slot| {
+            if let Some(source) = slot.source.get() {
+                let source = source.as_ref().expect("file is not valid");
+                source.lines().clone()
+            } else if let Some(bytes) = slot.file.get() {
+                let bytes = bytes.as_ref().expect("file is not valid");
+                Lines::try_from(bytes).expect("file is not valid utf-8")
+            } else {
+                panic!("file id does not point to any source file");
+            }
+        })
     }
 }
 
@@ -339,6 +349,11 @@ impl<T: Clone> SlotCell<T> {
         self.accessed = false;
     }
 
+    /// Gets the contents of the cell.
+    fn get(&self) -> Option<&FileResult<T>> {
+        self.data.as_ref()
+    }
+
     /// Gets the contents of the cell or initialize them.
     fn get_or_init(
         &mut self,

crates/typst-eval/src/lib.rs

@@ -18,7 +18,6 @@ pub use self::call::{eval_closure, CapturesVisitor};
 pub use self::flow::FlowEvent;
 pub use self::import::import;
 pub use self::vm::Vm;
-pub use typst_library::routines::EvalMode;
 
 use self::access::*;
 use self::binding::*;
@@ -32,7 +31,7 @@ use typst_library::introspection::Introspector;
 use typst_library::math::EquationElem;
 use typst_library::routines::Routines;
 use typst_library::World;
-use typst_syntax::{ast, parse, parse_code, parse_math, Source, Span};
+use typst_syntax::{ast, parse, parse_code, parse_math, Source, Span, SyntaxMode};
 
 /// Evaluate a source file and return the resulting module.
 #[comemo::memoize]
@@ -104,13 +103,13 @@ pub fn eval_string(
     sink: TrackedMut<Sink>,
     string: &str,
     span: Span,
-    mode: EvalMode,
+    mode: SyntaxMode,
     scope: Scope,
 ) -> SourceResult<Value> {
     let mut root = match mode {
-        EvalMode::Code => parse_code(string),
-        EvalMode::Markup => parse(string),
-        EvalMode::Math => parse_math(string),
+        SyntaxMode::Code => parse_code(string),
+        SyntaxMode::Markup => parse(string),
+        SyntaxMode::Math => parse_math(string),
     };
 
     root.synthesize(span);
@@ -141,11 +140,11 @@ pub fn eval_string(
 
     // Evaluate the code.
     let output = match mode {
-        EvalMode::Code => root.cast::<ast::Code>().unwrap().eval(&mut vm)?,
-        EvalMode::Markup => {
+        SyntaxMode::Code => root.cast::<ast::Code>().unwrap().eval(&mut vm)?,
+        SyntaxMode::Markup => {
             Value::Content(root.cast::<ast::Markup>().unwrap().eval(&mut vm)?)
         }
-        EvalMode::Math => Value::Content(
+        SyntaxMode::Math => Value::Content(
             EquationElem::new(root.cast::<ast::Math>().unwrap().eval(&mut vm)?)
                 .with_block(false)
                 .pack()

crates/typst-ide/src/complete.rs

@@ -298,13 +298,20 @@ fn complete_math(ctx: &mut CompletionContext) -> bool {
         return false;
     }
 
-    // Start of an interpolated identifier: "#|".
+    // Start of an interpolated identifier: "$#|$".
     if ctx.leaf.kind() == SyntaxKind::Hash {
         ctx.from = ctx.cursor;
         code_completions(ctx, true);
         return true;
     }
 
+    // Behind existing interpolated identifier: "$#pa|$".
+    if ctx.leaf.kind() == SyntaxKind::Ident {
+        ctx.from = ctx.leaf.offset();
+        code_completions(ctx, true);
+        return true;
+    }
+
     // Behind existing atom or identifier: "$a|$" or "$abc|$".
     if matches!(
         ctx.leaf.kind(),
@@ -1666,6 +1673,13 @@ mod tests {
         test("#{() .a}", -2).must_include(["at", "any", "all"]);
     }
 
+    /// Test that autocomplete in math uses the correct global scope.
+    #[test]
+    fn test_autocomplete_math_scope() {
+        test("$#col$", -2).must_include(["colbreak"]).must_exclude(["colon"]);
+        test("$col$", -2).must_include(["colon"]).must_exclude(["colbreak"]);
+    }
+
     /// Test that the `before_window` doesn't slice into invalid byte
     /// boundaries.
     #[test]
@@ -1684,7 +1698,7 @@ mod tests {
 
         // Then, add the invalid `#cite` call. Had the document been invalid
         // initially, we would have no populated document to autocomplete with.
-        let end = world.main.len_bytes();
+        let end = world.main.text().len();
         world.main.edit(end..end, " #cite()");
 
         test_with_doc(&world, -2, doc.as_ref())

crates/typst-ide/src/tests.rs

@@ -228,7 +228,7 @@ impl FilePos for (&str, isize) {
 #[track_caller]
 fn cursor(source: &Source, cursor: isize) -> usize {
     if cursor < 0 {
-        source.len_bytes().checked_add_signed(cursor + 1).unwrap()
+        source.text().len().checked_add_signed(cursor + 1).unwrap()
     } else {
         cursor as usize
     }

crates/typst-ide/src/utils.rs

@@ -114,7 +114,9 @@ pub fn globals<'a>(world: &'a dyn IdeWorld, leaf: &LinkedNode) -> &'a Scope {
             | Some(SyntaxKind::Math)
             | Some(SyntaxKind::MathFrac)
             | Some(SyntaxKind::MathAttach)
-    );
+    ) && leaf
+        .prev_leaf()
+        .is_none_or(|prev| !matches!(prev.kind(), SyntaxKind::Hash));
 
     let library = world.library();
     if in_math {

crates/typst-layout/Cargo.toml

@@ -30,6 +30,7 @@ icu_provider_adapters = { workspace = true }
 icu_provider_blob = { workspace = true }
 icu_segmenter = { workspace = true }
 kurbo = { workspace = true }
+memchr = { workspace = true }
 rustybuzz = { workspace = true }
 smallvec = { workspace = true }
 ttf-parser = { workspace = true }

crates/typst-layout/src/grid/layouter.rs

@@ -3,7 +3,9 @@ use std::fmt::Debug;
 use typst_library::diag::{bail, SourceResult};
 use typst_library::engine::Engine;
 use typst_library::foundations::{Resolve, StyleChain};
-use typst_library::layout::grid::resolve::{Cell, CellGrid, LinePosition, Repeatable};
+use typst_library::layout::grid::resolve::{
+    Cell, CellGrid, Header, LinePosition, Repeatable,
+};
 use typst_library::layout::{
     Abs, Axes, Dir, Fr, Fragment, Frame, FrameItem, Length, Point, Region, Regions, Rel,
     Size, Sizing,
@@ -30,10 +32,8 @@ pub struct GridLayouter<'a> {
     pub(super) rcols: Vec<Abs>,
     /// The sum of `rcols`.
     pub(super) width: Abs,
-    /// Resolve row sizes, by region.
+    /// Resolved row sizes, by region.
     pub(super) rrows: Vec<Vec<RowPiece>>,
-    /// Rows in the current region.
-    pub(super) lrows: Vec<Row>,
     /// The amount of unbreakable rows remaining to be laid out in the
     /// current unbreakable row group. While this is positive, no region breaks
     /// should occur.
@@ -41,24 +41,155 @@ pub struct GridLayouter<'a> {
     /// Rowspans not yet laid out because not all of their spanned rows were
     /// laid out yet.
     pub(super) rowspans: Vec<Rowspan>,
-    /// The initial size of the current region before we started subtracting.
-    pub(super) initial: Size,
+    /// Grid layout state for the current region.
+    pub(super) current: Current,
     /// Frames for finished regions.
     pub(super) finished: Vec<Frame>,
+    /// The amount and height of header rows on each finished region.
+    pub(super) finished_header_rows: Vec<FinishedHeaderRowInfo>,
     /// Whether this is an RTL grid.
     pub(super) is_rtl: bool,
-    /// The simulated header height.
-    /// This field is reset in `layout_header` and properly updated by
+    /// Currently repeating headers, one per level. Sorted by increasing
+    /// levels.
+    ///
+    /// Note that some levels may be absent, in particular level 0, which does
+    /// not exist (so all levels are >= 1).
+    pub(super) repeating_headers: Vec<&'a Header>,
+    /// Headers, repeating or not, awaiting their first successful layout.
+    /// Sorted by increasing levels.
+    pub(super) pending_headers: &'a [Repeatable<Header>],
+    /// Next headers to be processed.
+    pub(super) upcoming_headers: &'a [Repeatable<Header>],
+    /// State of the row being currently laid out.
+    ///
+    /// This is kept as a field to avoid passing down too many parameters from
+    /// `layout_row` into called functions, which would then have to pass them
+    /// down to `push_row`, which reads these values.
+    pub(super) row_state: RowState,
+    /// The span of the grid element.
+    pub(super) span: Span,
+}
+
+/// Grid layout state for the current region. This should be reset or updated
+/// on each region break.
+pub(super) struct Current {
+    /// The initial size of the current region before we started subtracting.
+    pub(super) initial: Size,
+    /// The height of the region after repeated headers were placed and footers
+    /// prepared. This also includes pending repeating headers from the start,
+    /// even if they were not repeated yet, since they will be repeated in the
+    /// next region anyway (bar orphan prevention).
+    ///
+    /// This is used to quickly tell if any additional space in the region has
+    /// been occupied since then, meaning that additional space will become
+    /// available after a region break (see
+    /// [`GridLayouter::may_progress_with_repeats`]).
+    pub(super) initial_after_repeats: Abs,
+    /// Whether `layouter.regions.may_progress()` was `true` at the top of the
+    /// region.
+    pub(super) could_progress_at_top: bool,
+    /// Rows in the current region.
+    pub(super) lrows: Vec<Row>,
+    /// The amount of repeated header rows at the start of the current region.
+    /// Thus, excludes rows from pending headers (which were placed for the
+    /// first time).
+    ///
+    /// Note that `repeating_headers` and `pending_headers` can change if we
+    /// find a new header inside the region (not at the top), so this field
+    /// is required to access information from the top of the region.
+    ///
+    /// This information is used on finish region to calculate the total height
+    /// of resolved header rows at the top of the region, which is used by
+    /// multi-page rowspans so they can properly skip the header rows at the
+    /// top of each region during layout.
+    pub(super) repeated_header_rows: usize,
+    /// The end bound of the row range of the last repeating header at the
+    /// start of the region.
+    ///
+    /// The last row might have disappeared from layout due to being empty, so
+    /// this is how we can become aware of where the last header ends without
+    /// having to check the vector of rows. Line layout uses this to determine
+    /// when to prioritize the last lines under a header.
+    ///
+    /// A value of zero indicates no repeated headers were placed.
+    pub(super) last_repeated_header_end: usize,
+    /// Stores the length of `lrows` before a sequence of rows equipped with
+    /// orphan prevention was laid out. In this case, if no more rows without
+    /// orphan prevention are laid out after those rows before the region ends,
+    /// the rows will be removed, and there may be an attempt to place them
+    /// again in the new region. Effectively, this is the mechanism used for
+    /// orphan prevention of rows.
+    ///
+    /// At the moment, this is only used by repeated headers (they aren't laid
+    /// out if alone in the region) and by new headers, which are moved to the
+    /// `pending_headers` vector and so will automatically be placed again
+    /// until they fit and are not orphans in at least one region (or exactly
+    /// one, for non-repeated headers).
+    pub(super) lrows_orphan_snapshot: Option<usize>,
+    /// The height of effectively repeating headers, that is, ignoring
+    /// non-repeating pending headers, in the current region.
+    ///
+    /// This is used by multi-page auto rows so they can inform cell layout on
+    /// how much space should be taken by headers if they break across regions.
+    /// In particular, non-repeating headers only occupy the initial region,
+    /// but disappear on new regions, so they can be ignored.
+    ///
+    /// This field is reset on each new region and properly updated by
     /// `layout_auto_row` and `layout_relative_row`, and should not be read
     /// before all header rows are fully laid out. It is usually fine because
     /// header rows themselves are unbreakable, and unbreakable rows do not
     /// need to read this field at all.
-    pub(super) header_height: Abs,
+    ///
+    /// This height is not only computed at the beginning of the region. It is
+    /// updated whenever a new header is found, subtracting the height of
+    /// headers which stopped repeating and adding the height of all new
+    /// headers.
+    pub(super) repeating_header_height: Abs,
+    /// The height for each repeating header that was placed in this region.
+    /// Note that this includes headers not at the top of the region, before
+    /// their first repetition (pending headers), and excludes headers removed
+    /// by virtue of a new, conflicting header being found (short-lived
+    /// headers).
+    ///
+    /// This is used to know how much to update `repeating_header_height` by
+    /// when finding a new header and causing existing repeating headers to
+    /// stop.
+    pub(super) repeating_header_heights: Vec<Abs>,
     /// The simulated footer height for this region.
+    ///
     /// The simulation occurs before any rows are laid out for a region.
     pub(super) footer_height: Abs,
-    /// The span of the grid element.
-    pub(super) span: Span,
+}
+
+/// Data about the row being laid out right now.
+#[derive(Debug, Default)]
+pub(super) struct RowState {
+    /// If this is `Some`, this will be updated by the currently laid out row's
+    /// height if it is auto or relative. This is used for header height
+    /// calculation.
+    pub(super) current_row_height: Option<Abs>,
+    /// This is `true` when laying out non-short lived headers and footers.
+    /// That is, headers and footers which are not immediately followed or
+    /// preceded (respectively) by conflicting headers and footers of same or
+    /// lower level, or the end or start of the table (respectively), which
+    /// would cause them to never repeat, even once.
+    ///
+    /// If this is `false`, the next row to be laid out will remove an active
+    /// orphan snapshot and will flush pending headers, as there is no risk
+    /// that they will be orphans anymore.
+    pub(super) in_active_repeatable: bool,
+}
+
+/// Data about laid out repeated header rows for a specific finished region.
+#[derive(Debug, Default)]
+pub(super) struct FinishedHeaderRowInfo {
+    /// The amount of repeated headers at the top of the region.
+    pub(super) repeated_amount: usize,
+    /// The end bound of the row range of the last repeated header at the top
+    /// of the region.
+    pub(super) last_repeated_header_end: usize,
+    /// The total height of repeated headers at the top of the region.
+    pub(super) repeated_height: Abs,
 }
 
 /// Details about a resulting row piece.
@@ -114,14 +245,27 @@ impl<'a> GridLayouter<'a> {
             rcols: vec![Abs::zero(); grid.cols.len()],
             width: Abs::zero(),
             rrows: vec![],
-            lrows: vec![],
             unbreakable_rows_left: 0,
             rowspans: vec![],
-            initial: regions.size,
             finished: vec![],
+            finished_header_rows: vec![],
             is_rtl: TextElem::dir_in(styles) == Dir::RTL,
-            header_height: Abs::zero(),
-            footer_height: Abs::zero(),
+            repeating_headers: vec![],
+            upcoming_headers: &grid.headers,
+            pending_headers: Default::default(),
+            row_state: RowState::default(),
+            current: Current {
+                initial: regions.size,
+                initial_after_repeats: regions.size.y,
+                could_progress_at_top: regions.may_progress(),
+                lrows: vec![],
+                repeated_header_rows: 0,
+                last_repeated_header_end: 0,
+                lrows_orphan_snapshot: None,
+                repeating_header_height: Abs::zero(),
+                repeating_header_heights: vec![],
+                footer_height: Abs::zero(),
+            },
             span,
         }
     }
@@ -130,38 +274,57 @@ impl<'a> GridLayouter<'a> {
     pub fn layout(mut self, engine: &mut Engine) -> SourceResult<Fragment> {
         self.measure_columns(engine)?;
 
-        if let Some(Repeatable::Repeated(footer)) = &self.grid.footer {
-            // Ensure rows in the first region will be aware of the possible
-            // presence of the footer.
-            self.prepare_footer(footer, engine, 0)?;
-            if matches!(self.grid.header, None | Some(Repeatable::NotRepeated(_))) {
-                // No repeatable header, so we won't subtract it later.
-                self.regions.size.y -= self.footer_height;
+        if let Some(footer) = &self.grid.footer {
+            if footer.repeated {
+                // Ensure rows in the first region will be aware of the
+                // possible presence of the footer.
+                self.prepare_footer(footer, engine, 0)?;
+                self.regions.size.y -= self.current.footer_height;
+                self.current.initial_after_repeats = self.regions.size.y;
             }
         }
 
-        for y in 0..self.grid.rows.len() {
-            if let Some(Repeatable::Repeated(header)) = &self.grid.header {
-                if y < header.end {
-                    if y == 0 {
-                        self.layout_header(header, engine, 0)?;
-                        self.regions.size.y -= self.footer_height;
-                    }
+        let mut y = 0;
+        let mut consecutive_header_count = 0;
+        while y < self.grid.rows.len() {
+            if let Some(next_header) = self.upcoming_headers.get(consecutive_header_count)
+            {
+                if next_header.range.contains(&y) {
+                    self.place_new_headers(&mut consecutive_header_count, engine)?;
+                    y = next_header.range.end;
+
                     // Skip header rows during normal layout.
                     continue;
                 }
             }
 
-            if let Some(Repeatable::Repeated(footer)) = &self.grid.footer {
-                if y >= footer.start {
+            if let Some(footer) = &self.grid.footer {
+                if footer.repeated && y >= footer.start {
                     if y == footer.start {
                         self.layout_footer(footer, engine, self.finished.len())?;
+                        self.flush_orphans();
                     }
+                    y = footer.end;
                     continue;
                 }
             }
 
             self.layout_row(y, engine, 0)?;
+
+            // After the first non-header row is placed, pending headers are no
+            // longer orphans and can repeat, so we move them to repeating
+            // headers.
+            //
+            // Note that this is usually done in `push_row`, since the call to
+            // `layout_row` above might trigger region breaks (for multi-page
+            // auto rows), whereas this needs to be called as soon as any part
+            // of a row is laid out. However, it's possible a row has no
+            // visible output and thus does not push any rows even though it
+            // was successfully laid out, in which case we additionally flush
+            // here just in case.
+            self.flush_orphans();
+
+            y += 1;
         }
 
         self.finish_region(engine, true)?;
@@ -184,12 +347,46 @@ impl<'a> GridLayouter<'a> {
         self.render_fills_strokes()
     }
 
-    /// Layout the given row.
+    /// Layout a row with a certain initial state, returning the final state.
+    #[inline]
+    pub(super) fn layout_row_with_state(
+        &mut self,
+        y: usize,
+        engine: &mut Engine,
+        disambiguator: usize,
+        initial_state: RowState,
+    ) -> SourceResult<RowState> {
+        // Keep a copy of the previous value in the stack, as this function can
+        // call itself recursively (e.g. if a region break is triggered and a
+        // header is placed), so we shouldn't outright overwrite it, but rather
+        // save and later restore the state when back to this call.
+        let previous = std::mem::replace(&mut self.row_state, initial_state);
+
+        // Keep it as a separate function to allow inlining the return below,
+        // as it's usually not needed.
+        self.layout_row_internal(y, engine, disambiguator)?;
+
+        Ok(std::mem::replace(&mut self.row_state, previous))
+    }
+
+    /// Layout the given row with the default row state.
+    #[inline]
     pub(super) fn layout_row(
         &mut self,
         y: usize,
         engine: &mut Engine,
         disambiguator: usize,
+    ) -> SourceResult<()> {
+        self.layout_row_with_state(y, engine, disambiguator, RowState::default())?;
+        Ok(())
+    }
+
+    /// Layout the given row using the current state.
+    pub(super) fn layout_row_internal(
+        &mut self,
+        y: usize,
+        engine: &mut Engine,
+        disambiguator: usize,
     ) -> SourceResult<()> {
         // Skip to next region if current one is full, but only for content
         // rows, not for gutter rows, and only if we aren't laying out an
@@ -206,13 +403,18 @@ impl<'a> GridLayouter<'a> {
         }
 
         // Don't layout gutter rows at the top of a region.
-        if is_content_row || !self.lrows.is_empty() {
+        if is_content_row || !self.current.lrows.is_empty() {
             match self.grid.rows[y] {
                 Sizing::Auto => self.layout_auto_row(engine, disambiguator, y)?,
                 Sizing::Rel(v) => {
                     self.layout_relative_row(engine, disambiguator, v, y)?
                 }
-                Sizing::Fr(v) => self.lrows.push(Row::Fr(v, y, disambiguator)),
+                Sizing::Fr(v) => {
+                    if !self.row_state.in_active_repeatable {
+                        self.flush_orphans();
+                    }
+                    self.current.lrows.push(Row::Fr(v, y, disambiguator))
+                }
             }
         }
 
@@ -225,8 +427,13 @@ impl<'a> GridLayouter<'a> {
     fn render_fills_strokes(mut self) -> SourceResult<Fragment> {
         let mut finished = std::mem::take(&mut self.finished);
         let frame_amount = finished.len();
-        for ((frame_index, frame), rows) in
-            finished.iter_mut().enumerate().zip(&self.rrows)
+        for (((frame_index, frame), rows), finished_header_rows) in
+            finished.iter_mut().enumerate().zip(&self.rrows).zip(
+                self.finished_header_rows
+                    .iter()
+                    .map(Some)
+                    .chain(std::iter::repeat(None)),
+            )
         {
             if self.rcols.is_empty() || rows.is_empty() {
                 continue;
@@ -347,7 +554,8 @@ impl<'a> GridLayouter<'a> {
             let hline_indices = rows
                 .iter()
                 .map(|piece| piece.y)
-                .chain(std::iter::once(self.grid.rows.len()));
+                .chain(std::iter::once(self.grid.rows.len()))
+                .enumerate();
 
             // Converts a row to the corresponding index in the vector of
             // hlines.
@@ -372,7 +580,7 @@ impl<'a> GridLayouter<'a> {
             };
 
             let mut prev_y = None;
-            for (y, dy) in hline_indices.zip(hline_offsets) {
+            for ((i, y), dy) in hline_indices.zip(hline_offsets) {
                 // Position of lines below the row index in the previous iteration.
                 let expected_prev_line_position = prev_y
                     .map(|prev_y| {
@@ -383,47 +591,40 @@ impl<'a> GridLayouter<'a> {
                     })
                     .unwrap_or(LinePosition::Before);
 
-                // FIXME: In the future, directly specify in 'self.rrows' when
-                // we place a repeated header rather than its original rows.
-                // That would let us remove most of those verbose checks, both
-                // in 'lines.rs' and here. Those checks also aren't fully
-                // accurate either, since they will also trigger when some rows
-                // have been removed between the header and what's below it.
-                let is_under_repeated_header = self
-                    .grid
-                    .header
-                    .as_ref()
-                    .and_then(Repeatable::as_repeated)
-                    .zip(prev_y)
-                    .is_some_and(|(header, prev_y)| {
-                        // Note: 'y == header.end' would mean we're right below
-                        // the NON-REPEATED header, so that case should return
-                        // false.
-                        prev_y < header.end && y > header.end
-                    });
+                // Header's lines at the bottom have priority when repeated.
+                // This will store the end bound of the last header if the
+                // current iteration is calculating lines under it.
+                let last_repeated_header_end_above = match finished_header_rows {
+                    Some(info) if prev_y.is_some() && i == info.repeated_amount => {
+                        Some(info.last_repeated_header_end)
+                    }
+                    _ => None,
+                };
 
                 // If some grid rows were omitted between the previous resolved
                 // row and the current one, we ensure lines below the previous
                 // row don't "disappear" and are considered, albeit with less
                 // priority. However, don't do this when we're below a header,
                 // as it must have more priority instead of less, so it is
-                // chained later instead of before. The exception is when the
+                // chained later instead of before (stored in the
+                // 'header_hlines' variable below). The exception is when the
                 // last row in the header is removed, in which case we append
                 // both the lines under the row above us and also (later) the
                 // lines under the header's (removed) last row.
-                let prev_lines = prev_y
-                    .filter(|prev_y| {
-                        prev_y + 1 != y
-                            && (!is_under_repeated_header
-                                || self
-                                    .grid
-                                    .header
-                                    .as_ref()
-                                    .and_then(Repeatable::as_repeated)
-                                    .is_some_and(|header| prev_y + 1 != header.end))
-                    })
-                    .map(|prev_y| get_hlines_at(prev_y + 1))
-                    .unwrap_or(&[]);
+                let prev_lines = match prev_y {
+                    Some(prev_y)
+                        if prev_y + 1 != y
+                            && last_repeated_header_end_above.is_none_or(
+                                |last_repeated_header_end| {
+                                    prev_y + 1 != last_repeated_header_end
+                                },
+                            ) =>
+                    {
+                        get_hlines_at(prev_y + 1)
+                    }
+
+                    _ => &[],
+                };
 
                 let expected_hline_position =
                     expected_line_position(y, y == self.grid.rows.len());
@@ -441,15 +642,13 @@ impl<'a> GridLayouter<'a> {
                 };
 
                 let mut expected_header_line_position = LinePosition::Before;
-                let header_hlines = if let Some((Repeatable::Repeated(header), prev_y)) =
-                    self.grid.header.as_ref().zip(prev_y)
-                {
-                    if is_under_repeated_header
-                        && (!self.grid.has_gutter
+                let header_hlines = match (last_repeated_header_end_above, prev_y) {
+                    (Some(header_end_above), Some(prev_y))
+                        if !self.grid.has_gutter
                             || matches!(
                                 self.grid.rows[prev_y],
                                 Sizing::Rel(length) if length.is_zero()
-                            ))
+                            ) =>
                     {
                         // For lines below a header, give priority to the
                         // lines originally below the header rather than
@@ -468,15 +667,13 @@ impl<'a> GridLayouter<'a> {
                         // column-gutter is specified, for example. In that
                         // case, we still repeat the line under the gutter.
                         expected_header_line_position = expected_line_position(
-                            header.end,
-                            header.end == self.grid.rows.len(),
+                            header_end_above,
+                            header_end_above == self.grid.rows.len(),
                         );
-                        get_hlines_at(header.end)
-                    } else {
-                        &[]
+                        get_hlines_at(header_end_above)
                     }
-                } else {
-                    &[]
+
+                    _ => &[],
                 };
 
                 // The effective hlines to be considered at this row index are
@@ -529,6 +726,7 @@ impl<'a> GridLayouter<'a> {
                             grid,
                             rows,
                             local_top_y,
+                            last_repeated_header_end_above,
                             in_last_region,
                             y,
                             x,
@@ -941,15 +1139,9 @@ impl<'a> GridLayouter<'a> {
             let frame = self.layout_single_row(engine, disambiguator, first, y)?;
             self.push_row(frame, y, true);
 
-            if self
-                .grid
-                .header
-                .as_ref()
-                .and_then(Repeatable::as_repeated)
-                .is_some_and(|header| y < header.end)
-            {
-                // Add to header height.
-                self.header_height += first;
+            if let Some(row_height) = &mut self.row_state.current_row_height {
+                // Add to header height, as we are in a header row.
+                *row_height += first;
             }
 
             return Ok(());
@@ -958,19 +1150,21 @@ impl<'a> GridLayouter<'a> {
         // Expand all but the last region.
         // Skip the first region if the space is eaten up by an fr row.
         let len = resolved.len();
-        for ((i, region), target) in self
-            .regions
-            .iter()
-            .enumerate()
-            .zip(&mut resolved[..len - 1])
-            .skip(self.lrows.iter().any(|row| matches!(row, Row::Fr(..))) as usize)
+        for ((i, region), target) in
+            self.regions
+                .iter()
+                .enumerate()
+                .zip(&mut resolved[..len - 1])
+                .skip(self.current.lrows.iter().any(|row| matches!(row, Row::Fr(..)))
+                    as usize)
         {
             // Subtract header and footer heights from the region height when
-            // it's not the first.
+            // it's not the first. Ignore non-repeating headers as they only
+            // appear on the first region by definition.
             target.set_max(
                 region.y
                     - if i > 0 {
-                        self.header_height + self.footer_height
+                        self.current.repeating_header_height + self.current.footer_height
                     } else {
                         Abs::zero()
                     },
@@ -1181,25 +1375,19 @@ impl<'a> GridLayouter<'a> {
         let resolved = v.resolve(self.styles).relative_to(self.regions.base().y);
         let frame = self.layout_single_row(engine, disambiguator, resolved, y)?;
 
-        if self
-            .grid
-            .header
-            .as_ref()
-            .and_then(Repeatable::as_repeated)
-            .is_some_and(|header| y < header.end)
-        {
-            // Add to header height.
-            self.header_height += resolved;
+        if let Some(row_height) = &mut self.row_state.current_row_height {
+            // Add to header height, as we are in a header row.
+            *row_height += resolved;
         }
 
         // Skip to fitting region, but only if we aren't part of an unbreakable
-        // row group. We use 'in_last_with_offset' so our 'in_last' call
-        // properly considers that a header and a footer would be added on each
-        // region break.
+        // row group. We use 'may_progress_with_repeats' to stop trying if we
+        // would skip to a region with the same height and where the same
+        // headers would be repeated.
         let height = frame.height();
         while self.unbreakable_rows_left == 0
             && !self.regions.size.y.fits(height)
-            && !in_last_with_offset(self.regions, self.header_height + self.footer_height)
+            && self.may_progress_with_repeats()
         {
             self.finish_region(engine, false)?;
 
@@ -1323,8 +1511,13 @@ impl<'a> GridLayouter<'a> {
     /// will be pushed for this particular row. It can be `false` for rows
     /// spanning multiple regions.
     fn push_row(&mut self, frame: Frame, y: usize, is_last: bool) {
+        if !self.row_state.in_active_repeatable {
+            // There is now a row after the rows equipped with orphan
+            // prevention, so no need to keep moving them anymore.
+            self.flush_orphans();
+        }
         self.regions.size.y -= frame.height();
-        self.lrows.push(Row::Frame(frame, y, is_last));
+        self.current.lrows.push(Row::Frame(frame, y, is_last));
     }
 
     /// Finish rows for one region.
@@ -1333,68 +1526,73 @@ impl<'a> GridLayouter<'a> {
         engine: &mut Engine,
         last: bool,
     ) -> SourceResult<()> {
+        // The latest rows have orphan prevention (headers) and no other rows
+        // were placed, so remove those rows and try again in a new region,
+        // unless this is the last region.
+        if let Some(orphan_snapshot) = self.current.lrows_orphan_snapshot.take() {
+            if !last {
+                self.current.lrows.truncate(orphan_snapshot);
+                self.current.repeated_header_rows =
+                    self.current.repeated_header_rows.min(orphan_snapshot);
+
+                if orphan_snapshot == 0 {
+                    // Removed all repeated headers.
+                    self.current.last_repeated_header_end = 0;
+                }
+            }
+        }
+
         if self
+            .current
             .lrows
             .last()
             .is_some_and(|row| self.grid.is_gutter_track(row.index()))
         {
             // Remove the last row in the region if it is a gutter row.
-            self.lrows.pop().unwrap();
+            self.current.lrows.pop().unwrap();
+            self.current.repeated_header_rows =
+                self.current.repeated_header_rows.min(self.current.lrows.len());
         }
 
-        // If no rows other than the footer have been laid out so far, and
-        // there are rows beside the footer, then don't lay it out at all.
-        // This check doesn't apply, and is thus overridden, when there is a
-        // header.
-        let mut footer_would_be_orphan = self.lrows.is_empty()
-            && !in_last_with_offset(
-                self.regions,
-                self.header_height + self.footer_height,
-            )
-            && self
-                .grid
-                .footer
-                .as_ref()
-                .and_then(Repeatable::as_repeated)
-                .is_some_and(|footer| footer.start != 0);
-
-        if let Some(Repeatable::Repeated(header)) = &self.grid.header {
-            if self.grid.rows.len() > header.end
-                && self
-                    .grid
-                    .footer
-                    .as_ref()
-                    .and_then(Repeatable::as_repeated)
-                    .is_none_or(|footer| footer.start != header.end)
-                && self.lrows.last().is_some_and(|row| row.index() < header.end)
-                && !in_last_with_offset(
-                    self.regions,
-                    self.header_height + self.footer_height,
-                )
-            {
-                // Header and footer would be alone in this region, but there are more
-                // rows beyond the header and the footer. Push an empty region.
-                self.lrows.clear();
-                footer_would_be_orphan = true;
-            }
-        }
+        // If no rows other than the footer have been laid out so far
+        // (e.g. due to header orphan prevention), and there are rows
+        // beside the footer, then don't lay it out at all.
+        //
+        // It is worth noting that the footer is made non-repeatable at
+        // the grid resolving stage if it is short-lived, that is, if
+        // it is at the start of the table (or right after headers at
+        // the start of the table).
+        //
+        // TODO(subfooters): explicitly check for short-lived footers.
+        // TODO(subfooters): widow prevention for non-repeated footers with a
+        // similar mechanism / when implementing multiple footers.
+        let footer_would_be_widow = matches!(&self.grid.footer, Some(footer) if footer.repeated)
+            && self.current.lrows.is_empty()
+            && self.current.could_progress_at_top;
 
         let mut laid_out_footer_start = None;
-        if let Some(Repeatable::Repeated(footer)) = &self.grid.footer {
-            // Don't layout the footer if it would be alone with the header in
-            // the page, and don't layout it twice.
-            if !footer_would_be_orphan
-                && self.lrows.iter().all(|row| row.index() < footer.start)
-            {
-                laid_out_footer_start = Some(footer.start);
-                self.layout_footer(footer, engine, self.finished.len())?;
+        if !footer_would_be_widow {
+            if let Some(footer) = &self.grid.footer {
+                // Don't layout the footer if it would be alone with the header
+                // in the page (hence the widow check), and don't layout it
+                // twice (check below).
+                //
+                // TODO(subfooters): this check can be replaced by a vector of
+                // repeating footers in the future, and/or some "pending
+                // footers" vector for footers we're about to place.
+                if footer.repeated
+                    && self.current.lrows.iter().all(|row| row.index() < footer.start)
+                {
+                    laid_out_footer_start = Some(footer.start);
+                    self.layout_footer(footer, engine, self.finished.len())?;
+                }
             }
         }
 
         // Determine the height of existing rows in the region.
         let mut used = Abs::zero();
         let mut fr = Fr::zero();
-        for row in &self.lrows {
+        for row in &self.current.lrows {
             match row {
                 Row::Frame(frame, _, _) => used += frame.height(),
                 Row::Fr(v, _, _) => fr += *v,
@@ -1403,9 +1601,9 @@ impl<'a> GridLayouter<'a> {
 
         // Determine the size of the grid in this region, expanding fully if
         // there are fr rows.
-        let mut size = Size::new(self.width, used).min(self.initial);
-        if fr.get() > 0.0 && self.initial.y.is_finite() {
-            size.y = self.initial.y;
+        let mut size = Size::new(self.width, used).min(self.current.initial);
+        if fr.get() > 0.0 && self.current.initial.y.is_finite() {
+            size.y = self.current.initial.y;
         }
 
         // The frame for the region.
@@ -1413,9 +1611,10 @@ impl<'a> GridLayouter<'a> {
         let mut pos = Point::zero();
         let mut rrows = vec![];
         let current_region = self.finished.len();
+        let mut repeated_header_row_height = Abs::zero();
 
         // Place finished rows and layout fractional rows.
-        for row in std::mem::take(&mut self.lrows) {
+        for (i, row) in std::mem::take(&mut self.current.lrows).into_iter().enumerate() {
             let (frame, y, is_last) = match row {
                 Row::Frame(frame, y, is_last) => (frame, y, is_last),
                 Row::Fr(v, y, disambiguator) => {
@@ -1426,6 +1625,9 @@ impl<'a> GridLayouter<'a> {
             };
 
             let height = frame.height();
+            if i < self.current.repeated_header_rows {
+                repeated_header_row_height += height;
+            }
 
             // Ensure rowspans which span this row will have enough space to
             // be laid out over it later.
@@ -1504,7 +1706,11 @@ impl<'a> GridLayouter<'a> {
                     // we have to check the same index again in the next
                     // iteration.
                     let rowspan = self.rowspans.remove(i);
-                    self.layout_rowspan(rowspan, Some((&mut output, &rrows)), engine)?;
+                    self.layout_rowspan(
+                        rowspan,
+                        Some((&mut output, repeated_header_row_height)),
+                        engine,
+                    )?;
                 } else {
                     i += 1;
                 }
@@ -1515,21 +1721,40 @@ impl<'a> GridLayouter<'a> {
             pos.y += height;
         }
 
-        self.finish_region_internal(output, rrows);
+        self.finish_region_internal(
+            output,
+            rrows,
+            FinishedHeaderRowInfo {
+                repeated_amount: self.current.repeated_header_rows,
+                last_repeated_header_end: self.current.last_repeated_header_end,
+                repeated_height: repeated_header_row_height,
+            },
+        );
 
         if !last {
+            self.current.repeated_header_rows = 0;
+            self.current.last_repeated_header_end = 0;
+            self.current.repeating_header_height = Abs::zero();
+            self.current.repeating_header_heights.clear();
+
             let disambiguator = self.finished.len();
-            if let Some(Repeatable::Repeated(footer)) = &self.grid.footer {
+            if let Some(footer) =
+                self.grid.footer.as_ref().and_then(Repeatable::as_repeated)
+            {
                 self.prepare_footer(footer, engine, disambiguator)?;
             }
 
-            if let Some(Repeatable::Repeated(header)) = &self.grid.header {
-                // Add a header to the new region.
-                self.layout_header(header, engine, disambiguator)?;
-            }
-
             // Ensure rows don't try to overrun the footer.
-            self.regions.size.y -= self.footer_height;
+            // Note that header layout will only subtract this again if it has
+            // to skip regions to fit headers, so there is no risk of
+            // subtracting this twice.
+            self.regions.size.y -= self.current.footer_height;
+            self.current.initial_after_repeats = self.regions.size.y;
+
+            if !self.repeating_headers.is_empty() || !self.pending_headers.is_empty() {
+                // Add headers to the new region.
+                self.layout_active_headers(engine)?;
+            }
         }
 
         Ok(())
@@ -1541,11 +1766,26 @@ impl<'a> GridLayouter<'a> {
         &mut self,
         output: Frame,
         resolved_rows: Vec<RowPiece>,
+        header_row_info: FinishedHeaderRowInfo,
     ) {
         self.finished.push(output);
         self.rrows.push(resolved_rows);
         self.regions.next();
-        self.initial = self.regions.size;
+        self.current.initial = self.regions.size;
+
+        // Repeats haven't been laid out yet, so in the meantime, this will
+        // represent the initial height after repeats laid out so far, and will
+        // be gradually updated when preparing footers and repeating headers.
+        self.current.initial_after_repeats = self.current.initial.y;
+
+        self.current.could_progress_at_top = self.regions.may_progress();
+
+        if !self.grid.headers.is_empty() {
+            self.finished_header_rows.push(header_row_info);
+        }
+
+        // Ensure orphan prevention is handled before resolving rows.
+        debug_assert!(self.current.lrows_orphan_snapshot.is_none());
     }
 }
 
@@ -1560,13 +1800,3 @@ pub(super) fn points(
         offset
     })
 }
-
-/// Checks if the first region of a sequence of regions is the last usable
-/// region, assuming that the last region will always be occupied by some
-/// specific offset height, even after calling `.next()`, due to some
-/// additional logic which adds content automatically on each region turn (in
-/// our case, headers).
-pub(super) fn in_last_with_offset(regions: Regions<'_>, offset: Abs) -> bool {
-    regions.backlog.is_empty()
-        && regions.last.is_none_or(|height| regions.size.y + offset == height)
-}

crates/typst-layout/src/grid/lines.rs

@@ -391,10 +391,12 @@ pub fn vline_stroke_at_row(
 ///
 /// This function assumes columns are sorted by increasing `x`, and rows are
 /// sorted by increasing `y`.
+#[allow(clippy::too_many_arguments)]
 pub fn hline_stroke_at_column(
     grid: &CellGrid,
     rows: &[RowPiece],
     local_top_y: Option<usize>,
+    header_end_above: Option<usize>,
     in_last_region: bool,
     y: usize,
     x: usize,
@@ -499,17 +501,15 @@ pub fn hline_stroke_at_column(
     // Top border stroke and header stroke are generally prioritized, unless
     // they don't have explicit hline overrides and one or more user-provided
     // hlines would appear at the same position, which then are prioritized.
-    let top_stroke_comes_from_header = grid
-        .header
-        .as_ref()
-        .and_then(Repeatable::as_repeated)
-        .zip(local_top_y)
-        .is_some_and(|(header, local_top_y)| {
-            // Ensure the row above us is a repeated header.
-            // FIXME: Make this check more robust when headers at arbitrary
-            // positions are added.
-            local_top_y < header.end && y > header.end
-        });
+    let top_stroke_comes_from_header = header_end_above.zip(local_top_y).is_some_and(
+        |(last_repeated_header_end, local_top_y)| {
+            // Check if the last repeated header row is above this line.
+            //
+            // Note that `y == last_repeated_header_end` is impossible for a
+            // strictly repeated header (not in its original position).
+            local_top_y < last_repeated_header_end && y > last_repeated_header_end
+        },
+    );
 
     // Prioritize the footer's top stroke as well where applicable.
     let bottom_stroke_comes_from_footer = grid
@@ -637,7 +637,7 @@ mod test {
             },
             vec![],
             vec![],
-            None,
+            vec![],
             None,
             entries,
         )
@@ -1175,7 +1175,7 @@ mod test {
             },
             vec![],
             vec![],
-            None,
+            vec![],
             None,
             entries,
         )
@@ -1268,6 +1268,7 @@ mod test {
                         grid,
                         &rows,
                         y.checked_sub(1),
+                        None,
                         true,
                         y,
                         x,
@@ -1461,6 +1462,7 @@ mod test {
                         grid,
                         &rows,
                         y.checked_sub(1),
+                        None,
                         true,
                         y,
                         x,
@@ -1506,6 +1508,7 @@ mod test {
                     grid,
                     &rows,
                     if y == 4 { Some(2) } else { y.checked_sub(1) },
+                    None,
                     true,
                     y,
                     x,

crates/typst-layout/src/grid/repeated.rs

@@ -1,57 +1,446 @@
+use std::ops::Deref;
+
 use typst_library::diag::SourceResult;
 use typst_library::engine::Engine;
 use typst_library::layout::grid::resolve::{Footer, Header, Repeatable};
 use typst_library::layout::{Abs, Axes, Frame, Regions};
 
-use super::layouter::GridLayouter;
+use super::layouter::{GridLayouter, RowState};
 use super::rowspans::UnbreakableRowGroup;
 
-impl GridLayouter<'_> {
-    /// Layouts the header's rows.
-    /// Skips regions as necessary.
-    pub fn layout_header(
+impl<'a> GridLayouter<'a> {
+    /// Checks whether a region break could help a situation where we're out of
+    /// space for the next row. The criteria are:
+    ///
+    /// 1. If we could progress at the top of the region, that indicates the
+    ///    region has a backlog, or (if we're at the first region) a region break
+    ///    is at all possible (`regions.last` is `Some()`), so that's sufficient.
+    ///
+    /// 2. Otherwise, we may progress if another region break is possible
+    ///    (`regions.last` is still `Some()`) and non-repeating rows have been
+    ///    placed, since that means the space they occupy will be available in the
+    ///    next region.
+    #[inline]
+    pub fn may_progress_with_repeats(&self) -> bool {
+        // TODO(subfooters): check below isn't enough to detect non-repeating
+        // footers... we can also change 'initial_after_repeats' to stop being
+        // calculated if there were any non-repeating footers.
+        self.current.could_progress_at_top
+            || self.regions.last.is_some()
+                && self.regions.size.y != self.current.initial_after_repeats
+    }
+
+    pub fn place_new_headers(
+        &mut self,
+        consecutive_header_count: &mut usize,
+        engine: &mut Engine,
+    ) -> SourceResult<()> {
+        *consecutive_header_count += 1;
+        let (consecutive_headers, new_upcoming_headers) =
+            self.upcoming_headers.split_at(*consecutive_header_count);
+
+        if new_upcoming_headers.first().is_some_and(|next_header| {
+            consecutive_headers.last().is_none_or(|latest_header| {
+                !latest_header.short_lived
+                    && next_header.range.start == latest_header.range.end
+            }) && !next_header.short_lived
+        }) {
+            // More headers coming, so wait until we reach them.
+            return Ok(());
+        }
+
+        self.upcoming_headers = new_upcoming_headers;
+        *consecutive_header_count = 0;
+
+        let [first_header, ..] = consecutive_headers else {
+            self.flush_orphans();
+            return Ok(());
+        };
+
+        // Assuming non-conflicting headers sorted by increasing y, this must
+        // be the header with the lowest level (sorted by increasing levels).
+        let first_level = first_header.level;
+
+        // Stop repeating conflicting headers, even if the new headers are
+        // short-lived or won't repeat.
+        //
+        // If we go to a new region before the new headers fit alongside their
+        // children (or in general, for short-lived), the old headers should
+        // not be displayed anymore.
+        let first_conflicting_pos =
+            self.repeating_headers.partition_point(|h| h.level < first_level);
+        self.repeating_headers.truncate(first_conflicting_pos);
+
+        // Ensure upcoming rows won't see that these headers will occupy any
+        // space in future regions anymore.
+        for removed_height in
+            self.current.repeating_header_heights.drain(first_conflicting_pos..)
+        {
+            self.current.repeating_header_height -= removed_height;
+        }
+
+        // Layout short-lived headers immediately.
+        if consecutive_headers.last().is_some_and(|h| h.short_lived) {
+            // No chance of orphans as we're immediately placing conflicting
+            // headers afterwards, which basically are not headers, for all intents
+            // and purposes. It is therefore guaranteed that all new headers have
+            // been placed at least once.
+            self.flush_orphans();
+
+            // Layout each conflicting header independently, without orphan
+            // prevention (as they don't go into 'pending_headers').
+            // These headers are short-lived as they are immediately followed by a
+            // header of the same or lower level, such that they never actually get
+            // to repeat.
+            self.layout_new_headers(consecutive_headers, true, engine)?;
+        } else {
+            // Let's try to place pending headers at least once.
+            // This might be a waste as we could generate an orphan and thus have
+            // to try to place old and new headers all over again, but that happens
+            // for every new region anyway, so it's rather unavoidable.
+            let snapshot_created =
+                self.layout_new_headers(consecutive_headers, false, engine)?;
+
+            // Queue the new headers for layout. They will remain in this
+            // vector due to orphan prevention.
+            //
+            // After the first subsequent row is laid out, move to repeating, as
+            // it's then confirmed the headers won't be moved due to orphan
+            // prevention anymore.
+            self.pending_headers = consecutive_headers;
+
+            if !snapshot_created {
+                // Region probably couldn't progress.
+                //
+                // Mark new pending headers as final and ensure there isn't a
+                // snapshot.
+                self.flush_orphans();
+            }
+        }
+
+        Ok(())
+    }
+
+    /// Lays out rows belonging to a header, returning the calculated header
+    /// height only for that header. Indicates to the laid out rows that they
+    /// should inform their laid out heights if appropriate (auto or fixed
+    /// size rows only).
+    #[inline]
+    fn layout_header_rows(
         &mut self,
         header: &Header,
         engine: &mut Engine,
         disambiguator: usize,
-    ) -> SourceResult<()> {
-        let header_rows =
-            self.simulate_header(header, &self.regions, engine, disambiguator)?;
-        let mut skipped_region = false;
-        while self.unbreakable_rows_left == 0
-            && !self.regions.size.y.fits(header_rows.height + self.footer_height)
-            && self.regions.may_progress()
-        {
-            // Advance regions without any output until we can place the
-            // header and the footer.
-            self.finish_region_internal(Frame::soft(Axes::splat(Abs::zero())), vec![]);
-            skipped_region = true;
+        as_short_lived: bool,
+    ) -> SourceResult<Abs> {
+        let mut header_height = Abs::zero();
+        for y in header.range.clone() {
+            header_height += self
+                .layout_row_with_state(
+                    y,
+                    engine,
+                    disambiguator,
+                    RowState {
+                        current_row_height: Some(Abs::zero()),
+                        in_active_repeatable: !as_short_lived,
+                    },
+                )?
+                .current_row_height
+                .unwrap_or_default();
+        }
+        Ok(header_height)
+    }
+
+    /// This function should be called each time an additional row has been
+    /// laid out in a region to indicate that orphan prevention has succeeded.
+    ///
+    /// It removes the current orphan snapshot and flushes pending headers,
+    /// such that a non-repeating header won't try to be laid out again
+    /// anymore, and a repeating header will begin to be part of
+    /// `repeating_headers`.
+    pub fn flush_orphans(&mut self) {
+        self.current.lrows_orphan_snapshot = None;
+        self.flush_pending_headers();
+    }
+
+    /// Indicates all currently pending headers have been successfully placed
+    /// once, since another row has been placed after them, so they are
+    /// certainly not orphans.
+    pub fn flush_pending_headers(&mut self) {
+        if self.pending_headers.is_empty() {
+            return;
         }
 
-        // Reset the header height for this region.
-        // It will be re-calculated when laying out each header row.
-        self.header_height = Abs::zero();
-
-        if let Some(Repeatable::Repeated(footer)) = &self.grid.footer {
-            if skipped_region {
-                // Simulate the footer again; the region's 'full' might have
-                // changed.
-                self.footer_height = self
-                    .simulate_footer(footer, &self.regions, engine, disambiguator)?
-                    .height;
+        for header in self.pending_headers {
+            if header.repeated {
+                // Vector remains sorted by increasing levels:
+                // - 'pending_headers' themselves are sorted, since we only
+                // push non-mutually-conflicting headers at a time.
+                // - Before pushing new pending headers in
+                // 'layout_new_pending_headers', we truncate repeating headers
+                // to remove anything with the same or higher levels as the
+                // first pending header.
+                // - Assuming it was sorted before, that truncation only keeps
+                // elements with a lower level.
+                // - Therefore, by pushing this header to the end, it will have
+                // a level larger than all the previous headers, and is thus
+                // in its 'correct' position.
+                self.repeating_headers.push(header);
             }
         }
 
-        // Header is unbreakable.
+        self.pending_headers = Default::default();
+    }
+
+    /// Lays out the rows of repeating and pending headers at the top of the
+    /// region.
+    ///
+    /// Assumes the footer height for the current region has already been
+    /// calculated. Skips regions as necessary to fit all headers and all
+    /// footers.
+    pub fn layout_active_headers(&mut self, engine: &mut Engine) -> SourceResult<()> {
+        // Generate different locations for content in headers across its
+        // repetitions by assigning a unique number for each one.
+        let disambiguator = self.finished.len();
+
+        let header_height = self.simulate_header_height(
+            self.repeating_headers
+                .iter()
+                .copied()
+                .chain(self.pending_headers.iter().map(Repeatable::deref)),
+            &self.regions,
+            engine,
+            disambiguator,
+        )?;
+
+        // We already take the footer into account below.
+        // While skipping regions, footer height won't be automatically
+        // re-calculated until the end.
+        let mut skipped_region = false;
+        while self.unbreakable_rows_left == 0
+            && !self.regions.size.y.fits(header_height)
+            && self.may_progress_with_repeats()
+        {
+            // Advance regions without any output until we can place the
+            // header and the footer.
+            self.finish_region_internal(
+                Frame::soft(Axes::splat(Abs::zero())),
+                vec![],
+                Default::default(),
+            );
+
+            // TODO(layout model): re-calculate heights of headers and footers
+            // on each region if 'full' changes? (Assuming height doesn't
+            // change for now...)
+            //
+            // Would remove the footer height update below (move it here).
+            skipped_region = true;
+
+            self.regions.size.y -= self.current.footer_height;
+            self.current.initial_after_repeats = self.regions.size.y;
+        }
+
+        if let Some(footer) = &self.grid.footer {
+            if footer.repeated && skipped_region {
+                // Simulate the footer again; the region's 'full' might have
+                // changed.
+                self.regions.size.y += self.current.footer_height;
+                self.current.footer_height = self
+                    .simulate_footer(footer, &self.regions, engine, disambiguator)?
+                    .height;
+                self.regions.size.y -= self.current.footer_height;
+            }
+        }
+
+        let repeating_header_rows =
+            total_header_row_count(self.repeating_headers.iter().copied());
+
+        let pending_header_rows =
+            total_header_row_count(self.pending_headers.iter().map(Repeatable::deref));
+
+        // Group of headers is unbreakable.
         // Thus, no risk of 'finish_region' being recursively called from
         // within 'layout_row'.
-        self.unbreakable_rows_left += header.end;
-        for y in 0..header.end {
-            self.layout_row(y, engine, disambiguator)?;
+        self.unbreakable_rows_left += repeating_header_rows + pending_header_rows;
+
+        self.current.last_repeated_header_end =
+            self.repeating_headers.last().map(|h| h.range.end).unwrap_or_default();
+
+        // Reset the header height for this region.
+        // It will be re-calculated when laying out each header row.
+        self.current.repeating_header_height = Abs::zero();
+        self.current.repeating_header_heights.clear();
+
+        debug_assert!(self.current.lrows.is_empty());
+        debug_assert!(self.current.lrows_orphan_snapshot.is_none());
+        let may_progress = self.may_progress_with_repeats();
+
+        if may_progress {
+            // Enable orphan prevention for headers at the top of the region.
+            // Otherwise, we will flush pending headers below, after laying
+            // them out.
+            //
+            // It is very rare for this to make a difference as we're usually
+            // at the 'last' region after the first skip, at which the snapshot
+            // is handled by 'layout_new_headers'. Either way, we keep this
+            // here for correctness.
+            self.current.lrows_orphan_snapshot = Some(self.current.lrows.len());
         }
+
+        // Use indices to avoid double borrow. We don't mutate headers in
+        // 'layout_row' so this is fine.
+        let mut i = 0;
+        while let Some(&header) = self.repeating_headers.get(i) {
+            let header_height =
+                self.layout_header_rows(header, engine, disambiguator, false)?;
+            self.current.repeating_header_height += header_height;
+
+            // We assume that this vector will be sorted according
+            // to increasing levels like 'repeating_headers' and
+            // 'pending_headers' - and, in particular, their union, as this
+            // vector is pushed repeating heights from both.
+            //
+            // This is guaranteed by:
+            // 1. We always push pending headers after repeating headers,
+            // as we assume they don't conflict because we remove
+            // conflicting repeating headers when pushing a new pending
+            // header.
+            //
+            // 2. We push in the same order as each.
+            //
+            // 3. This vector is also modified when pushing a new pending
+            // header, where we remove heights for conflicting repeating
+            // headers which have now stopped repeating. They are always at
+            // the end and new pending headers respect the existing sort,
+            // so the vector will remain sorted.
+            self.current.repeating_header_heights.push(header_height);
+
+            i += 1;
+        }
+
+        self.current.repeated_header_rows = self.current.lrows.len();
+        self.current.initial_after_repeats = self.regions.size.y;
+
+        let mut has_non_repeated_pending_header = false;
+        for header in self.pending_headers {
+            if !header.repeated {
+                self.current.initial_after_repeats = self.regions.size.y;
+                has_non_repeated_pending_header = true;
+            }
+            let header_height =
+                self.layout_header_rows(header, engine, disambiguator, false)?;
+            if header.repeated {
+                self.current.repeating_header_height += header_height;
+                self.current.repeating_header_heights.push(header_height);
+            }
+        }
+
+        if !has_non_repeated_pending_header {
+            self.current.initial_after_repeats = self.regions.size.y;
+        }
+
+        if !may_progress {
+            // Flush pending headers immediately, as placing them again later
+            // won't help.
+            self.flush_orphans();
+        }
+
         Ok(())
     }
 
+    /// Lays out headers found for the first time during row layout.
+    ///
+    /// If 'short_lived' is true, these headers are immediately followed by
+    /// a conflicting header, so it is assumed they will not be pushed to
+    /// pending headers.
+    ///
+    /// Returns whether orphan prevention was successfully setup, or couldn't
+    /// due to short-lived headers or the region couldn't progress.
+    pub fn layout_new_headers(
+        &mut self,
+        headers: &'a [Repeatable<Header>],
+        short_lived: bool,
+        engine: &mut Engine,
+    ) -> SourceResult<bool> {
+        // At first, only consider the height of the given headers. However,
+        // for upcoming regions, we will have to consider repeating headers as
+        // well.
+        let header_height = self.simulate_header_height(
+            headers.iter().map(Repeatable::deref),
+            &self.regions,
+            engine,
+            0,
+        )?;
+
+        while self.unbreakable_rows_left == 0
+            && !self.regions.size.y.fits(header_height)
+            && self.may_progress_with_repeats()
+        {
+            // Note that, after the first region skip, the new headers will go
+            // at the top of the region, but after the repeating headers that
+            // remained (which will be automatically placed in 'finish_region').
+            self.finish_region(engine, false)?;
+        }
+
+        // Remove new headers at the end of the region if the upcoming row
+        // doesn't fit.
+        // TODO(subfooters): what if there is a footer right after it?
+        let should_snapshot = !short_lived
+            && self.current.lrows_orphan_snapshot.is_none()
+            && self.may_progress_with_repeats();
+
+        if should_snapshot {
+            // If we don't enter this branch while laying out non-short lived
+            // headers, that means we will have to immediately flush pending
+            // headers and mark them as final, since trying to place them in
+            // the next page won't help get more space.
+            self.current.lrows_orphan_snapshot = Some(self.current.lrows.len());
+        }
+
+        let mut at_top = self.regions.size.y == self.current.initial_after_repeats;
+
+        self.unbreakable_rows_left +=
+            total_header_row_count(headers.iter().map(Repeatable::deref));
+
+        for header in headers {
+            let header_height = self.layout_header_rows(header, engine, 0, false)?;
+
+            // Only store this header height if it is actually going to
+            // become a pending header. Otherwise, pretend it's not a
+            // header... This is fine for consumers of 'header_height' as
+            // it is guaranteed this header won't appear in a future
+            // region, so multi-page rows and cells can effectively ignore
+            // this header.
+            if !short_lived && header.repeated {
+                self.current.repeating_header_height += header_height;
+                self.current.repeating_header_heights.push(header_height);
+                if at_top {
+                    self.current.initial_after_repeats = self.regions.size.y;
+                }
+            } else {
+                at_top = false;
+            }
+        }
+
+        Ok(should_snapshot)
+    }
+
+    /// Calculates the total expected height of several headers.
+    pub fn simulate_header_height<'h: 'a>(
+        &self,
+        headers: impl IntoIterator<Item = &'h Header>,
+        regions: &Regions<'_>,
+        engine: &mut Engine,
+        disambiguator: usize,
+    ) -> SourceResult<Abs> {
+        let mut height = Abs::zero();
+        for header in headers {
+            height +=
+                self.simulate_header(header, regions, engine, disambiguator)?.height;
+        }
+        Ok(height)
+    }
+
     /// Simulate the header's group of rows.
     pub fn simulate_header(
         &self,
@@ -66,8 +455,8 @@ impl GridLayouter<'_> {
         // assume that the amount of unbreakable rows following the first row
         // in the header will be precisely the rows in the header.
         self.simulate_unbreakable_row_group(
-            0,
-            Some(header.end),
+            header.range.start,
+            Some(header.range.end - header.range.start),
             regions,
             engine,
             disambiguator,
@@ -91,11 +480,22 @@ impl GridLayouter<'_> {
         {
             // Advance regions without any output until we can place the
             // footer.
-            self.finish_region_internal(Frame::soft(Axes::splat(Abs::zero())), vec![]);
+            self.finish_region_internal(
+                Frame::soft(Axes::splat(Abs::zero())),
+                vec![],
+                Default::default(),
+            );
             skipped_region = true;
         }
 
-        self.footer_height = if skipped_region {
+        // TODO(subfooters): Consider resetting header height etc. if we skip
+        // region. (Maybe move that step to `finish_region_internal`.)
+        //
+        // That is unnecessary at the moment as 'prepare_footers' is only
+        // called at the start of the region, so header height is always zero
+        // and no headers were placed so far, but what about when we can have
+        // footers in the middle of the region? Let's think about this then.
+        self.current.footer_height = if skipped_region {
             // Simulate the footer again; the region's 'full' might have
             // changed.
             self.simulate_footer(footer, &self.regions, engine, disambiguator)?
@@ -118,12 +518,22 @@ impl GridLayouter<'_> {
         // Ensure footer rows have their own height available.
         // Won't change much as we're creating an unbreakable row group
         // anyway, so this is mostly for correctness.
-        self.regions.size.y += self.footer_height;
+        self.regions.size.y += self.current.footer_height;
 
+        let repeats = self.grid.footer.as_ref().is_some_and(|f| f.repeated);
         let footer_len = self.grid.rows.len() - footer.start;
         self.unbreakable_rows_left += footer_len;
+
         for y in footer.start..self.grid.rows.len() {
-            self.layout_row(y, engine, disambiguator)?;
+            self.layout_row_with_state(
+                y,
+                engine,
+                disambiguator,
+                RowState {
+                    in_active_repeatable: repeats,
+                    ..Default::default()
+                },
+            )?;
         }
 
         Ok(())
@@ -144,10 +554,18 @@ impl GridLayouter<'_> {
         // in the footer will be precisely the rows in the footer.
         self.simulate_unbreakable_row_group(
             footer.start,
-            Some(self.grid.rows.len() - footer.start),
+            Some(footer.end - footer.start),
             regions,
             engine,
             disambiguator,
         )
     }
 }
+
+/// The total amount of rows in the given list of headers.
+#[inline]
+pub fn total_header_row_count<'h>(
+    headers: impl IntoIterator<Item = &'h Header>,
+) -> usize {
+    headers.into_iter().map(|h| h.range.end - h.range.start).sum()
+}

crates/typst-layout/src/grid/rowspans.rs

@@ -4,7 +4,7 @@ use typst_library::foundations::Resolve;
 use typst_library::layout::grid::resolve::Repeatable;
 use typst_library::layout::{Abs, Axes, Frame, Point, Region, Regions, Size, Sizing};
 
-use super::layouter::{in_last_with_offset, points, Row, RowPiece};
+use super::layouter::{points, Row};
 use super::{layout_cell, Cell, GridLayouter};
 
 /// All information needed to layout a single rowspan.
@@ -90,10 +90,10 @@ pub struct CellMeasurementData<'layouter> {
 
 impl GridLayouter<'_> {
     /// Layout a rowspan over the already finished regions, plus the current
-    /// region's frame and resolved rows, if it wasn't finished yet (because
-    /// we're being called from `finish_region`, but note that this function is
-    /// also called once after all regions are finished, in which case
-    /// `current_region_data` is `None`).
+    /// region's frame and height of resolved header rows, if it wasn't
+    /// finished yet (because we're being called from `finish_region`, but note
+    /// that this function is also called once after all regions are finished,
+    /// in which case `current_region_data` is `None`).
     ///
     /// We need to do this only once we already know the heights of all
     /// spanned rows, which is only possible after laying out the last row
@@ -101,7 +101,7 @@ impl GridLayouter<'_> {
     pub fn layout_rowspan(
         &mut self,
         rowspan_data: Rowspan,
-        current_region_data: Option<(&mut Frame, &[RowPiece])>,
+        current_region_data: Option<(&mut Frame, Abs)>,
         engine: &mut Engine,
     ) -> SourceResult<()> {
         let Rowspan {
@@ -146,11 +146,31 @@ impl GridLayouter<'_> {
 
         // Push the layouted frames directly into the finished frames.
         let fragment = layout_cell(cell, engine, disambiguator, self.styles, pod)?;
-        let (current_region, current_rrows) = current_region_data.unzip();
-        for ((i, finished), frame) in self
+        let (current_region, current_header_row_height) = current_region_data.unzip();
+
+        // Clever trick to process finished header rows:
+        // - If there are grid headers, the vector will be filled with one
+        // finished header row height per region, so, chaining with the height
+        // for the current one, we get the header row height for each region.
+        //
+        // - But if there are no grid headers, the vector will be empty, so in
+        // theory the regions and resolved header row heights wouldn't match.
+        // But that's fine - 'current_header_row_height' can only be either
+        // 'Some(zero)' or 'None' in such a case, and for all other rows we
+        // append infinite zeros. That is, in such a case, the resolved header
+        // row height is always zero, so that's our fallback.
+        let finished_header_rows = self
+            .finished_header_rows
+            .iter()
+            .map(|info| info.repeated_height)
+            .chain(current_header_row_height)
+            .chain(std::iter::repeat(Abs::zero()));
+
+        for ((i, (finished, header_dy)), frame) in self
             .finished
             .iter_mut()
             .chain(current_region.into_iter())
+            .zip(finished_header_rows)
             .skip(first_region)
             .enumerate()
             .zip(fragment)
@@ -162,22 +182,9 @@ impl GridLayouter<'_> {
             } else {
                 // The rowspan continuation starts after the header (thus,
                 // at a position after the sum of the laid out header
-                // rows).
-                if let Some(Repeatable::Repeated(header)) = &self.grid.header {
-                    let header_rows = self
-                        .rrows
-                        .get(i)
-                        .map(Vec::as_slice)
-                        .or(current_rrows)
-                        .unwrap_or(&[])
-                        .iter()
-                        .take_while(|row| row.y < header.end);
-
-                    header_rows.map(|row| row.height).sum()
-                } else {
-                    // Without a header, start at the very top of the region.
-                    Abs::zero()
-                }
+                // rows). Without a header, this is zero, so the rowspan can
+                // start at the very top of the region as usual.
+                header_dy
             };
 
             finished.push_frame(Point::new(dx, dy), frame);
@@ -231,15 +238,13 @@ impl GridLayouter<'_> {
             // current row is dynamic and depends on the amount of upcoming
             // unbreakable cells (with or without a rowspan setting).
             let mut amount_unbreakable_rows = None;
-            if let Some(Repeatable::NotRepeated(header)) = &self.grid.header {
-                if current_row < header.end {
-                    // Non-repeated header, so keep it unbreakable.
-                    amount_unbreakable_rows = Some(header.end);
-                }
-            }
-            if let Some(Repeatable::NotRepeated(footer)) = &self.grid.footer {
-                if current_row >= footer.start {
+            if let Some(footer) = &self.grid.footer {
+                if !footer.repeated && current_row >= footer.start {
                     // Non-repeated footer, so keep it unbreakable.
+                    //
+                    // TODO(subfooters): This will become unnecessary
+                    // once non-repeated footers are treated differently and
+                    // have widow prevention.
                     amount_unbreakable_rows = Some(self.grid.rows.len() - footer.start);
                 }
             }
@@ -254,10 +259,7 @@ impl GridLayouter<'_> {
 
             // Skip to fitting region.
             while !self.regions.size.y.fits(row_group.height)
-                && !in_last_with_offset(
-                    self.regions,
-                    self.header_height + self.footer_height,
-                )
+                && self.may_progress_with_repeats()
             {
                 self.finish_region(engine, false)?;
             }
@@ -396,16 +398,29 @@ impl GridLayouter<'_> {
             // auto rows don't depend on the backlog, as they only span one
             // region.
             if breakable
-                && (matches!(self.grid.header, Some(Repeatable::Repeated(_)))
-                    || matches!(self.grid.footer, Some(Repeatable::Repeated(_))))
+                && (!self.repeating_headers.is_empty()
+                    || !self.pending_headers.is_empty()
+                    || matches!(&self.grid.footer, Some(footer) if footer.repeated))
             {
                 // Subtract header and footer height from all upcoming regions
                 // when measuring the cell, including the last repeated region.
                 //
                 // This will update the 'custom_backlog' vector with the
                 // updated heights of the upcoming regions.
+                //
+                // We predict that header height will only include that of
+                // repeating headers, as we can assume non-repeating headers in
+                // the first region have been successfully placed, unless
+                // something didn't fit on the first region of the auto row,
+                // but we will only find that out after measurement, and if
+                // that happens, we discard the measurement and try again.
                 let mapped_regions = self.regions.map(&mut custom_backlog, |size| {
-                    Size::new(size.x, size.y - self.header_height - self.footer_height)
+                    Size::new(
+                        size.x,
+                        size.y
+                            - self.current.repeating_header_height
+                            - self.current.footer_height,
+                    )
                 });
 
                 // Callees must use the custom backlog instead of the current
@@ -459,6 +474,7 @@ impl GridLayouter<'_> {
             // Height of the rowspan covered by spanned rows in the current
             // region.
             let laid_out_height: Abs = self
+                .current
                 .lrows
                 .iter()
                 .filter_map(|row| match row {
@@ -506,7 +522,12 @@ impl GridLayouter<'_> {
                     .iter()
                     .copied()
                     .chain(std::iter::once(if breakable {
-                        self.initial.y - self.header_height - self.footer_height
+                        // Here we are calculating the available height for a
+                        // rowspan from the top of the current region, so
+                        // we have to use initial header heights (note that
+                        // header height can change in the middle of the
+                        // region).
+                        self.current.initial_after_repeats
                     } else {
                         // When measuring unbreakable auto rows, infinite
                         // height is available for content to expand.
@@ -518,11 +539,13 @@ impl GridLayouter<'_> {
                     // rowspan's already laid out heights with the current
                     // region's height and current backlog to ensure a good
                     // level of accuracy in the measurements.
-                    let backlog = self
-                        .regions
-                        .backlog
-                        .iter()
-                        .map(|&size| size - self.header_height - self.footer_height);
+                    //
+                    // Assume only repeating headers will survive starting at
+                    // the next region.
+                    let backlog = self.regions.backlog.iter().map(|&size| {
+                        size - self.current.repeating_header_height
+                            - self.current.footer_height
+                    });
 
                     heights_up_to_current_region.chain(backlog).collect::<Vec<_>>()
                 } else {
@@ -536,10 +559,10 @@ impl GridLayouter<'_> {
                 height = *rowspan_height;
                 backlog = None;
                 full = rowspan_full;
-                last = self
-                    .regions
-                    .last
-                    .map(|size| size - self.header_height - self.footer_height);
+                last = self.regions.last.map(|size| {
+                    size - self.current.repeating_header_height
+                        - self.current.footer_height
+                });
             } else {
                 // The rowspan started in the current region, as its vector
                 // of heights in regions is currently empty.
@@ -741,10 +764,11 @@ impl GridLayouter<'_> {
             simulated_regions.next();
             disambiguator += 1;
 
-            // Subtract the initial header and footer height, since that's the
-            // height we used when subtracting from the region backlog's
+            // Subtract the repeating header and footer height, since that's
+            // the height we used when subtracting from the region backlog's
             // heights while measuring cells.
-            simulated_regions.size.y -= self.header_height + self.footer_height;
+            simulated_regions.size.y -=
+                self.current.repeating_header_height + self.current.footer_height;
         }
 
         if let Some(original_last_resolved_size) = last_resolved_size {
@@ -876,12 +900,8 @@ impl GridLayouter<'_> {
         // which, when used and combined with upcoming spanned rows, covers all
         // of the requested rowspan height, we give up.
         for _attempt in 0..5 {
-            let rowspan_simulator = RowspanSimulator::new(
-                disambiguator,
-                simulated_regions,
-                self.header_height,
-                self.footer_height,
-            );
+            let rowspan_simulator =
+                RowspanSimulator::new(disambiguator, simulated_regions, &self.current);
 
             let total_spanned_height = rowspan_simulator.simulate_rowspan_layout(
                 y,
@@ -963,7 +983,8 @@ impl GridLayouter<'_> {
             {
                 extra_amount_to_grow -= simulated_regions.size.y.max(Abs::zero());
                 simulated_regions.next();
-                simulated_regions.size.y -= self.header_height + self.footer_height;
+                simulated_regions.size.y -=
+                    self.current.repeating_header_height + self.current.footer_height;
                 disambiguator += 1;
             }
             simulated_regions.size.y -= extra_amount_to_grow;
@@ -980,10 +1001,17 @@ struct RowspanSimulator<'a> {
     finished: usize,
     /// The state of regions during the simulation.
     regions: Regions<'a>,
-    /// The height of the header in the currently simulated region.
+    /// The total height of headers in the currently simulated region.
     header_height: Abs,
-    /// The height of the footer in the currently simulated region.
+    /// The total height of footers in the currently simulated region.
     footer_height: Abs,
+    /// Whether `self.regions.may_progress()` was `true` at the top of the
+    /// region, indicating we can progress anywhere in the current region,
+    /// even right after a repeated header.
+    could_progress_at_top: bool,
+    /// Available height after laying out repeated headers at the top of the
+    /// currently simulated region.
+    initial_after_repeats: Abs,
     /// The total spanned height so far in the simulation.
     total_spanned_height: Abs,
     /// Height of the latest spanned gutter row in the simulation.
@@ -997,14 +1025,19 @@ impl<'a> RowspanSimulator<'a> {
     fn new(
         finished: usize,
         regions: Regions<'a>,
-        header_height: Abs,
-        footer_height: Abs,
+        current: &super::layouter::Current,
     ) -> Self {
         Self {
             finished,
             regions,
-            header_height,
-            footer_height,
+            // There can be no new headers or footers within a multi-page
+            // rowspan, since headers and footers are unbreakable, so
+            // assuming the repeating header height and footer height
+            // won't change is safe.
+            header_height: current.repeating_header_height,
+            footer_height: current.footer_height,
+            could_progress_at_top: current.could_progress_at_top,
+            initial_after_repeats: current.initial_after_repeats,
             total_spanned_height: Abs::zero(),
             latest_spanned_gutter_height: Abs::zero(),
         }
@@ -1053,10 +1086,7 @@ impl<'a> RowspanSimulator<'a> {
                     0,
                 )?;
                 while !self.regions.size.y.fits(row_group.height)
-                    && !in_last_with_offset(
-                        self.regions,
-                        self.header_height + self.footer_height,
-                    )
+                    && self.may_progress_with_repeats()
                 {
                     self.finish_region(layouter, engine)?;
                 }
@@ -1078,10 +1108,7 @@ impl<'a> RowspanSimulator<'a> {
                     let mut skipped_region = false;
                     while unbreakable_rows_left == 0
                         && !self.regions.size.y.fits(height)
-                        && !in_last_with_offset(
-                            self.regions,
-                            self.header_height + self.footer_height,
-                        )
+                        && self.may_progress_with_repeats()
                     {
                         self.finish_region(layouter, engine)?;
 
@@ -1127,23 +1154,37 @@ impl<'a> RowspanSimulator<'a> {
         // our simulation checks what happens AFTER the auto row, so we can
         // just use the original backlog from `self.regions`.
         let disambiguator = self.finished;
-        let header_height =
-            if let Some(Repeatable::Repeated(header)) = &layouter.grid.header {
-                layouter
-                    .simulate_header(header, &self.regions, engine, disambiguator)?
-                    .height
-            } else {
-                Abs::zero()
-            };
 
-        let footer_height =
-            if let Some(Repeatable::Repeated(footer)) = &layouter.grid.footer {
-                layouter
-                    .simulate_footer(footer, &self.regions, engine, disambiguator)?
-                    .height
-            } else {
-                Abs::zero()
-            };
+        let (repeating_headers, header_height) = if !layouter.repeating_headers.is_empty()
+            || !layouter.pending_headers.is_empty()
+        {
+            // Only repeating headers have survived after the first region
+            // break.
+            let repeating_headers = layouter.repeating_headers.iter().copied().chain(
+                layouter.pending_headers.iter().filter_map(Repeatable::as_repeated),
+            );
+
+            let header_height = layouter.simulate_header_height(
+                repeating_headers.clone(),
+                &self.regions,
+                engine,
+                disambiguator,
+            )?;
+
+            (Some(repeating_headers), header_height)
+        } else {
+            (None, Abs::zero())
+        };
+
+        let footer_height = if let Some(footer) =
+            layouter.grid.footer.as_ref().and_then(Repeatable::as_repeated)
+        {
+            layouter
+                .simulate_footer(footer, &self.regions, engine, disambiguator)?
+                .height
+        } else {
+            Abs::zero()
+        };
 
         let mut skipped_region = false;
 
@@ -1156,19 +1197,24 @@ impl<'a> RowspanSimulator<'a> {
             skipped_region = true;
         }
 
-        if let Some(Repeatable::Repeated(header)) = &layouter.grid.header {
+        if let Some(repeating_headers) = repeating_headers {
             self.header_height = if skipped_region {
                 // Simulate headers again, at the new region, as
                 // the full region height may change.
-                layouter
-                    .simulate_header(header, &self.regions, engine, disambiguator)?
-                    .height
+                layouter.simulate_header_height(
+                    repeating_headers,
+                    &self.regions,
+                    engine,
+                    disambiguator,
+                )?
             } else {
                 header_height
             };
         }
 
-        if let Some(Repeatable::Repeated(footer)) = &layouter.grid.footer {
+        if let Some(footer) =
+            layouter.grid.footer.as_ref().and_then(Repeatable::as_repeated)
+        {
             self.footer_height = if skipped_region {
                 // Simulate footers again, at the new region, as
                 // the full region height may change.
@@ -1185,6 +1231,7 @@ impl<'a> RowspanSimulator<'a> {
         // header or footer (as an invariant, any rowspans spanning any header
         // or footer rows are fully contained within that header's or footer's rows).
         self.regions.size.y -= self.header_height + self.footer_height;
+        self.initial_after_repeats = self.regions.size.y;
 
         Ok(())
     }
@@ -1201,8 +1248,18 @@ impl<'a> RowspanSimulator<'a> {
         self.regions.next();
         self.finished += 1;
 
+        self.could_progress_at_top = self.regions.may_progress();
         self.simulate_header_footer_layout(layouter, engine)
     }
+
+    /// Similar to [`GridLayouter::may_progress_with_repeats`] but for rowspan
+    /// simulation.
+    #[inline]
+    fn may_progress_with_repeats(&self) -> bool {
+        self.could_progress_at_top
+            || self.regions.last.is_some()
+                && self.regions.size.y != self.initial_after_repeats
+    }
 }
 
 /// Subtracts some size from the end of a vector of sizes.

crates/typst-layout/src/image.rs

@@ -1,6 +1,6 @@
 use std::ffi::OsStr;
 
-use typst_library::diag::{warning, At, SourceResult, StrResult};
+use typst_library::diag::{warning, At, LoadedWithin, SourceResult, StrResult};
 use typst_library::engine::Engine;
 use typst_library::foundations::{Bytes, Derived, Packed, Smart, StyleChain};
 use typst_library::introspection::Locator;
@@ -27,17 +27,17 @@ pub fn layout_image(
 
     // Take the format that was explicitly defined, or parse the extension,
     // or try to detect the format.
-    let Derived { source, derived: data } = &elem.source;
+    let Derived { source, derived: loaded } = &elem.source;
     let format = match elem.format(styles) {
         Smart::Custom(v) => v,
-        Smart::Auto => determine_format(source, data).at(span)?,
+        Smart::Auto => determine_format(source, &loaded.data).at(span)?,
     };
 
     // Warn the user if the image contains a foreign object. Not perfect
     // because the svg could also be encoded, but that's an edge case.
     if format == ImageFormat::Vector(VectorFormat::Svg) {
         let has_foreign_object =
-            data.as_str().is_ok_and(|s| s.contains("<foreignObject"));
+            memchr::memmem::find(&loaded.data, b"<foreignObject").is_some();
 
         if has_foreign_object {
             engine.sink.warn(warning!(
@@ -53,7 +53,7 @@ pub fn layout_image(
     let kind = match format {
         ImageFormat::Raster(format) => ImageKind::Raster(
             RasterImage::new(
-                data.clone(),
+                loaded.data.clone(),
                 format,
                 elem.icc(styles).as_ref().map(|icc| icc.derived.clone()),
             )
@@ -61,11 +61,11 @@ pub fn layout_image(
         ),
         ImageFormat::Vector(VectorFormat::Svg) => ImageKind::Svg(
             SvgImage::with_fonts(
-                data.clone(),
+                loaded.data.clone(),
                 engine.world,
                 &families(styles).map(|f| f.as_str()).collect::<Vec<_>>(),
             )
-            .at(span)?,
+            .within(loaded)?,
         ),
     };
 

crates/typst-layout/src/inline/mod.rs

@@ -9,6 +9,7 @@ mod prepare;
 mod shaping;
 
 pub use self::box_::layout_box;
+pub use self::shaping::create_shape_plan;
 
 use comemo::{Track, Tracked, TrackedMut};
 use typst_library::diag::SourceResult;

crates/typst-layout/src/inline/shaping.rs

@@ -1,18 +1,16 @@
 use std::borrow::Cow;
 use std::fmt::{self, Debug, Formatter};
-use std::str::FromStr;
 use std::sync::Arc;
 
 use az::SaturatingAs;
-use ecow::EcoString;
 use rustybuzz::{BufferFlags, ShapePlan, UnicodeBuffer};
 use ttf_parser::Tag;
 use typst_library::engine::Engine;
 use typst_library::foundations::{Smart, StyleChain};
 use typst_library::layout::{Abs, Dir, Em, Frame, FrameItem, Point, Size};
 use typst_library::text::{
-    families, features, is_default_ignorable, variant, Font, FontFamily, FontVariant,
-    Glyph, Lang, Region, TextEdgeBounds, TextElem, TextItem,
+    families, features, is_default_ignorable, language, variant, Font, FontFamily,
+    FontVariant, Glyph, Lang, Region, TextEdgeBounds, TextElem, TextItem,
 };
 use typst_library::World;
 use typst_utils::SliceExt;
@@ -295,6 +293,8 @@ impl<'a> ShapedText<'a> {
                             + justification_left
                             + justification_right,
                         x_offset: shaped.x_offset + justification_left,
+                        y_advance: Em::zero(),
+                        y_offset: Em::zero(),
                         range: (shaped.range.start - range.start).saturating_as()
                             ..(shaped.range.end - range.start).saturating_as(),
                         span,
@@ -934,7 +934,7 @@ fn shape_segment<'a>(
 
 /// Create a shape plan.
 #[comemo::memoize]
-fn create_shape_plan(
+pub fn create_shape_plan(
     font: &Font,
     direction: rustybuzz::Direction,
     script: rustybuzz::Script,
@@ -952,7 +952,7 @@ fn create_shape_plan(
 
 /// Shape the text with tofus from the given font.
 fn shape_tofus(ctx: &mut ShapingContext, base: usize, text: &str, font: Font) {
-    let x_advance = font.advance(0).unwrap_or_default();
+    let x_advance = font.x_advance(0).unwrap_or_default();
     let add_glyph = |(cluster, c): (usize, char)| {
         let start = base + cluster;
         let end = start + c.len_utf8();
@@ -1044,20 +1044,8 @@ fn calculate_adjustability(ctx: &mut ShapingContext, lang: Lang, region: Option<
 
 /// Difference between non-breaking and normal space.
 fn nbsp_delta(font: &Font) -> Option<Em> {
-    let space = font.ttf().glyph_index(' ')?.0;
     let nbsp = font.ttf().glyph_index('\u{00A0}')?.0;
-    Some(font.advance(nbsp)? - font.advance(space)?)
-}
-
-/// Process the language and region of a style chain into a
-/// rustybuzz-compatible BCP 47 language.
-fn language(styles: StyleChain) -> rustybuzz::Language {
-    let mut bcp: EcoString = TextElem::lang_in(styles).as_str().into();
-    if let Some(region) = TextElem::region_in(styles) {
-        bcp.push('-');
-        bcp.push_str(region.as_str());
-    }
-    rustybuzz::Language::from_str(&bcp).unwrap()
+    Some(font.x_advance(nbsp)? - font.space_width()?)
 }
 
 /// Returns true if all glyphs in `glyphs` have ranges within the range `range`.

crates/typst-layout/src/math/accent.rs

@@ -3,7 +3,10 @@ use typst_library::foundations::{Packed, StyleChain};
 use typst_library::layout::{Em, Frame, Point, Size};
 use typst_library::math::AccentElem;
 
-use super::{style_cramped, FrameFragment, GlyphFragment, MathContext, MathFragment};
+use super::{
+    style_cramped, style_dtls, style_flac, FrameFragment, GlyphFragment, MathContext,
+    MathFragment,
+};
 
 /// How much the accent can be shorter than the base.
 const ACCENT_SHORT_FALL: Em = Em::new(0.5);
@@ -15,40 +18,40 @@ pub fn layout_accent(
     ctx: &mut MathContext,
     styles: StyleChain,
 ) -> SourceResult<()> {
-    let cramped = style_cramped();
-    let mut base = ctx.layout_into_fragment(&elem.base, styles.chain(&cramped))?;
-
     let accent = elem.accent;
     let top_accent = !accent.is_bottom();
 
-    // Try to replace base glyph with its dotless variant.
-    if top_accent && elem.dotless(styles) {
-        if let MathFragment::Glyph(glyph) = &mut base {
-            glyph.make_dotless_form(ctx);
-        }
-    }
+    // Try to replace the base glyph with its dotless variant.
+    let dtls = style_dtls();
+    let base_styles =
+        if top_accent && elem.dotless(styles) { styles.chain(&dtls) } else { styles };
+
+    let cramped = style_cramped();
+    let base = ctx.layout_into_fragment(&elem.base, base_styles.chain(&cramped))?;
 
     // Preserve class to preserve automatic spacing.
     let base_class = base.class();
     let base_attach = base.accent_attach();
 
-    let mut glyph = GlyphFragment::new(ctx, styles, accent.0, elem.span());
+    // Try to replace the accent glyph with its flattened variant.
+    let flattened_base_height = scaled!(ctx, styles, flattened_accent_base_height);
+    let flac = style_flac();
+    let accent_styles = if top_accent && base.ascent() > flattened_base_height {
+        styles.chain(&flac)
+    } else {
+        styles
+    };
 
-    // Try to replace accent glyph with its flattened variant.
-    if top_accent {
-        let flattened_base_height = scaled!(ctx, styles, flattened_accent_base_height);
-        if base.ascent() > flattened_base_height {
-            glyph.make_flattened_accent_form(ctx);
-        }
-    }
+    let mut glyph =
+        GlyphFragment::new_char(ctx.font, accent_styles, accent.0, elem.span())?;
 
-    // Forcing the accent to be at least as large as the base makes it too
-    // wide in many case.
+    // Forcing the accent to be at least as large as the base makes it too wide
+    // in many cases.
     let width = elem.size(styles).relative_to(base.width());
-    let short_fall = ACCENT_SHORT_FALL.at(glyph.font_size);
-    let variant = glyph.stretch_horizontal(ctx, width - short_fall);
-    let accent = variant.frame;
-    let accent_attach = variant.accent_attach.0;
+    let short_fall = ACCENT_SHORT_FALL.at(glyph.item.size);
+    glyph.stretch_horizontal(ctx, width - short_fall);
+    let accent_attach = glyph.accent_attach.0;
+    let accent = glyph.into_frame();
 
     let (gap, accent_pos, base_pos) = if top_accent {
         // Descent is negative because the accent's ink bottom is above the

crates/typst-layout/src/math/attach.rs

@@ -66,7 +66,6 @@ pub fn layout_attach(
         let relative_to_width = measure!(t, width).max(measure!(b, width));
         stretch_fragment(
             ctx,
-            styles,
             &mut base,
             Some(Axis::X),
             Some(relative_to_width),
@@ -220,7 +219,6 @@ fn layout_attachments(
     // Calculate the distance each pre-script extends to the left of the base's
     // width.
     let (tl_pre_width, bl_pre_width) = compute_pre_script_widths(
-        ctx,
         &base,
         [tl.as_ref(), bl.as_ref()],
         (tx_shift, bx_shift),
@@ -231,7 +229,6 @@ fn layout_attachments(
     // base's width. Also calculate each post-script's kerning (we need this for
     // its position later).
     let ((tr_post_width, tr_kern), (br_post_width, br_kern)) = compute_post_script_widths(
-        ctx,
         &base,
         [tr.as_ref(), br.as_ref()],
         (tx_shift, bx_shift),
@@ -287,14 +284,13 @@ fn layout_attachments(
 /// post-script's kerning value. The first tuple is for the post-superscript,
 /// and the second is for the post-subscript.
 fn compute_post_script_widths(
-    ctx: &MathContext,
     base: &MathFragment,
     [tr, br]: [Option<&MathFragment>; 2],
     (tr_shift, br_shift): (Abs, Abs),
     space_after_post_script: Abs,
 ) -> ((Abs, Abs), (Abs, Abs)) {
     let tr_values = tr.map_or_default(|tr| {
-        let kern = math_kern(ctx, base, tr, tr_shift, Corner::TopRight);
+        let kern = math_kern(base, tr, tr_shift, Corner::TopRight);
         (space_after_post_script + tr.width() + kern, kern)
     });
 
@@ -302,7 +298,7 @@ fn compute_post_script_widths(
     // need to shift the post-subscript left by the base's italic correction
     // (see the kerning algorithm as described in the OpenType MATH spec).
     let br_values = br.map_or_default(|br| {
-        let kern = math_kern(ctx, base, br, br_shift, Corner::BottomRight)
+        let kern = math_kern(base, br, br_shift, Corner::BottomRight)
             - base.italics_correction();
         (space_after_post_script + br.width() + kern, kern)
     });
@@ -317,19 +313,18 @@ fn compute_post_script_widths(
 /// extends left of the base's width and the second being the distance the
 /// pre-subscript extends left of the base's width.
 fn compute_pre_script_widths(
-    ctx: &MathContext,
     base: &MathFragment,
     [tl, bl]: [Option<&MathFragment>; 2],
     (tl_shift, bl_shift): (Abs, Abs),
     space_before_pre_script: Abs,
 ) -> (Abs, Abs) {
     let tl_pre_width = tl.map_or_default(|tl| {
-        let kern = math_kern(ctx, base, tl, tl_shift, Corner::TopLeft);
+        let kern = math_kern(base, tl, tl_shift, Corner::TopLeft);
         space_before_pre_script + tl.width() + kern
     });
 
     let bl_pre_width = bl.map_or_default(|bl| {
-        let kern = math_kern(ctx, base, bl, bl_shift, Corner::BottomLeft);
+        let kern = math_kern(base, bl, bl_shift, Corner::BottomLeft);
         space_before_pre_script + bl.width() + kern
     });
 
@@ -471,13 +466,7 @@ fn compute_script_shifts(
 /// a negative value means shifting the script closer to the base. Requires the
 /// distance from the base's baseline to the script's baseline, as well as the
 /// script's corner (tl, tr, bl, br).
-fn math_kern(
-    ctx: &MathContext,
-    base: &MathFragment,
-    script: &MathFragment,
-    shift: Abs,
-    pos: Corner,
-) -> Abs {
+fn math_kern(base: &MathFragment, script: &MathFragment, shift: Abs, pos: Corner) -> Abs {
     // This process is described under the MathKernInfo table in the OpenType
     // MATH spec.
 
@@ -502,8 +491,8 @@ fn math_kern(
 
     // Calculate the sum of kerning values for each correction height.
     let summed_kern = |height| {
-        let base_kern = base.kern_at_height(ctx, pos, height);
-        let attach_kern = script.kern_at_height(ctx, pos.inv(), height);
+        let base_kern = base.kern_at_height(pos, height);
+        let attach_kern = script.kern_at_height(pos.inv(), height);
         base_kern + attach_kern
     };
 

crates/typst-layout/src/math/frac.rs

@@ -109,14 +109,14 @@ fn layout_frac_like(
     frame.push_frame(denom_pos, denom);
 
     if binom {
-        let mut left = GlyphFragment::new(ctx, styles, '(', span)
-            .stretch_vertical(ctx, height - short_fall);
-        left.center_on_axis(ctx);
+        let mut left = GlyphFragment::new_char(ctx.font, styles, '(', span)?;
+        left.stretch_vertical(ctx, height - short_fall);
+        left.center_on_axis();
         ctx.push(left);
         ctx.push(FrameFragment::new(styles, frame));
-        let mut right = GlyphFragment::new(ctx, styles, ')', span)
-            .stretch_vertical(ctx, height - short_fall);
-        right.center_on_axis(ctx);
+        let mut right = GlyphFragment::new_char(ctx.font, styles, ')', span)?;
+        right.stretch_vertical(ctx, height - short_fall);
+        right.center_on_axis();
         ctx.push(right);
     } else {
         frame.push(

crates/typst-layout/src/math/lr.rs

@@ -45,20 +45,20 @@ pub fn layout_lr(
 
     // Scale up fragments at both ends.
     match inner_fragments {
-        [one] => scale(ctx, styles, one, relative_to, height, None),
+        [one] => scale_if_delimiter(ctx, one, relative_to, height, None),
         [first, .., last] => {
-            scale(ctx, styles, first, relative_to, height, Some(MathClass::Opening));
-            scale(ctx, styles, last, relative_to, height, Some(MathClass::Closing));
+            scale_if_delimiter(ctx, first, relative_to, height, Some(MathClass::Opening));
+            scale_if_delimiter(ctx, last, relative_to, height, Some(MathClass::Closing));
         }
-        _ => {}
+        [] => {}
     }
 
-    // Handle MathFragment::Variant fragments that should be scaled up.
+    // Handle MathFragment::Glyph fragments that should be scaled up.
     for fragment in inner_fragments.iter_mut() {
-        if let MathFragment::Variant(ref mut variant) = fragment {
-            if variant.mid_stretched == Some(false) {
-                variant.mid_stretched = Some(true);
-                scale(ctx, styles, fragment, relative_to, height, Some(MathClass::Large));
+        if let MathFragment::Glyph(ref mut glyph) = fragment {
+            if glyph.mid_stretched == Some(false) {
+                glyph.mid_stretched = Some(true);
+                scale(ctx, fragment, relative_to, height);
             }
         }
     }
@@ -95,18 +95,9 @@ pub fn layout_mid(
     let mut fragments = ctx.layout_into_fragments(&elem.body, styles)?;
 
     for fragment in &mut fragments {
-        match fragment {
-            MathFragment::Glyph(glyph) => {
-                let mut new = glyph.clone().into_variant();
-                new.mid_stretched = Some(false);
-                new.class = MathClass::Fence;
-                *fragment = MathFragment::Variant(new);
-            }
-            MathFragment::Variant(variant) => {
-                variant.mid_stretched = Some(false);
-                variant.class = MathClass::Fence;
-            }
-            _ => {}
+        if let MathFragment::Glyph(ref mut glyph) = fragment {
+            glyph.mid_stretched = Some(false);
+            glyph.class = MathClass::Relation;
         }
     }
 
@@ -114,10 +105,13 @@ pub fn layout_mid(
     Ok(())
 }
 
-/// Scale a math fragment to a height.
-fn scale(
+/// Scales a math fragment to a height if it has the class Opening, Closing, or
+/// Fence.
+///
+/// In case `apply` is `Some(class)`, `class` will be applied to the fragment if
+/// it is a delimiter, in a way that cannot be overridden by the user.
+fn scale_if_delimiter(
     ctx: &mut MathContext,
-    styles: StyleChain,
     fragment: &mut MathFragment,
     relative_to: Abs,
     height: Rel<Abs>,
@@ -127,21 +121,23 @@ fn scale(
         fragment.class(),
         MathClass::Opening | MathClass::Closing | MathClass::Fence
     ) {
-        // This unwrap doesn't really matter. If it is None, then the fragment
-        // won't be stretchable anyways.
-        let short_fall = DELIM_SHORT_FALL.at(fragment.font_size().unwrap_or_default());
-        stretch_fragment(
-            ctx,
-            styles,
-            fragment,
-            Some(Axis::Y),
-            Some(relative_to),
-            height,
-            short_fall,
-        );
+        scale(ctx, fragment, relative_to, height);
 
         if let Some(class) = apply {
             fragment.set_class(class);
         }
     }
 }
+
+/// Scales a math fragment to a height.
+fn scale(
+    ctx: &mut MathContext,
+    fragment: &mut MathFragment,
+    relative_to: Abs,
+    height: Rel<Abs>,
+) {
+    // This unwrap doesn't really matter. If it is None, then the fragment
+    // won't be stretchable anyways.
+    let short_fall = DELIM_SHORT_FALL.at(fragment.font_size().unwrap_or_default());
+    stretch_fragment(ctx, fragment, Some(Axis::Y), Some(relative_to), height, short_fall);
+}

crates/typst-layout/src/math/mat.rs

@@ -9,8 +9,8 @@ use typst_library::visualize::{FillRule, FixedStroke, Geometry, LineCap, Shape};
 use typst_syntax::Span;
 
 use super::{
-    alignments, delimiter_alignment, style_for_denominator, AlignmentResult,
-    FrameFragment, GlyphFragment, LeftRightAlternator, MathContext, DELIM_SHORT_FALL,
+    alignments, style_for_denominator, AlignmentResult, FrameFragment, GlyphFragment,
+    LeftRightAlternator, MathContext, DELIM_SHORT_FALL,
 };
 
 const VERTICAL_PADDING: Ratio = Ratio::new(0.1);
@@ -183,8 +183,12 @@ fn layout_body(
     // We pad ascent and descent with the ascent and descent of the paren
     // to ensure that normal matrices are aligned with others unless they are
     // way too big.
-    let paren =
-        GlyphFragment::new(ctx, styles.chain(&denom_style), '(', Span::detached());
+    let paren = GlyphFragment::new_char(
+        ctx.font,
+        styles.chain(&denom_style),
+        '(',
+        Span::detached(),
+    )?;
 
     for (column, col) in columns.iter().zip(&mut cols) {
         for (cell, (ascent, descent)) in column.iter().zip(&mut heights) {
@@ -202,8 +206,8 @@ fn layout_body(
                 ));
             }
 
-            ascent.set_max(cell.ascent().max(paren.ascent));
-            descent.set_max(cell.descent().max(paren.descent));
+            ascent.set_max(cell.ascent().max(paren.ascent()));
+            descent.set_max(cell.descent().max(paren.descent()));
 
             col.push(cell);
         }
@@ -312,19 +316,19 @@ fn layout_delimiters(
     let target = height + VERTICAL_PADDING.of(height);
     frame.set_baseline(height / 2.0 + axis);
 
-    if let Some(left) = left {
-        let mut left = GlyphFragment::new(ctx, styles, left, span)
-            .stretch_vertical(ctx, target - short_fall);
-        left.align_on_axis(ctx, delimiter_alignment(left.c));
+    if let Some(left_c) = left {
+        let mut left = GlyphFragment::new_char(ctx.font, styles, left_c, span)?;
+        left.stretch_vertical(ctx, target - short_fall);
+        left.center_on_axis();
         ctx.push(left);
     }
 
     ctx.push(FrameFragment::new(styles, frame));
 
-    if let Some(right) = right {
-        let mut right = GlyphFragment::new(ctx, styles, right, span)
-            .stretch_vertical(ctx, target - short_fall);
-        right.align_on_axis(ctx, delimiter_alignment(right.c));
+    if let Some(right_c) = right {
+        let mut right = GlyphFragment::new_char(ctx.font, styles, right_c, span)?;
+        right.stretch_vertical(ctx, target - short_fall);
+        right.center_on_axis();
         ctx.push(right);
     }
 

crates/typst-layout/src/math/mod.rs

@@ -13,8 +13,6 @@ mod stretch;
 mod text;
 mod underover;
 
-use rustybuzz::Feature;
-use ttf_parser::Tag;
 use typst_library::diag::{bail, SourceResult};
 use typst_library::engine::Engine;
 use typst_library::foundations::{
@@ -30,7 +28,7 @@ use typst_library::math::*;
 use typst_library::model::ParElem;
 use typst_library::routines::{Arenas, RealizationKind};
 use typst_library::text::{
-    families, features, variant, Font, LinebreakElem, SpaceElem, TextEdgeBounds, TextElem,
+    families, variant, Font, LinebreakElem, SpaceElem, TextEdgeBounds, TextElem,
 };
 use typst_library::World;
 use typst_syntax::Span;
@@ -38,11 +36,11 @@ use typst_utils::Numeric;
 use unicode_math_class::MathClass;
 
 use self::fragment::{
-    FrameFragment, GlyphFragment, GlyphwiseSubsts, Limits, MathFragment, VariantFragment,
+    has_dtls_feat, stretch_axes, FrameFragment, GlyphFragment, Limits, MathFragment,
 };
 use self::run::{LeftRightAlternator, MathRun, MathRunFrameBuilder};
 use self::shared::*;
-use self::stretch::{stretch_fragment, stretch_glyph};
+use self::stretch::stretch_fragment;
 
 /// Layout an inline equation (in a paragraph).
 #[typst_macros::time(span = elem.span())]
@@ -58,7 +56,7 @@ pub fn layout_equation_inline(
     let font = find_math_font(engine, styles, elem.span())?;
 
     let mut locator = locator.split();
-    let mut ctx = MathContext::new(engine, &mut locator, styles, region, &font);
+    let mut ctx = MathContext::new(engine, &mut locator, region, &font);
 
     let scale_style = style_for_script_scale(&ctx);
     let styles = styles.chain(&scale_style);
@@ -113,7 +111,7 @@ pub fn layout_equation_block(
     let font = find_math_font(engine, styles, span)?;
 
     let mut locator = locator.split();
-    let mut ctx = MathContext::new(engine, &mut locator, styles, regions.base(), &font);
+    let mut ctx = MathContext::new(engine, &mut locator, regions.base(), &font);
 
     let scale_style = style_for_script_scale(&ctx);
     let styles = styles.chain(&scale_style);
@@ -374,14 +372,7 @@ struct MathContext<'a, 'v, 'e> {
     region: Region,
     // Font-related.
     font: &'a Font,
-    ttf: &'a ttf_parser::Face<'a>,
-    table: ttf_parser::math::Table<'a>,
     constants: ttf_parser::math::Constants<'a>,
-    dtls_table: Option<GlyphwiseSubsts<'a>>,
-    flac_table: Option<GlyphwiseSubsts<'a>>,
-    ssty_table: Option<GlyphwiseSubsts<'a>>,
-    glyphwise_tables: Option<Vec<GlyphwiseSubsts<'a>>>,
-    space_width: Em,
     // Mutable.
     fragments: Vec<MathFragment>,
 }
@@ -391,46 +382,20 @@ impl<'a, 'v, 'e> MathContext<'a, 'v, 'e> {
     fn new(
         engine: &'v mut Engine<'e>,
         locator: &'v mut SplitLocator<'a>,
-        styles: StyleChain<'a>,
         base: Size,
         font: &'a Font,
     ) -> Self {
-        let math_table = font.ttf().tables().math.unwrap();
-        let gsub_table = font.ttf().tables().gsub;
-        let constants = math_table.constants.unwrap();
-
-        let feat = |tag: &[u8; 4]| {
-            GlyphwiseSubsts::new(gsub_table, Feature::new(Tag::from_bytes(tag), 0, ..))
-        };
-
-        let features = features(styles);
-        let glyphwise_tables = Some(
-            features
-                .into_iter()
-                .filter_map(|feature| GlyphwiseSubsts::new(gsub_table, feature))
-                .collect(),
-        );
-
-        let ttf = font.ttf();
-        let space_width = ttf
-            .glyph_index(' ')
-            .and_then(|id| ttf.glyph_hor_advance(id))
-            .map(|advance| font.to_em(advance))
-            .unwrap_or(THICK);
+        // These unwraps are safe as the font given is one returned by the
+        // find_math_font function, which only returns fonts that have a math
+        // constants table.
+        let constants = font.ttf().tables().math.unwrap().constants.unwrap();
 
         Self {
             engine,
             locator,
             region: Region::new(base, Axes::splat(false)),
             font,
-            ttf,
-            table: math_table,
             constants,
-            dtls_table: feat(b"dtls"),
-            flac_table: feat(b"flac"),
-            ssty_table: feat(b"ssty"),
-            glyphwise_tables,
-            space_width,
             fragments: vec![],
         }
     }
@@ -529,7 +494,8 @@ fn layout_realized(
     if let Some(elem) = elem.to_packed::<TagElem>() {
         ctx.push(MathFragment::Tag(elem.tag.clone()));
     } else if elem.is::<SpaceElem>() {
-        ctx.push(MathFragment::Space(ctx.space_width.resolve(styles)));
+        let space_width = ctx.font.space_width().unwrap_or(THICK);
+        ctx.push(MathFragment::Space(space_width.resolve(styles)));
     } else if elem.is::<LinebreakElem>() {
         ctx.push(MathFragment::Linebreak);
     } else if let Some(elem) = elem.to_packed::<HElem>() {

crates/typst-layout/src/math/root.rs

@@ -49,9 +49,9 @@ pub fn layout_root(
 
     // Layout root symbol.
     let target = radicand.height() + thickness + gap;
-    let sqrt = GlyphFragment::new(ctx, styles, '√', span)
-        .stretch_vertical(ctx, target)
-        .frame;
+    let mut sqrt = GlyphFragment::new_char(ctx.font, styles, '√', span)?;
+    sqrt.stretch_vertical(ctx, target);
+    let sqrt = sqrt.into_frame();
 
     // Layout the index.
     let sscript = EquationElem::set_size(MathSize::ScriptScript).wrap();

crates/typst-layout/src/math/shared.rs

@@ -1,7 +1,9 @@
 use ttf_parser::math::MathValue;
+use ttf_parser::Tag;
 use typst_library::foundations::{Style, StyleChain};
-use typst_library::layout::{Abs, Em, FixedAlignment, Frame, Point, Size, VAlignment};
+use typst_library::layout::{Abs, Em, FixedAlignment, Frame, Point, Size};
 use typst_library::math::{EquationElem, MathSize};
+use typst_library::text::{FontFeatures, TextElem};
 use typst_utils::LazyHash;
 
 use super::{LeftRightAlternator, MathContext, MathFragment, MathRun};
@@ -59,6 +61,16 @@ pub fn style_cramped() -> LazyHash<Style> {
     EquationElem::set_cramped(true).wrap()
 }
 
+/// Sets flac OpenType feature.
+pub fn style_flac() -> LazyHash<Style> {
+    TextElem::set_features(FontFeatures(vec![(Tag::from_bytes(b"flac"), 1)])).wrap()
+}
+
+/// Sets dtls OpenType feature.
+pub fn style_dtls() -> LazyHash<Style> {
+    TextElem::set_features(FontFeatures(vec![(Tag::from_bytes(b"dtls"), 1)])).wrap()
+}
+
 /// The style for subscripts in the current style.
 pub fn style_for_subscript(styles: StyleChain) -> [LazyHash<Style>; 2] {
     [style_for_superscript(styles), EquationElem::set_cramped(true).wrap()]
@@ -97,15 +109,6 @@ pub fn style_for_script_scale(ctx: &MathContext) -> LazyHash<Style> {
     .wrap()
 }
 
-/// How a delimieter should be aligned when scaling.
-pub fn delimiter_alignment(delimiter: char) -> VAlignment {
-    match delimiter {
-        '⌜' | '⌝' => VAlignment::Top,
-        '⌞' | '⌟' => VAlignment::Bottom,
-        _ => VAlignment::Horizon,
-    }
-}
-
 /// Stack rows on top of each other.
 ///
 /// Add a `gap` between each row and uses the baseline of the `baseline`-th

crates/typst-layout/src/math/stretch.rs

@@ -1,19 +1,10 @@
-use ttf_parser::math::{GlyphAssembly, GlyphConstruction, GlyphPart};
-use ttf_parser::LazyArray16;
 use typst_library::diag::{warning, SourceResult};
 use typst_library::foundations::{Packed, StyleChain};
-use typst_library::layout::{Abs, Axis, Frame, Point, Rel, Size};
+use typst_library::layout::{Abs, Axis, Rel};
 use typst_library::math::StretchElem;
 use typst_utils::Get;
 
-use super::{
-    delimiter_alignment, GlyphFragment, MathContext, MathFragment, Scaled,
-    VariantFragment,
-};
-use crate::modifiers::FrameModify;
-
-/// Maximum number of times extenders can be repeated.
-const MAX_REPEATS: usize = 1024;
+use super::{stretch_axes, MathContext, MathFragment};
 
 /// Lays out a [`StretchElem`].
 #[typst_macros::time(name = "math.stretch", span = elem.span())]
@@ -23,15 +14,7 @@ pub fn layout_stretch(
     styles: StyleChain,
 ) -> SourceResult<()> {
     let mut fragment = ctx.layout_into_fragment(&elem.body, styles)?;
-    stretch_fragment(
-        ctx,
-        styles,
-        &mut fragment,
-        None,
-        None,
-        elem.size(styles),
-        Abs::zero(),
-    );
+    stretch_fragment(ctx, &mut fragment, None, None, elem.size(styles), Abs::zero());
     ctx.push(fragment);
     Ok(())
 }
@@ -39,266 +22,49 @@ pub fn layout_stretch(
 /// Attempts to stretch the given fragment by/to the amount given in stretch.
 pub fn stretch_fragment(
     ctx: &mut MathContext,
-    styles: StyleChain,
     fragment: &mut MathFragment,
     axis: Option<Axis>,
     relative_to: Option<Abs>,
     stretch: Rel<Abs>,
     short_fall: Abs,
 ) {
-    let glyph = match fragment {
-        MathFragment::Glyph(glyph) => glyph.clone(),
-        MathFragment::Variant(variant) => {
-            GlyphFragment::new(ctx, styles, variant.c, variant.span)
-        }
-        _ => return,
-    };
+    let size = fragment.size();
+
+    let MathFragment::Glyph(ref mut glyph) = fragment else { return };
 
     // Return if we attempt to stretch along an axis which isn't stretchable,
     // so that the original fragment isn't modified.
-    let Some(stretch_axis) = stretch_axis(ctx, &glyph) else { return };
-    let axis = axis.unwrap_or(stretch_axis);
-    if axis != stretch_axis {
-        return;
-    }
-
-    let relative_to_size = relative_to.unwrap_or_else(|| fragment.size().get(axis));
-
-    let mut variant = stretch_glyph(
-        ctx,
-        glyph,
-        stretch.relative_to(relative_to_size) - short_fall,
-        axis,
-    );
-
-    if axis == Axis::Y {
-        variant.align_on_axis(ctx, delimiter_alignment(variant.c));
-    }
-
-    *fragment = MathFragment::Variant(variant);
-}
-
-/// Return whether the glyph is stretchable and if it is, along which axis it
-/// can be stretched.
-fn stretch_axis(ctx: &mut MathContext, base: &GlyphFragment) -> Option<Axis> {
-    let base_id = base.id;
-    let vertical = ctx
-        .table
-        .variants
-        .and_then(|variants| variants.vertical_constructions.get(base_id))
-        .map(|_| Axis::Y);
-    let horizontal = ctx
-        .table
-        .variants
-        .and_then(|variants| variants.horizontal_constructions.get(base_id))
-        .map(|_| Axis::X);
-
-    match (vertical, horizontal) {
-        (vertical, None) => vertical,
-        (None, horizontal) => horizontal,
-        _ => {
-            // As far as we know, there aren't any glyphs that have both
-            // vertical and horizontal constructions. So for the time being, we
-            // will assume that a glyph cannot have both.
-            ctx.engine.sink.warn(warning!(
-               base.span,
-               "glyph has both vertical and horizontal constructions";
-               hint: "this is probably a font bug";
-               hint: "please file an issue at https://github.com/typst/typst/issues"
-            ));
-
-            None
+    let axes = stretch_axes(&glyph.item.font, glyph.base_glyph.id);
+    let stretch_axis = if let Some(axis) = axis {
+        if !axes.get(axis) {
+            return;
         }
-    }
-}
-
-/// Try to stretch a glyph to a desired width or height.
-///
-/// The resulting frame may not have the exact desired width.
-pub fn stretch_glyph(
-    ctx: &mut MathContext,
-    mut base: GlyphFragment,
-    target: Abs,
-    axis: Axis,
-) -> VariantFragment {
-    // If the base glyph is good enough, use it.
-    let advance = match axis {
-        Axis::X => base.width,
-        Axis::Y => base.height(),
-    };
-    if target <= advance {
-        return base.into_variant();
-    }
-
-    let mut min_overlap = Abs::zero();
-    let construction = ctx
-        .table
-        .variants
-        .and_then(|variants| {
-            min_overlap = variants.min_connector_overlap.scaled(ctx, base.font_size);
-            match axis {
-                Axis::X => variants.horizontal_constructions,
-                Axis::Y => variants.vertical_constructions,
+        axis
+    } else {
+        match (axes.x, axes.y) {
+            (true, false) => Axis::X,
+            (false, true) => Axis::Y,
+            (false, false) => return,
+            (true, true) => {
+                // As far as we know, there aren't any glyphs that have both
+                // vertical and horizontal constructions. So for the time being, we
+                // will assume that a glyph cannot have both.
+                ctx.engine.sink.warn(warning!(
+                   glyph.item.glyphs[0].span.0,
+                   "glyph has both vertical and horizontal constructions";
+                   hint: "this is probably a font bug";
+                   hint: "please file an issue at https://github.com/typst/typst/issues"
+                ));
+                return;
             }
-            .get(base.id)
-        })
-        .unwrap_or(GlyphConstruction { assembly: None, variants: LazyArray16::new(&[]) });
-
-    // Search for a pre-made variant with a good advance.
-    let mut best_id = base.id;
-    let mut best_advance = base.width;
-    for variant in construction.variants {
-        best_id = variant.variant_glyph;
-        best_advance = base.font.to_em(variant.advance_measurement).at(base.font_size);
-        if target <= best_advance {
-            break;
         }
-    }
-
-    // This is either good or the best we've got.
-    if target <= best_advance || construction.assembly.is_none() {
-        base.set_id(ctx, best_id);
-        return base.into_variant();
-    }
-
-    // Assemble from parts.
-    let assembly = construction.assembly.unwrap();
-    assemble(ctx, base, assembly, min_overlap, target, axis)
-}
-
-/// Assemble a glyph from parts.
-fn assemble(
-    ctx: &mut MathContext,
-    base: GlyphFragment,
-    assembly: GlyphAssembly,
-    min_overlap: Abs,
-    target: Abs,
-    axis: Axis,
-) -> VariantFragment {
-    // Determine the number of times the extenders need to be repeated as well
-    // as a ratio specifying how much to spread the parts apart
-    // (0 = maximal overlap, 1 = minimal overlap).
-    let mut full;
-    let mut ratio;
-    let mut repeat = 0;
-    loop {
-        full = Abs::zero();
-        ratio = 0.0;
-
-        let mut parts = parts(assembly, repeat).peekable();
-        let mut growable = Abs::zero();
-
-        while let Some(part) = parts.next() {
-            let mut advance = part.full_advance.scaled(ctx, base.font_size);
-            if let Some(next) = parts.peek() {
-                let max_overlap = part
-                    .end_connector_length
-                    .min(next.start_connector_length)
-                    .scaled(ctx, base.font_size);
-                if max_overlap < min_overlap {
-                    // This condition happening is indicative of a bug in the
-                    // font.
-                    ctx.engine.sink.warn(warning!(
-                       base.span,
-                       "glyph has assembly parts with overlap less than minConnectorOverlap";
-                       hint: "its rendering may appear broken - this is probably a font bug";
-                       hint: "please file an issue at https://github.com/typst/typst/issues"
-                    ));
-                }
-
-                advance -= max_overlap;
-                growable += max_overlap - min_overlap;
-            }
-
-            full += advance;
-        }
-
-        if full < target {
-            let delta = target - full;
-            ratio = (delta / growable).min(1.0);
-            full += ratio * growable;
-        }
-
-        if target <= full || repeat >= MAX_REPEATS {
-            break;
-        }
-
-        repeat += 1;
-    }
-
-    let mut selected = vec![];
-    let mut parts = parts(assembly, repeat).peekable();
-    while let Some(part) = parts.next() {
-        let mut advance = part.full_advance.scaled(ctx, base.font_size);
-        if let Some(next) = parts.peek() {
-            let max_overlap = part
-                .end_connector_length
-                .min(next.start_connector_length)
-                .scaled(ctx, base.font_size);
-            advance -= max_overlap;
-            advance += ratio * (max_overlap - min_overlap);
-        }
-
-        let mut fragment = base.clone();
-        fragment.set_id(ctx, part.glyph_id);
-        selected.push((fragment, advance));
-    }
-
-    let size;
-    let baseline;
-    match axis {
-        Axis::X => {
-            let height = base.ascent + base.descent;
-            size = Size::new(full, height);
-            baseline = base.ascent;
-        }
-        Axis::Y => {
-            let axis = ctx.constants.axis_height().scaled(ctx, base.font_size);
-            let width = selected.iter().map(|(f, _)| f.width).max().unwrap_or_default();
-            size = Size::new(width, full);
-            baseline = full / 2.0 + axis;
-        }
-    }
-
-    let mut frame = Frame::soft(size);
-    let mut offset = Abs::zero();
-    frame.set_baseline(baseline);
-    frame.modify(&base.modifiers);
-
-    for (fragment, advance) in selected {
-        let pos = match axis {
-            Axis::X => Point::new(offset, frame.baseline() - fragment.ascent),
-            Axis::Y => Point::with_y(full - offset - fragment.height()),
-        };
-        frame.push_frame(pos, fragment.into_frame());
-        offset += advance;
-    }
-
-    let accent_attach = match axis {
-        Axis::X => (frame.width() / 2.0, frame.width() / 2.0),
-        Axis::Y => base.accent_attach,
     };
 
-    VariantFragment {
-        c: base.c,
-        frame,
-        font_size: base.font_size,
-        italics_correction: Abs::zero(),
-        accent_attach,
-        class: base.class,
-        math_size: base.math_size,
-        span: base.span,
-        limits: base.limits,
-        mid_stretched: None,
-        extended_shape: true,
+    let relative_to_size = relative_to.unwrap_or_else(|| size.get(stretch_axis));
+
+    glyph.stretch(ctx, stretch.relative_to(relative_to_size) - short_fall, stretch_axis);
+
+    if stretch_axis == Axis::Y {
+        glyph.center_on_axis();
     }
 }
-
-/// Return an iterator over the assembly's parts with extenders repeated the
-/// specified number of times.
-fn parts(assembly: GlyphAssembly, repeat: usize) -> impl Iterator<Item = GlyphPart> + '_ {
-    assembly.parts.into_iter().flat_map(move |part| {
-        let count = if part.part_flags.extender() { repeat } else { 1 };
-        std::iter::repeat_n(part, count)
-    })
-}

crates/typst-layout/src/math/text.rs

@@ -12,7 +12,10 @@ use typst_syntax::{is_newline, Span};
 use unicode_math_class::MathClass;
 use unicode_segmentation::UnicodeSegmentation;
 
-use super::{FrameFragment, GlyphFragment, MathContext, MathFragment, MathRun};
+use super::{
+    has_dtls_feat, style_dtls, FrameFragment, GlyphFragment, MathContext, MathFragment,
+    MathRun,
+};
 
 /// Lays out a [`TextElem`].
 pub fn layout_text(
@@ -67,12 +70,7 @@ fn layout_inline_text(
         let mut fragments = vec![];
         for unstyled_c in text.chars() {
             let c = styled_char(styles, unstyled_c, false);
-            let mut glyph = GlyphFragment::new(ctx, styles, c, span);
-            match EquationElem::size_in(styles) {
-                MathSize::Script => glyph.make_script_size(ctx),
-                MathSize::ScriptScript => glyph.make_script_script_size(ctx),
-                _ => {}
-            }
+            let glyph = GlyphFragment::new_char(ctx.font, styles, c, span)?;
             fragments.push(glyph.into());
         }
         let frame = MathRun::new(fragments).into_frame(styles);
@@ -121,54 +119,45 @@ pub fn layout_symbol(
 ) -> SourceResult<()> {
     // Switch dotless char to normal when we have the dtls OpenType feature.
     // This should happen before the main styling pass.
-    let (unstyled_c, dtls) = match try_dotless(elem.text) {
-        Some(c) if ctx.dtls_table.is_some() => (c, true),
-        _ => (elem.text, false),
+    let dtls = style_dtls();
+    let (unstyled_c, symbol_styles) = match try_dotless(elem.text) {
+        Some(c) if has_dtls_feat(ctx.font) => (c, styles.chain(&dtls)),
+        _ => (elem.text, styles),
     };
     let c = styled_char(styles, unstyled_c, true);
-    let fragment = match GlyphFragment::try_new(ctx, styles, c, elem.span()) {
-        Some(glyph) => layout_glyph(glyph, dtls, ctx, styles),
-        None => {
-            // Not in the math font, fallback to normal inline text layout.
-            layout_inline_text(c.encode_utf8(&mut [0; 4]), elem.span(), ctx, styles)?
-                .into()
-        }
-    };
+    let fragment: MathFragment =
+        match GlyphFragment::new_char(ctx.font, symbol_styles, c, elem.span()) {
+            Ok(mut glyph) => {
+                adjust_glyph_layout(&mut glyph, ctx, styles);
+                glyph.into()
+            }
+            Err(_) => {
+                // Not in the math font, fallback to normal inline text layout.
+                // TODO: Should replace this with proper fallback in [`GlyphFragment::new`].
+                layout_inline_text(c.encode_utf8(&mut [0; 4]), elem.span(), ctx, styles)?
+                    .into()
+            }
+        };
     ctx.push(fragment);
     Ok(())
 }
 
-/// Layout a [`GlyphFragment`].
-fn layout_glyph(
-    mut glyph: GlyphFragment,
-    dtls: bool,
+/// Centers large glyphs vertically on the axis, scaling them if in display
+/// style.
+fn adjust_glyph_layout(
+    glyph: &mut GlyphFragment,
     ctx: &mut MathContext,
     styles: StyleChain,
-) -> MathFragment {
-    if dtls {
-        glyph.make_dotless_form(ctx);
-    }
-    let math_size = EquationElem::size_in(styles);
-    match math_size {
-        MathSize::Script => glyph.make_script_size(ctx),
-        MathSize::ScriptScript => glyph.make_script_script_size(ctx),
-        _ => {}
-    }
-
+) {
     if glyph.class == MathClass::Large {
-        let mut variant = if math_size == MathSize::Display {
+        if EquationElem::size_in(styles) == MathSize::Display {
             let height = scaled!(ctx, styles, display_operator_min_height)
-                .max(SQRT_2 * glyph.height());
-            glyph.stretch_vertical(ctx, height)
-        } else {
-            glyph.into_variant()
+                .max(SQRT_2 * glyph.size.y);
+            glyph.stretch_vertical(ctx, height);
         };
         // TeXbook p 155. Large operators are always vertically centered on the
         // axis.
-        variant.center_on_axis(ctx);
-        variant.into()
-    } else {
-        glyph.into()
+        glyph.center_on_axis();
     }
 }
 

crates/typst-layout/src/math/underover.rs

@@ -285,14 +285,14 @@ fn layout_underoverspreader(
     let body = ctx.layout_into_run(body, styles)?;
     let body_class = body.class();
     let body = body.into_fragment(styles);
-    let glyph = GlyphFragment::new(ctx, styles, c, span);
-    let stretched = glyph.stretch_horizontal(ctx, body.width());
+    let mut glyph = GlyphFragment::new_char(ctx.font, styles, c, span)?;
+    glyph.stretch_horizontal(ctx, body.width());
 
     let mut rows = vec![];
     let baseline = match position {
         Position::Under => {
             rows.push(MathRun::new(vec![body]));
-            rows.push(stretched.into());
+            rows.push(glyph.into());
             if let Some(annotation) = annotation {
                 let under_style = style_for_subscript(styles);
                 let annotation_styles = styles.chain(&under_style);
@@ -306,7 +306,7 @@ fn layout_underoverspreader(
                 let annotation_styles = styles.chain(&over_style);
                 rows.extend(ctx.layout_into_run(annotation, annotation_styles)?.rows());
             }
-            rows.push(stretched.into());
+            rows.push(glyph.into());
             rows.push(MathRun::new(vec![body]));
             rows.len() - 1
         }

crates/typst-library/Cargo.toml

@@ -66,6 +66,7 @@ unicode-normalization = { workspace = true }
 unicode-segmentation = { workspace = true }
 unscanny = { workspace = true }
 usvg = { workspace = true }
+utf8_iter = { workspace = true }
 wasmi = { workspace = true }
 xmlwriter = { workspace = true }
 

crates/typst-library/src/diag.rs

@@ -1,17 +1,20 @@
 //! Diagnostics.
 
-use std::fmt::{self, Display, Formatter};
+use std::fmt::{self, Display, Formatter, Write as _};
 use std::io;
 use std::path::{Path, PathBuf};
 use std::str::Utf8Error;
 use std::string::FromUtf8Error;
 
+use az::SaturatingAs;
 use comemo::Tracked;
 use ecow::{eco_vec, EcoVec};
 use typst_syntax::package::{PackageSpec, PackageVersion};
-use typst_syntax::{Span, Spanned, SyntaxError};
+use typst_syntax::{Lines, Span, Spanned, SyntaxError};
+use utf8_iter::ErrorReportingUtf8Chars;
 
 use crate::engine::Engine;
+use crate::loading::{LoadSource, Loaded};
 use crate::{World, WorldExt};
 
 /// Early-return with a [`StrResult`] or [`SourceResult`].
@@ -148,7 +151,7 @@ pub struct Warned<T> {
     pub warnings: EcoVec<SourceDiagnostic>,
 }
 
-/// An error or warning in a source file.
+/// An error or warning in a source or text file.
 ///
 /// The contained spans will only be detached if any of the input source files
 /// were detached.
@@ -568,31 +571,287 @@ impl From<PackageError> for EcoString {
     }
 }
 
+/// A result type with a data-loading-related error.
+pub type LoadResult<T> = Result<T, LoadError>;
+
+/// A call site independent error that occurred during data loading. This avoids
+/// polluting the memoization with [`Span`]s and [`FileId`]s from source files.
+/// Can be turned into a [`SourceDiagnostic`] using the [`LoadedWithin::within`]
+/// method available on [`LoadResult`].
+///
+/// [`FileId`]: typst_syntax::FileId
+#[derive(Clone, Debug, PartialEq, Eq, Hash)]
+pub struct LoadError {
+    /// The position in the file at which the error occured.
+    pos: ReportPos,
+    /// Must contain a message formatted like this: `"failed to do thing (cause)"`.
+    message: EcoString,
+}
+
+impl LoadError {
+    /// Creates a new error from a position in a file, a base message
+    /// (e.g. `failed to parse JSON`) and a concrete error (e.g. `invalid
+    /// number`)
+    pub fn new(
+        pos: impl Into<ReportPos>,
+        message: impl std::fmt::Display,
+        error: impl std::fmt::Display,
+    ) -> Self {
+        Self {
+            pos: pos.into(),
+            message: eco_format!("{message} ({error})"),
+        }
+    }
+}
+
+impl From<Utf8Error> for LoadError {
+    fn from(err: Utf8Error) -> Self {
+        let start = err.valid_up_to();
+        let end = start + err.error_len().unwrap_or(0);
+        LoadError::new(
+            start..end,
+            "failed to convert to string",
+            "file is not valid utf-8",
+        )
+    }
+}
+
+/// Convert a [`LoadResult`] to a [`SourceResult`] by adding the [`Loaded`]
+/// context.
+pub trait LoadedWithin<T> {
+    /// Report an error, possibly in an external file.
+    fn within(self, loaded: &Loaded) -> SourceResult<T>;
+}
+
+impl<T, E> LoadedWithin<T> for Result<T, E>
+where
+    E: Into<LoadError>,
+{
+    fn within(self, loaded: &Loaded) -> SourceResult<T> {
+        self.map_err(|err| {
+            let LoadError { pos, message } = err.into();
+            load_err_in_text(loaded, pos, message)
+        })
+    }
+}
+
+/// Report an error, possibly in an external file. This will delegate to
+/// [`load_err_in_invalid_text`] if the data isn't valid utf-8.
+fn load_err_in_text(
+    loaded: &Loaded,
+    pos: impl Into<ReportPos>,
+    mut message: EcoString,
+) -> EcoVec<SourceDiagnostic> {
+    let pos = pos.into();
+    // This also does utf-8 validation. Only report an error in an external
+    // file if it is human readable (valid utf-8), otherwise fall back to
+    // `load_err_in_invalid_text`.
+    let lines = Lines::try_from(&loaded.data);
+    match (loaded.source.v, lines) {
+        (LoadSource::Path(file_id), Ok(lines)) => {
+            if let Some(range) = pos.range(&lines) {
+                let span = Span::from_range(file_id, range);
+                return eco_vec![SourceDiagnostic::error(span, message)];
+            }
+
+            // Either `ReportPos::None` was provided, or resolving the range
+            // from the line/column failed. If present report the possibly
+            // wrong line/column in the error message anyway.
+            let span = Span::from_range(file_id, 0..loaded.data.len());
+            if let Some(pair) = pos.line_col(&lines) {
+                message.pop();
+                let (line, col) = pair.numbers();
+                write!(&mut message, " at {line}:{col})").ok();
+            }
+            eco_vec![SourceDiagnostic::error(span, message)]
+        }
+        (LoadSource::Bytes, Ok(lines)) => {
+            if let Some(pair) = pos.line_col(&lines) {
+                message.pop();
+                let (line, col) = pair.numbers();
+                write!(&mut message, " at {line}:{col})").ok();
+            }
+            eco_vec![SourceDiagnostic::error(loaded.source.span, message)]
+        }
+        _ => load_err_in_invalid_text(loaded, pos, message),
+    }
+}
+
+/// Report an error (possibly from an external file) that isn't valid utf-8.
+fn load_err_in_invalid_text(
+    loaded: &Loaded,
+    pos: impl Into<ReportPos>,
+    mut message: EcoString,
+) -> EcoVec<SourceDiagnostic> {
+    let line_col = pos.into().try_line_col(&loaded.data).map(|p| p.numbers());
+    match (loaded.source.v, line_col) {
+        (LoadSource::Path(file), _) => {
+            message.pop();
+            if let Some(package) = file.package() {
+                write!(
+                    &mut message,
+                    " in {package}{}",
+                    file.vpath().as_rooted_path().display()
+                )
+                .ok();
+            } else {
+                write!(&mut message, " in {}", file.vpath().as_rootless_path().display())
+                    .ok();
+            };
+            if let Some((line, col)) = line_col {
+                write!(&mut message, ":{line}:{col}").ok();
+            }
+            message.push(')');
+        }
+        (LoadSource::Bytes, Some((line, col))) => {
+            message.pop();
+            write!(&mut message, " at {line}:{col})").ok();
+        }
+        (LoadSource::Bytes, None) => (),
+    }
+    eco_vec![SourceDiagnostic::error(loaded.source.span, message)]
+}
+
+/// A position at which an error was reported.
+#[derive(Clone, Debug, Default, PartialEq, Eq, Hash)]
+pub enum ReportPos {
+    /// Contains a range, and a line/column pair.
+    Full(std::ops::Range<u32>, LineCol),
+    /// Contains a range.
+    Range(std::ops::Range<u32>),
+    /// Contains a line/column pair.
+    LineCol(LineCol),
+    #[default]
+    None,
+}
+
+impl From<std::ops::Range<usize>> for ReportPos {
+    fn from(value: std::ops::Range<usize>) -> Self {
+        Self::Range(value.start.saturating_as()..value.end.saturating_as())
+    }
+}
+
+impl From<LineCol> for ReportPos {
+    fn from(value: LineCol) -> Self {
+        Self::LineCol(value)
+    }
+}
+
+impl ReportPos {
+    /// Creates a position from a pre-existing range and line-column pair.
+    pub fn full(range: std::ops::Range<usize>, pair: LineCol) -> Self {
+        let range = range.start.saturating_as()..range.end.saturating_as();
+        Self::Full(range, pair)
+    }
+
+    /// Tries to determine the byte range for this position.
+    fn range(&self, lines: &Lines<String>) -> Option<std::ops::Range<usize>> {
+        match self {
+            ReportPos::Full(range, _) => Some(range.start as usize..range.end as usize),
+            ReportPos::Range(range) => Some(range.start as usize..range.end as usize),
+            &ReportPos::LineCol(pair) => {
+                let i =
+                    lines.line_column_to_byte(pair.line as usize, pair.col as usize)?;
+                Some(i..i)
+            }
+            ReportPos::None => None,
+        }
+    }
+
+    /// Tries to determine the line/column for this position.
+    fn line_col(&self, lines: &Lines<String>) -> Option<LineCol> {
+        match self {
+            &ReportPos::Full(_, pair) => Some(pair),
+            ReportPos::Range(range) => {
+                let (line, col) = lines.byte_to_line_column(range.start as usize)?;
+                Some(LineCol::zero_based(line, col))
+            }
+            &ReportPos::LineCol(pair) => Some(pair),
+            ReportPos::None => None,
+        }
+    }
+
+    /// Either gets the line/column pair, or tries to compute it from possibly
+    /// invalid utf-8 data.
+    fn try_line_col(&self, bytes: &[u8]) -> Option<LineCol> {
+        match self {
+            &ReportPos::Full(_, pair) => Some(pair),
+            ReportPos::Range(range) => {
+                LineCol::try_from_byte_pos(range.start as usize, bytes)
+            }
+            &ReportPos::LineCol(pair) => Some(pair),
+            ReportPos::None => None,
+        }
+    }
+}
+
+/// A line/column pair.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub struct LineCol {
+    /// The 0-based line.
+    line: u32,
+    /// The 0-based column.
+    col: u32,
+}
+
+impl LineCol {
+    /// Constructs the line/column pair from 0-based indices.
+    pub fn zero_based(line: usize, col: usize) -> Self {
+        Self {
+            line: line.saturating_as(),
+            col: col.saturating_as(),
+        }
+    }
+
+    /// Constructs the line/column pair from 1-based numbers.
+    pub fn one_based(line: usize, col: usize) -> Self {
+        Self::zero_based(line.saturating_sub(1), col.saturating_sub(1))
+    }
+
+    /// Try to compute a line/column pair from possibly invalid utf-8 data.
+    pub fn try_from_byte_pos(pos: usize, bytes: &[u8]) -> Option<Self> {
+        let bytes = &bytes[..pos];
+        let mut line = 0;
+        #[allow(clippy::double_ended_iterator_last)]
+        let line_start = memchr::memchr_iter(b'\n', bytes)
+            .inspect(|_| line += 1)
+            .last()
+            .map(|i| i + 1)
+            .unwrap_or(bytes.len());
+
+        let col = ErrorReportingUtf8Chars::new(&bytes[line_start..]).count();
+        Some(LineCol::zero_based(line, col))
+    }
+
+    /// Returns the 0-based line/column indices.
+    pub fn indices(&self) -> (usize, usize) {
+        (self.line as usize, self.col as usize)
+    }
+
+    /// Returns the 1-based line/column numbers.
+    pub fn numbers(&self) -> (usize, usize) {
+        (self.line as usize + 1, self.col as usize + 1)
+    }
+}
+
 /// Format a user-facing error message for an XML-like file format.
-pub fn format_xml_like_error(format: &str, error: roxmltree::Error) -> EcoString {
-    match error {
-        roxmltree::Error::UnexpectedCloseTag(expected, actual, pos) => {
-            eco_format!(
-                "failed to parse {format} (found closing tag '{actual}' \
-                 instead of '{expected}' in line {})",
-                pos.row
-            )
+pub fn format_xml_like_error(format: &str, error: roxmltree::Error) -> LoadError {
+    let pos = LineCol::one_based(error.pos().row as usize, error.pos().col as usize);
+    let message = match error {
+        roxmltree::Error::UnexpectedCloseTag(expected, actual, _) => {
+            eco_format!("failed to parse {format} (found closing tag '{actual}' instead of '{expected}')")
         }
-        roxmltree::Error::UnknownEntityReference(entity, pos) => {
-            eco_format!(
-                "failed to parse {format} (unknown entity '{entity}' in line {})",
-                pos.row
-            )
+        roxmltree::Error::UnknownEntityReference(entity, _) => {
+            eco_format!("failed to parse {format} (unknown entity '{entity}')")
         }
-        roxmltree::Error::DuplicatedAttribute(attr, pos) => {
-            eco_format!(
-                "failed to parse {format} (duplicate attribute '{attr}' in line {})",
-                pos.row
-            )
+        roxmltree::Error::DuplicatedAttribute(attr, _) => {
+            eco_format!("failed to parse {format} (duplicate attribute '{attr}')")
         }
         roxmltree::Error::NoRootNode => {
             eco_format!("failed to parse {format} (missing root node)")
         }
         err => eco_format!("failed to parse {format} ({err})"),
-    }
+    };
+
+    LoadError { pos: pos.into(), message }
 }

crates/typst-library/src/foundations/bytes.rs

@@ -7,6 +7,7 @@ use std::sync::Arc;
 
 use ecow::{eco_format, EcoString};
 use serde::{Serialize, Serializer};
+use typst_syntax::Lines;
 use typst_utils::LazyHash;
 
 use crate::diag::{bail, StrResult};
@@ -286,6 +287,16 @@ impl Serialize for Bytes {
     }
 }
 
+impl TryFrom<&Bytes> for Lines<String> {
+    type Error = Utf8Error;
+
+    #[comemo::memoize]
+    fn try_from(value: &Bytes) -> Result<Lines<String>, Utf8Error> {
+        let text = value.as_str()?;
+        Ok(Lines::new(text.to_string()))
+    }
+}
+
 /// Any type that can back a byte buffer.
 trait Bytelike: Send + Sync {
     fn as_bytes(&self) -> &[u8];

crates/typst-library/src/foundations/cast.rs

@@ -9,7 +9,7 @@ use std::ops::Add;
 
 use ecow::eco_format;
 use smallvec::SmallVec;
-use typst_syntax::{Span, Spanned};
+use typst_syntax::{Span, Spanned, SyntaxMode};
 use unicode_math_class::MathClass;
 
 use crate::diag::{At, HintedStrResult, HintedString, SourceResult, StrResult};
@@ -459,6 +459,21 @@ impl FromValue for Never {
     }
 }
 
+cast! {
+    SyntaxMode,
+    self => IntoValue::into_value(match self {
+        SyntaxMode::Markup => "markup",
+        SyntaxMode::Math => "math",
+        SyntaxMode::Code => "code",
+    }),
+    /// Evaluate as markup, as in a Typst file.
+    "markup" => SyntaxMode::Markup,
+    /// Evaluate as math, as in an equation.
+    "math" => SyntaxMode::Math,
+    /// Evaluate as code, as after a hash.
+    "code" => SyntaxMode::Code,
+}
+
 cast! {
     MathClass,
     self => IntoValue::into_value(match self {

crates/typst-library/src/foundations/int.rs

@@ -1,4 +1,6 @@
-use std::num::{NonZeroI64, NonZeroIsize, NonZeroU64, NonZeroUsize, ParseIntError};
+use std::num::{
+    NonZeroI64, NonZeroIsize, NonZeroU32, NonZeroU64, NonZeroUsize, ParseIntError,
+};
 
 use ecow::{eco_format, EcoString};
 use smallvec::SmallVec;
@@ -482,3 +484,16 @@ cast! {
             "number too large"
         })?,
 }
+
+cast! {
+    NonZeroU32,
+    self => Value::Int(self.get() as _),
+    v: i64 => v
+        .try_into()
+        .and_then(|v: u32| v.try_into())
+        .map_err(|_| if v <= 0 {
+            "number must be positive"
+        } else {
+            "number too large"
+        })?,
+}

crates/typst-library/src/foundations/mod.rs

@@ -69,6 +69,7 @@ pub use self::ty::*;
 pub use self::value::*;
 pub use self::version::*;
 pub use typst_macros::{scope, ty};
+use typst_syntax::SyntaxMode;
 
 #[rustfmt::skip]
 #[doc(hidden)]
@@ -83,7 +84,6 @@ use typst_syntax::Spanned;
 
 use crate::diag::{bail, SourceResult, StrResult};
 use crate::engine::Engine;
-use crate::routines::EvalMode;
 use crate::{Feature, Features};
 
 /// Hook up all `foundations` definitions.
@@ -273,8 +273,8 @@ pub fn eval(
     /// #eval("1_2^3", mode: "math")
     /// ```
     #[named]
-    #[default(EvalMode::Code)]
-    mode: EvalMode,
+    #[default(SyntaxMode::Code)]
+    mode: SyntaxMode,
     /// A scope of definitions that are made available.
     ///
     /// ```example

crates/typst-library/src/foundations/module.rs

@@ -7,9 +7,10 @@ use typst_syntax::FileId;
 use crate::diag::{bail, DeprecationSink, StrResult};
 use crate::foundations::{repr, ty, Content, Scope, Value};
 
-/// A module of definitions.
+/// A collection of variables and functions that are commonly related to
+/// a single theme.
 ///
-/// A module
+/// A module can
 /// - be built-in
 /// - stem from a [file import]($scripting/#modules)
 /// - stem from a [package import]($scripting/#packages) (and thus indirectly

crates/typst-library/src/foundations/plugin.rs

@@ -151,8 +151,8 @@ pub fn plugin(
     /// A [path]($syntax/#paths) to a WebAssembly file or raw WebAssembly bytes.
     source: Spanned<DataSource>,
 ) -> SourceResult<Module> {
-    let data = source.load(engine.world)?;
-    Plugin::module(data).at(source.span)
+    let loaded = source.load(engine.world)?;
+    Plugin::module(loaded.data).at(source.span)
 }
 
 #[scope]

crates/typst-library/src/foundations/symbol.rs

@@ -1,8 +1,8 @@
-use std::cmp::Reverse;
 use std::collections::{BTreeSet, HashMap};
 use std::fmt::{self, Debug, Display, Formatter, Write};
 use std::sync::Arc;
 
+use codex::ModifierSet;
 use ecow::{eco_format, EcoString};
 use serde::{Serialize, Serializer};
 use typst_syntax::{is_ident, Span, Spanned};
@@ -54,18 +54,18 @@ enum Repr {
     /// A native symbol that has no named variant.
     Single(char),
     /// A native symbol with multiple named variants.
-    Complex(&'static [(&'static str, char)]),
+    Complex(&'static [(ModifierSet<&'static str>, char)]),
     /// A symbol with multiple named variants, where some modifiers may have
     /// been applied. Also used for symbols defined at runtime by the user with
     /// no modifier applied.
-    Modified(Arc<(List, EcoString)>),
+    Modified(Arc<(List, ModifierSet<EcoString>)>),
 }
 
 /// A collection of symbols.
 #[derive(Clone, Eq, PartialEq, Hash)]
 enum List {
-    Static(&'static [(&'static str, char)]),
-    Runtime(Box<[(EcoString, char)]>),
+    Static(&'static [(ModifierSet<&'static str>, char)]),
+    Runtime(Box<[(ModifierSet<EcoString>, char)]>),
 }
 
 impl Symbol {
@@ -76,24 +76,26 @@ impl Symbol {
 
     /// Create a symbol with a static variant list.
     #[track_caller]
-    pub const fn list(list: &'static [(&'static str, char)]) -> Self {
+    pub const fn list(list: &'static [(ModifierSet<&'static str>, char)]) -> Self {
         debug_assert!(!list.is_empty());
         Self(Repr::Complex(list))
     }
 
     /// Create a symbol with a runtime variant list.
     #[track_caller]
-    pub fn runtime(list: Box<[(EcoString, char)]>) -> Self {
+    pub fn runtime(list: Box<[(ModifierSet<EcoString>, char)]>) -> Self {
         debug_assert!(!list.is_empty());
-        Self(Repr::Modified(Arc::new((List::Runtime(list), EcoString::new()))))
+        Self(Repr::Modified(Arc::new((List::Runtime(list), ModifierSet::default()))))
     }
 
     /// Get the symbol's character.
     pub fn get(&self) -> char {
         match &self.0 {
             Repr::Single(c) => *c,
-            Repr::Complex(_) => find(self.variants(), "").unwrap(),
-            Repr::Modified(arc) => find(self.variants(), &arc.1).unwrap(),
+            Repr::Complex(_) => ModifierSet::<&'static str>::default()
+                .best_match_in(self.variants())
+                .unwrap(),
+            Repr::Modified(arc) => arc.1.best_match_in(self.variants()).unwrap(),
         }
     }
 
@@ -128,16 +130,14 @@ impl Symbol {
     /// Apply a modifier to the symbol.
     pub fn modified(mut self, modifier: &str) -> StrResult<Self> {
         if let Repr::Complex(list) = self.0 {
-            self.0 = Repr::Modified(Arc::new((List::Static(list), EcoString::new())));
+            self.0 =
+                Repr::Modified(Arc::new((List::Static(list), ModifierSet::default())));
         }
 
         if let Repr::Modified(arc) = &mut self.0 {
             let (list, modifiers) = Arc::make_mut(arc);
-            if !modifiers.is_empty() {
-                modifiers.push('.');
-            }
-            modifiers.push_str(modifier);
-            if find(list.variants(), modifiers).is_some() {
+            modifiers.insert_raw(modifier);
+            if modifiers.best_match_in(list.variants()).is_some() {
                 return Ok(self);
             }
         }
@@ -146,7 +146,7 @@ impl Symbol {
     }
 
     /// The characters that are covered by this symbol.
-    pub fn variants(&self) -> impl Iterator<Item = (&str, char)> {
+    pub fn variants(&self) -> impl Iterator<Item = (ModifierSet<&str>, char)> {
         match &self.0 {
             Repr::Single(c) => Variants::Single(Some(*c).into_iter()),
             Repr::Complex(list) => Variants::Static(list.iter()),
@@ -156,17 +156,15 @@ impl Symbol {
 
     /// Possible modifiers.
     pub fn modifiers(&self) -> impl Iterator<Item = &str> + '_ {
-        let mut set = BTreeSet::new();
         let modifiers = match &self.0 {
-            Repr::Modified(arc) => arc.1.as_str(),
-            _ => "",
+            Repr::Modified(arc) => arc.1.as_deref(),
+            _ => ModifierSet::default(),
         };
-        for modifier in self.variants().flat_map(|(name, _)| name.split('.')) {
-            if !modifier.is_empty() && !contained(modifiers, modifier) {
-                set.insert(modifier);
-            }
-        }
-        set.into_iter()
+        self.variants()
+            .flat_map(|(m, _)| m)
+            .filter(|modifier| !modifier.is_empty() && !modifiers.contains(modifier))
+            .collect::<BTreeSet<_>>()
+            .into_iter()
     }
 }
 
@@ -256,7 +254,10 @@ impl Symbol {
             seen.insert(hash, i);
         }
 
-        let list = variants.into_iter().map(|s| (s.v.0, s.v.1)).collect();
+        let list = variants
+            .into_iter()
+            .map(|s| (ModifierSet::from_raw_dotted(s.v.0), s.v.1))
+            .collect();
         Ok(Symbol::runtime(list))
     }
 }
@@ -291,14 +292,23 @@ impl crate::foundations::Repr for Symbol {
         match &self.0 {
             Repr::Single(c) => eco_format!("symbol(\"{}\")", *c),
             Repr::Complex(variants) => {
-                eco_format!("symbol{}", repr_variants(variants.iter().copied(), ""))
+                eco_format!(
+                    "symbol{}",
+                    repr_variants(variants.iter().copied(), ModifierSet::default())
+                )
             }
             Repr::Modified(arc) => {
                 let (list, modifiers) = arc.as_ref();
                 if modifiers.is_empty() {
-                    eco_format!("symbol{}", repr_variants(list.variants(), ""))
+                    eco_format!(
+                        "symbol{}",
+                        repr_variants(list.variants(), ModifierSet::default())
+                    )
                 } else {
-                    eco_format!("symbol{}", repr_variants(list.variants(), modifiers))
+                    eco_format!(
+                        "symbol{}",
+                        repr_variants(list.variants(), modifiers.as_deref())
+                    )
                 }
             }
         }
@@ -306,24 +316,24 @@ impl crate::foundations::Repr for Symbol {
 }
 
 fn repr_variants<'a>(
-    variants: impl Iterator<Item = (&'a str, char)>,
-    applied_modifiers: &str,
+    variants: impl Iterator<Item = (ModifierSet<&'a str>, char)>,
+    applied_modifiers: ModifierSet<&str>,
 ) -> String {
     crate::foundations::repr::pretty_array_like(
         &variants
-            .filter(|(variant, _)| {
+            .filter(|(modifiers, _)| {
                 // Only keep variants that can still be accessed, i.e., variants
                 // that contain all applied modifiers.
-                parts(applied_modifiers).all(|am| variant.split('.').any(|m| m == am))
+                applied_modifiers.iter().all(|am| modifiers.contains(am))
             })
-            .map(|(variant, c)| {
-                let trimmed_variant = variant
-                    .split('.')
-                    .filter(|&m| parts(applied_modifiers).all(|am| m != am));
-                if trimmed_variant.clone().all(|m| m.is_empty()) {
+            .map(|(modifiers, c)| {
+                let trimmed_modifiers =
+                    modifiers.into_iter().filter(|&m| !applied_modifiers.contains(m));
+                if trimmed_modifiers.clone().all(|m| m.is_empty()) {
                     eco_format!("\"{c}\"")
                 } else {
-                    let trimmed_modifiers = trimmed_variant.collect::<Vec<_>>().join(".");
+                    let trimmed_modifiers =
+                        trimmed_modifiers.collect::<Vec<_>>().join(".");
                     eco_format!("(\"{}\", \"{}\")", trimmed_modifiers, c)
                 }
             })
@@ -369,67 +379,22 @@ cast! {
 /// Iterator over variants.
 enum Variants<'a> {
     Single(std::option::IntoIter<char>),
-    Static(std::slice::Iter<'static, (&'static str, char)>),
-    Runtime(std::slice::Iter<'a, (EcoString, char)>),
+    Static(std::slice::Iter<'static, (ModifierSet<&'static str>, char)>),
+    Runtime(std::slice::Iter<'a, (ModifierSet<EcoString>, char)>),
 }
 
 impl<'a> Iterator for Variants<'a> {
-    type Item = (&'a str, char);
+    type Item = (ModifierSet<&'a str>, char);
 
     fn next(&mut self) -> Option<Self::Item> {
         match self {
-            Self::Single(iter) => Some(("", iter.next()?)),
+            Self::Single(iter) => Some((ModifierSet::default(), iter.next()?)),
             Self::Static(list) => list.next().copied(),
-            Self::Runtime(list) => list.next().map(|(s, c)| (s.as_str(), *c)),
+            Self::Runtime(list) => list.next().map(|(m, c)| (m.as_deref(), *c)),
         }
     }
 }
 
-/// Find the best symbol from the list.
-fn find<'a>(
-    variants: impl Iterator<Item = (&'a str, char)>,
-    modifiers: &str,
-) -> Option<char> {
-    let mut best = None;
-    let mut best_score = None;
-
-    // Find the best table entry with this name.
-    'outer: for candidate in variants {
-        for modifier in parts(modifiers) {
-            if !contained(candidate.0, modifier) {
-                continue 'outer;
-            }
-        }
-
-        let mut matching = 0;
-        let mut total = 0;
-        for modifier in parts(candidate.0) {
-            if contained(modifiers, modifier) {
-                matching += 1;
-            }
-            total += 1;
-        }
-
-        let score = (matching, Reverse(total));
-        if best_score.is_none_or(|b| score > b) {
-            best = Some(candidate.1);
-            best_score = Some(score);
-        }
-    }
-
-    best
-}
-
-/// Split a modifier list into its parts.
-fn parts(modifiers: &str) -> impl Iterator<Item = &str> {
-    modifiers.split('.').filter(|s| !s.is_empty())
-}
-
-/// Whether the modifier string contains the modifier `m`.
-fn contained(modifiers: &str, m: &str) -> bool {
-    parts(modifiers).any(|part| part == m)
-}
-
 /// A single character.
 #[elem(Repr, PlainText)]
 pub struct SymbolElem {

crates/typst-library/src/layout/grid/mod.rs

@@ -1,6 +1,6 @@
 pub mod resolve;
 
-use std::num::NonZeroUsize;
+use std::num::{NonZeroU32, NonZeroUsize};
 use std::sync::Arc;
 
 use comemo::Track;
@@ -468,6 +468,17 @@ pub struct GridHeader {
     #[default(true)]
     pub repeat: bool,
 
+    /// The level of the header. Must not be zero.
+    ///
+    /// This allows repeating multiple headers at once. Headers with different
+    /// levels can repeat together, as long as they have ascending levels.
+    ///
+    /// Notably, when a header with a lower level starts repeating, all higher
+    /// or equal level headers stop repeating (they are "replaced" by the new
+    /// header).
+    #[default(NonZeroU32::ONE)]
+    pub level: NonZeroU32,
+
     /// The cells and lines within the header.
     #[variadic]
     pub children: Vec<GridItem>,
@@ -755,7 +766,14 @@ impl Show for Packed<GridCell> {
 
 impl Default for Packed<GridCell> {
     fn default() -> Self {
-        Packed::new(GridCell::new(Content::default()))
+        Packed::new(
+            // Explicitly set colspan and rowspan to ensure they won't be
+            // overridden by set rules (default cells are created after
+            // colspans and rowspans are processed in the resolver)
+            GridCell::new(Content::default())
+                .with_colspan(NonZeroUsize::ONE)
+                .with_rowspan(NonZeroUsize::ONE),
+        )
     }
 }
 

crates/typst-library/src/layout/grid/resolve.rs

@@ -1,5 +1,5 @@
-use std::num::NonZeroUsize;
-use std::ops::Range;
+use std::num::{NonZeroU32, NonZeroUsize};
+use std::ops::{Deref, DerefMut, Range};
 use std::sync::Arc;
 
 use ecow::eco_format;
@@ -48,6 +48,7 @@ pub fn grid_to_cellgrid<'a>(
     let children = elem.children.iter().map(|child| match child {
         GridChild::Header(header) => ResolvableGridChild::Header {
             repeat: header.repeat(styles),
+            level: header.level(styles),
             span: header.span(),
             items: header.children.iter().map(resolve_item),
         },
@@ -101,6 +102,7 @@ pub fn table_to_cellgrid<'a>(
     let children = elem.children.iter().map(|child| match child {
         TableChild::Header(header) => ResolvableGridChild::Header {
             repeat: header.repeat(styles),
+            level: header.level(styles),
             span: header.span(),
             items: header.children.iter().map(resolve_item),
         },
@@ -426,8 +428,20 @@ pub struct Line {
 /// A repeatable grid header. Starts at the first row.
 #[derive(Debug)]
 pub struct Header {
-    /// The index after the last row included in this header.
-    pub end: usize,
+    /// The range of rows included in this header.
+    pub range: Range<usize>,
+    /// The header's level.
+    ///
+    /// Higher level headers repeat together with lower level headers. If a
+    /// lower level header stops repeating, all higher level headers do as
+    /// well.
+    pub level: u32,
+    /// Whether this header cannot be repeated nor should have orphan
+    /// prevention because it would be about to cease repetition, either
+    /// because it is followed by headers of conflicting levels, or because
+    /// it is at the end of the table (possibly followed by some footers at the
+    /// end).
+    pub short_lived: bool,
 }
 
 /// A repeatable grid footer. Stops at the last row.
@@ -435,32 +449,56 @@ pub struct Header {
 pub struct Footer {
     /// The first row included in this footer.
     pub start: usize,
+    /// The index after the last row included in this footer.
+    pub end: usize,
+    /// The footer's level.
+    ///
+    /// Used similarly to header level.
+    pub level: u32,
 }
 
-/// A possibly repeatable grid object.
+impl Footer {
+    /// The footer's range of included rows.
+    #[inline]
+    pub fn range(&self) -> Range<usize> {
+        self.start..self.end
+    }
+}
+
+/// A possibly repeatable grid child (header or footer).
+///
 /// It still exists even when not repeatable, but must not have additional
 /// considerations by grid layout, other than for consistency (such as making
 /// a certain group of rows unbreakable).
-pub enum Repeatable<T> {
-    Repeated(T),
-    NotRepeated(T),
+pub struct Repeatable<T> {
+    inner: T,
+
+    /// Whether the user requested the child to repeat.
+    pub repeated: bool,
+}
+
+impl<T> Deref for Repeatable<T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        &self.inner
+    }
+}
+
+impl<T> DerefMut for Repeatable<T> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        &mut self.inner
+    }
 }
 
 impl<T> Repeatable<T> {
-    /// Gets the value inside this repeatable, regardless of whether
-    /// it repeats.
-    pub fn unwrap(&self) -> &T {
-        match self {
-            Self::Repeated(repeated) => repeated,
-            Self::NotRepeated(not_repeated) => not_repeated,
-        }
-    }
-
     /// Returns `Some` if the value is repeated, `None` otherwise.
+    #[inline]
     pub fn as_repeated(&self) -> Option<&T> {
-        match self {
-            Self::Repeated(repeated) => Some(repeated),
-            Self::NotRepeated(_) => None,
+        if self.repeated {
+            Some(&self.inner)
+        } else {
+            None
         }
     }
 }
@@ -617,7 +655,7 @@ impl<'a> Entry<'a> {
 
 /// Any grid child, which can be either a header or an item.
 pub enum ResolvableGridChild<T: ResolvableCell, I> {
-    Header { repeat: bool, span: Span, items: I },
+    Header { repeat: bool, level: NonZeroU32, span: Span, items: I },
     Footer { repeat: bool, span: Span, items: I },
     Item(ResolvableGridItem<T>),
 }
@@ -638,8 +676,8 @@ pub struct CellGrid<'a> {
     /// Gutter rows are not included.
     /// Contains up to 'rows_without_gutter.len() + 1' vectors of lines.
     pub hlines: Vec<Vec<Line>>,
-    /// The repeatable header of this grid.
-    pub header: Option<Repeatable<Header>>,
+    /// The repeatable headers of this grid.
+    pub headers: Vec<Repeatable<Header>>,
     /// The repeatable footer of this grid.
     pub footer: Option<Repeatable<Footer>>,
     /// Whether this grid has gutters.
@@ -654,7 +692,7 @@ impl<'a> CellGrid<'a> {
         cells: impl IntoIterator<Item = Cell<'a>>,
     ) -> Self {
         let entries = cells.into_iter().map(Entry::Cell).collect();
-        Self::new_internal(tracks, gutter, vec![], vec![], None, None, entries)
+        Self::new_internal(tracks, gutter, vec![], vec![], vec![], None, entries)
     }
 
     /// Generates the cell grid, given the tracks and resolved entries.
@@ -663,7 +701,7 @@ impl<'a> CellGrid<'a> {
         gutter: Axes<&[Sizing]>,
         vlines: Vec<Vec<Line>>,
         hlines: Vec<Vec<Line>>,
-        header: Option<Repeatable<Header>>,
+        headers: Vec<Repeatable<Header>>,
         footer: Option<Repeatable<Footer>>,
         entries: Vec<Entry<'a>>,
     ) -> Self {
@@ -717,7 +755,7 @@ impl<'a> CellGrid<'a> {
             entries,
             vlines,
             hlines,
-            header,
+            headers,
             footer,
             has_gutter,
         }
@@ -852,6 +890,11 @@ impl<'a> CellGrid<'a> {
             self.cols.len()
         }
     }
+
+    #[inline]
+    pub fn has_repeated_headers(&self) -> bool {
+        self.headers.iter().any(|h| h.repeated)
+    }
 }
 
 /// Resolves and positions all cells in the grid before creating it.
@@ -937,6 +980,12 @@ struct RowGroupData {
     span: Span,
     kind: RowGroupKind,
 
+    /// Whether this header or footer may repeat.
+    repeat: bool,
+
+    /// Level of this header or footer.
+    repeatable_level: NonZeroU32,
+
     /// Start of the range of indices of hlines at the top of the row group.
     /// This is always the first index after the last hline before we started
     /// building the row group - any upcoming hlines would appear at least at
@@ -984,14 +1033,17 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
         let mut pending_vlines: Vec<(Span, Line)> = vec![];
         let has_gutter = self.gutter.any(|tracks| !tracks.is_empty());
 
-        let mut header: Option<Header> = None;
-        let mut repeat_header = false;
+        let mut headers: Vec<Repeatable<Header>> = vec![];
 
         // Stores where the footer is supposed to end, its span, and the
         // actual footer structure.
         let mut footer: Option<(usize, Span, Footer)> = None;
         let mut repeat_footer = false;
 
+        // If true, there has been at least one cell besides headers and
+        // footers. When false, footers at the end are forced to not repeat.
+        let mut at_least_one_cell = false;
+
         // We can't just use the cell's index in the 'cells' vector to
         // determine its automatic position, since cells could have arbitrary
         // positions, so the position of a cell in 'cells' can differ from its
@@ -1008,6 +1060,11 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
         // automatically-positioned cell.
         let mut auto_index: usize = 0;
 
+        // The next header after the latest auto-positioned cell. This is used
+        // to avoid checking for collision with headers that were already
+        // skipped.
+        let mut next_header = 0;
+
         // We have to rebuild the grid to account for fixed cell positions.
         //
         // Create at least 'children.len()' positions, since there could be at
@@ -1028,12 +1085,13 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                 columns,
                 &mut pending_hlines,
                 &mut pending_vlines,
-                &mut header,
-                &mut repeat_header,
+                &mut headers,
                 &mut footer,
                 &mut repeat_footer,
                 &mut auto_index,
+                &mut next_header,
                 &mut resolved_cells,
+                &mut at_least_one_cell,
                 child,
             )?;
         }
@@ -1049,13 +1107,13 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
             row_amount,
         )?;
 
-        let (header, footer) = self.finalize_headers_and_footers(
+        let footer = self.finalize_headers_and_footers(
             has_gutter,
-            header,
-            repeat_header,
+            &mut headers,
             footer,
             repeat_footer,
             row_amount,
+            at_least_one_cell,
         )?;
 
         Ok(CellGrid::new_internal(
@@ -1063,7 +1121,7 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
             self.gutter,
             vlines,
             hlines,
-            header,
+            headers,
             footer,
             resolved_cells,
         ))
@@ -1083,12 +1141,13 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
         columns: usize,
         pending_hlines: &mut Vec<(Span, Line, bool)>,
         pending_vlines: &mut Vec<(Span, Line)>,
-        header: &mut Option<Header>,
-        repeat_header: &mut bool,
+        headers: &mut Vec<Repeatable<Header>>,
         footer: &mut Option<(usize, Span, Footer)>,
         repeat_footer: &mut bool,
         auto_index: &mut usize,
+        next_header: &mut usize,
         resolved_cells: &mut Vec<Option<Entry<'x>>>,
+        at_least_one_cell: &mut bool,
         child: ResolvableGridChild<T, I>,
     ) -> SourceResult<()>
     where
@@ -1112,7 +1171,32 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
         // position than it would usually be if it would be in a non-empty
         // row, so we must step a local index inside headers and footers
         // instead, and use a separate counter outside them.
-        let mut local_auto_index = *auto_index;
+        let local_auto_index = if matches!(child, ResolvableGridChild::Item(_)) {
+            auto_index
+        } else {
+            // Although 'usize' is Copy, we need to be explicit here that we
+            // aren't reborrowing the original auto index but rather making a
+            // mutable copy of it using 'clone'.
+            &mut (*auto_index).clone()
+        };
+
+        // NOTE: usually, if 'next_header' were to be updated inside a row
+        // group (indicating a header was skipped by a cell), that would
+        // indicate a collision between the row group and that header, which
+        // is an error. However, the exception is for the first auto cell of
+        // the row group, which may skip headers while searching for a position
+        // where to begin the row group in the first place.
+        //
+        // Therefore, we cannot safely share the counter in the row group with
+        // the counter used by auto cells outside, as it might update it in a
+        // valid situation, whereas it must not, since its auto cells use a
+        // different auto index counter and will have seen different headers,
+        // so we copy the next header counter while inside a row group.
+        let local_next_header = if matches!(child, ResolvableGridChild::Item(_)) {
+            next_header
+        } else {
+            &mut (*next_header).clone()
+        };
 
         // The first row in which this table group can fit.
         //
@@ -1123,23 +1207,19 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
         let mut first_available_row = 0;
 
         let (header_footer_items, simple_item) = match child {
-            ResolvableGridChild::Header { repeat, span, items, .. } => {
-                if header.is_some() {
-                    bail!(span, "cannot have more than one header");
-                }
-
+            ResolvableGridChild::Header { repeat, level, span, items, .. } => {
                 row_group_data = Some(RowGroupData {
                     range: None,
                     span,
                     kind: RowGroupKind::Header,
+                    repeat,
+                    repeatable_level: level,
                     top_hlines_start: pending_hlines.len(),
                     top_hlines_end: None,
                 });
 
-                *repeat_header = repeat;
-
                 first_available_row =
-                    find_next_empty_row(resolved_cells, local_auto_index, columns);
+                    find_next_empty_row(resolved_cells, *local_auto_index, columns);
 
                 // If any cell in the header is automatically positioned,
                 // have it skip to the next empty row. This is to avoid
@@ -1150,7 +1230,7 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                 // latest auto-position cell, since each auto-position cell
                 // always occupies the first available position after the
                 // previous one. Therefore, this will be >= auto_index.
-                local_auto_index = first_available_row * columns;
+                *local_auto_index = first_available_row * columns;
 
                 (Some(items), None)
             }
@@ -1162,21 +1242,27 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                 row_group_data = Some(RowGroupData {
                     range: None,
                     span,
+                    repeat,
                     kind: RowGroupKind::Footer,
+                    repeatable_level: NonZeroU32::ONE,
                     top_hlines_start: pending_hlines.len(),
                     top_hlines_end: None,
                 });
 
-                *repeat_footer = repeat;
-
                 first_available_row =
-                    find_next_empty_row(resolved_cells, local_auto_index, columns);
+                    find_next_empty_row(resolved_cells, *local_auto_index, columns);
 
-                local_auto_index = first_available_row * columns;
+                *local_auto_index = first_available_row * columns;
 
                 (Some(items), None)
             }
-            ResolvableGridChild::Item(item) => (None, Some(item)),
+            ResolvableGridChild::Item(item) => {
+                if matches!(item, ResolvableGridItem::Cell(_)) {
+                    *at_least_one_cell = true;
+                }
+
+                (None, Some(item))
+            }
         };
 
         let items = header_footer_items.into_iter().flatten().chain(simple_item);
@@ -1191,7 +1277,7 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                         // gutter.
                         skip_auto_index_through_fully_merged_rows(
                             resolved_cells,
-                            &mut local_auto_index,
+                            local_auto_index,
                             columns,
                         );
 
@@ -1266,7 +1352,7 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                         // automatically positioned cell. Same for footers.
                         local_auto_index
                             .checked_sub(1)
-                            .filter(|_| local_auto_index > first_available_row * columns)
+                            .filter(|_| *local_auto_index > first_available_row * columns)
                             .map_or(0, |last_auto_index| last_auto_index % columns + 1)
                     });
                     if end.is_some_and(|end| end.get() < start) {
@@ -1295,10 +1381,11 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                     cell_y,
                     colspan,
                     rowspan,
-                    header.as_ref(),
+                    headers,
                     footer.as_ref(),
                     resolved_cells,
-                    &mut local_auto_index,
+                    local_auto_index,
+                    local_next_header,
                     first_available_row,
                     columns,
                     row_group_data.is_some(),
@@ -1350,7 +1437,7 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                 );
 
                 if top_hlines_end.is_none()
-                    && local_auto_index > first_available_row * columns
+                    && *local_auto_index > first_available_row * columns
                 {
                     // Auto index was moved, so upcoming auto-pos hlines should
                     // no longer appear at the top.
@@ -1437,7 +1524,7 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                 None => {
                     // Empty header/footer: consider the header/footer to be
                     // at the next empty row after the latest auto index.
-                    local_auto_index = first_available_row * columns;
+                    *local_auto_index = first_available_row * columns;
                     let group_start = first_available_row;
                     let group_end = group_start + 1;
 
@@ -1454,8 +1541,8 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                     // 'find_next_empty_row' will skip through any existing headers
                     // and footers without having to loop through them each time.
                     // Cells themselves, unfortunately, still have to.
-                    assert!(resolved_cells[local_auto_index].is_none());
-                    resolved_cells[local_auto_index] =
+                    assert!(resolved_cells[*local_auto_index].is_none());
+                    resolved_cells[*local_auto_index] =
                         Some(Entry::Cell(self.resolve_cell(
                             T::default(),
                             0,
@@ -1483,21 +1570,38 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
 
             match row_group.kind {
                 RowGroupKind::Header => {
-                    if group_range.start != 0 {
-                        bail!(
-                            row_group.span,
-                            "header must start at the first row";
-                            hint: "remove any rows before the header"
-                        );
-                    }
-
-                    *header = Some(Header {
-                        // Later on, we have to correct this number in case there
+                    let data = Header {
+                        // Later on, we have to correct this range in case there
                         // is gutter. But only once all cells have been analyzed
                         // and the header has fully expanded in the fixup loop
                         // below.
-                        end: group_range.end,
-                    });
+                        range: group_range,
+
+                        level: row_group.repeatable_level.get(),
+
+                        // This can only change at a later iteration, if we
+                        // find a conflicting header or footer right away.
+                        short_lived: false,
+                    };
+
+                    // Mark consecutive headers right before this one as short
+                    // lived if they would have a higher or equal level, as
+                    // then they would immediately stop repeating during
+                    // layout.
+                    let mut consecutive_header_start = data.range.start;
+                    for conflicting_header in
+                        headers.iter_mut().rev().take_while(move |h| {
+                            let conflicts = h.range.end == consecutive_header_start
+                                && h.level >= data.level;
+
+                            consecutive_header_start = h.range.start;
+                            conflicts
+                        })
+                    {
+                        conflicting_header.short_lived = true;
+                    }
+
+                    headers.push(Repeatable { inner: data, repeated: row_group.repeat });
                 }
 
                 RowGroupKind::Footer => {
@@ -1514,15 +1618,14 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                             // before the footer might not be included as part of
                             // the footer if it is contained within the header.
                             start: group_range.start,
+                            end: group_range.end,
+                            level: 1,
                         },
                     ));
+
+                    *repeat_footer = row_group.repeat;
                 }
             }
-        } else {
-            // The child was a single cell outside headers or footers.
-            // Therefore, 'local_auto_index' for this table child was
-            // simply an alias for 'auto_index', so we update it as needed.
-            *auto_index = local_auto_index;
         }
 
         Ok(())
@@ -1689,47 +1792,64 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
     fn finalize_headers_and_footers(
         &self,
         has_gutter: bool,
-        header: Option<Header>,
-        repeat_header: bool,
+        headers: &mut [Repeatable<Header>],
         footer: Option<(usize, Span, Footer)>,
         repeat_footer: bool,
         row_amount: usize,
-    ) -> SourceResult<(Option<Repeatable<Header>>, Option<Repeatable<Footer>>)> {
-        let header = header
-            .map(|mut header| {
-                // Repeat the gutter below a header (hence why we don't
-                // subtract 1 from the gutter case).
-                // Don't do this if there are no rows under the header.
-                if has_gutter {
-                    // - 'header.end' is always 'last y + 1'. The header stops
-                    // before that row.
-                    // - Therefore, '2 * header.end' will be 2 * (last y + 1),
-                    // which is the adjusted index of the row before which the
-                    // header stops, meaning it will still stop right before it
-                    // even with gutter thanks to the multiplication below.
-                    // - This means that it will span all rows up to
-                    // '2 * (last y + 1) - 1 = 2 * last y + 1', which equates
-                    // to the index of the gutter row right below the header,
-                    // which is what we want (that gutter spacing should be
-                    // repeated across pages to maintain uniformity).
-                    header.end *= 2;
+        at_least_one_cell: bool,
+    ) -> SourceResult<Option<Repeatable<Footer>>> {
+        // Mark consecutive headers right before the end of the table, or the
+        // final footer, as short lived, given that there are no normal rows
+        // after them, so repeating them is pointless.
+        //
+        // It is important to do this BEFORE we update header and footer ranges
+        // due to gutter below as 'row_amount' doesn't consider gutter.
+        //
+        // TODO(subfooters): take the last footer if it is at the end and
+        // backtrack through consecutive footers until the first one in the
+        // sequence is found. If there is no footer at the end, there are no
+        // haeders to turn short-lived.
+        let mut consecutive_header_start =
+            footer.as_ref().map(|(_, _, f)| f.start).unwrap_or(row_amount);
+        for header_at_the_end in headers.iter_mut().rev().take_while(move |h| {
+            let at_the_end = h.range.end == consecutive_header_start;
 
-                    // If the header occupies the entire grid, ensure we don't
-                    // include an extra gutter row when it doesn't exist, since
-                    // the last row of the header is at the very bottom,
-                    // therefore '2 * last y + 1' is not a valid index.
-                    let row_amount = (2 * row_amount).saturating_sub(1);
-                    header.end = header.end.min(row_amount);
-                }
-                header
-            })
-            .map(|header| {
-                if repeat_header {
-                    Repeatable::Repeated(header)
-                } else {
-                    Repeatable::NotRepeated(header)
-                }
-            });
+            consecutive_header_start = h.range.start;
+            at_the_end
+        }) {
+            header_at_the_end.short_lived = true;
+        }
+
+        // Repeat the gutter below a header (hence why we don't
+        // subtract 1 from the gutter case).
+        // Don't do this if there are no rows under the header.
+        if has_gutter {
+            for header in &mut *headers {
+                // Index of first y is doubled, as each row before it
+                // receives a gutter row below.
+                header.range.start *= 2;
+
+                // - 'header.end' is always 'last y + 1'. The header stops
+                // before that row.
+                // - Therefore, '2 * header.end' will be 2 * (last y + 1),
+                // which is the adjusted index of the row before which the
+                // header stops, meaning it will still stop right before it
+                // even with gutter thanks to the multiplication below.
+                // - This means that it will span all rows up to
+                // '2 * (last y + 1) - 1 = 2 * last y + 1', which equates
+                // to the index of the gutter row right below the header,
+                // which is what we want (that gutter spacing should be
+                // repeated across pages to maintain uniformity).
+                header.range.end *= 2;
+
+                // If the header occupies the entire grid, ensure we don't
+                // include an extra gutter row when it doesn't exist, since
+                // the last row of the header is at the very bottom,
+                // therefore '2 * last y + 1' is not a valid index.
+                let row_amount = (2 * row_amount).saturating_sub(1);
+                header.range.end = header.range.end.min(row_amount);
+            }
+        }
 
         let footer = footer
             .map(|(footer_end, footer_span, mut footer)| {
@@ -1737,8 +1857,17 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                     bail!(footer_span, "footer must end at the last row");
                 }
 
-                let header_end =
-                    header.as_ref().map(Repeatable::unwrap).map(|header| header.end);
+                // TODO(subfooters): will need a global slice of headers and
+                // footers for when we have multiple footers
+                // Alternatively, never include the gutter in the footer's
+                // range and manually add it later on layout. This would allow
+                // laying out the gutter as part of both the header and footer,
+                // and, if the page only has headers, the gutter row below the
+                // header is automatically removed (as it becomes the last), so
+                // only the gutter above the footer is kept, ensuring the same
+                // gutter row isn't laid out two times in a row. When laying
+                // out the footer for real, the mechanism can be disabled.
+                let last_header_end = headers.last().map(|header| header.range.end);
 
                 if has_gutter {
                     // Convert the footer's start index to post-gutter coordinates.
@@ -1747,23 +1876,38 @@ impl<'x> CellGridResolver<'_, '_, 'x> {
                     // Include the gutter right before the footer, unless there is
                     // none, or the gutter is already included in the header (no
                     // rows between the header and the footer).
-                    if header_end != Some(footer.start) {
+                    if last_header_end != Some(footer.start) {
                         footer.start = footer.start.saturating_sub(1);
                     }
+
+                    // Adapt footer end but DO NOT include the gutter below it,
+                    // if it exists. Calculation:
+                    // - Starts as 'last y + 1'.
+                    // - The result will be
+                    // 2 * (last_y + 1) - 1 = 2 * last_y + 1,
+                    // which is the new index of the last footer row plus one,
+                    // meaning we do exclude any gutter below this way.
+                    //
+                    // It also keeps us within the total amount of rows, so we
+                    // don't need to '.min()' later.
+                    footer.end = (2 * footer.end).saturating_sub(1);
                 }
 
                 Ok(footer)
             })
             .transpose()?
             .map(|footer| {
-                if repeat_footer {
-                    Repeatable::Repeated(footer)
-                } else {
-                    Repeatable::NotRepeated(footer)
+                // Don't repeat footers when the table only has headers and
+                // footers.
+                // TODO(subfooters): Switch this to marking the last N
+                // consecutive footers as short lived.
+                Repeatable {
+                    inner: footer,
+                    repeated: repeat_footer && at_least_one_cell,
                 }
             });
 
-        Ok((header, footer))
+        Ok(footer)
     }
 
     /// Resolves the cell's fields based on grid-wide properties.
@@ -1934,28 +2078,28 @@ fn expand_row_group(
 
 /// Check if a cell's fixed row would conflict with a header or footer.
 fn check_for_conflicting_cell_row(
-    header: Option<&Header>,
+    headers: &[Repeatable<Header>],
     footer: Option<&(usize, Span, Footer)>,
     cell_y: usize,
     rowspan: usize,
 ) -> HintedStrResult<()> {
-    if let Some(header) = header {
-        // TODO: check start (right now zero, always satisfied)
-        if cell_y < header.end {
-            bail!(
-                "cell would conflict with header spanning the same position";
-                hint: "try moving the cell or the header"
-            );
-        }
+    // NOTE: y + rowspan >, not >=, header.start, to check if the rowspan
+    // enters the header. For example, consider a rowspan of 1: if
+    // `y + 1 = header.start` holds, that means `y < header.start`, and it
+    // only occupies one row (`y`), so the cell is actually not in
+    // conflict.
+    if headers
+        .iter()
+        .any(|header| cell_y < header.range.end && cell_y + rowspan > header.range.start)
+    {
+        bail!(
+            "cell would conflict with header spanning the same position";
+            hint: "try moving the cell or the header"
+        );
     }
 
-    if let Some((footer_end, _, footer)) = footer {
-        // NOTE: y + rowspan >, not >=, footer.start, to check if the rowspan
-        // enters the footer. For example, consider a rowspan of 1: if
-        // `y + 1 = footer.start` holds, that means `y < footer.start`, and it
-        // only occupies one row (`y`), so the cell is actually not in
-        // conflict.
-        if cell_y < *footer_end && cell_y + rowspan > footer.start {
+    if let Some((_, _, footer)) = footer {
+        if cell_y < footer.end && cell_y + rowspan > footer.start {
             bail!(
                 "cell would conflict with footer spanning the same position";
                 hint: "try reducing the cell's rowspan or moving the footer"
@@ -1981,10 +2125,11 @@ fn resolve_cell_position(
     cell_y: Smart<usize>,
     colspan: usize,
     rowspan: usize,
-    header: Option<&Header>,
+    headers: &[Repeatable<Header>],
     footer: Option<&(usize, Span, Footer)>,
     resolved_cells: &[Option<Entry>],
     auto_index: &mut usize,
+    next_header: &mut usize,
     first_available_row: usize,
     columns: usize,
     in_row_group: bool,
@@ -2005,12 +2150,14 @@ fn resolve_cell_position(
             // Note that the counter ignores any cells with fixed positions,
             // but automatically-positioned cells will avoid conflicts by
             // simply skipping existing cells, headers and footers.
-            let resolved_index = find_next_available_position::<false>(
-                header,
+            let resolved_index = find_next_available_position(
+                headers,
                 footer,
                 resolved_cells,
                 columns,
                 *auto_index,
+                next_header,
+                false,
             )?;
 
             // Ensure the next cell with automatic position will be
@@ -2046,7 +2193,7 @@ fn resolve_cell_position(
                 // footer (but only if it isn't already in one, otherwise there
                 // will already be a separate check).
                 if !in_row_group {
-                    check_for_conflicting_cell_row(header, footer, cell_y, rowspan)?;
+                    check_for_conflicting_cell_row(headers, footer, cell_y, rowspan)?;
                 }
 
                 cell_index(cell_x, cell_y)
@@ -2063,12 +2210,28 @@ fn resolve_cell_position(
                 // requested column ('Some(None)') or an out of bounds position
                 // ('None'), in which case we'd create a new row to place this
                 // cell in.
-                find_next_available_position::<true>(
-                    header,
+                find_next_available_position(
+                    headers,
                     footer,
                     resolved_cells,
                     columns,
                     initial_index,
+                    // Make our own copy of the 'next_header' counter, since it
+                    // should only be updated by auto cells. However, we cannot
+                    // start with the same value as we are searching from the
+                    // start, and not from 'auto_index', so auto cells might
+                    // have skipped some headers already which this cell will
+                    // also need to skip.
+                    //
+                    // We could, in theory, keep a separate 'next_header'
+                    // counter for cells with fixed columns. But then we would
+                    // need one for every column, and much like how there isn't
+                    // an index counter for each column either, the potential
+                    // speed gain seems less relevant for a less used feature.
+                    // Still, it is something to consider for the future if
+                    // this turns out to be a bottleneck in important cases.
+                    &mut 0,
+                    true,
                 )
             }
         }
@@ -2078,7 +2241,7 @@ fn resolve_cell_position(
             // footer (but only if it isn't already in one, otherwise there
             // will already be a separate check).
             if !in_row_group {
-                check_for_conflicting_cell_row(header, footer, cell_y, rowspan)?;
+                check_for_conflicting_cell_row(headers, footer, cell_y, rowspan)?;
             }
 
             // Let's find the first column which has that row available.
@@ -2110,13 +2273,18 @@ fn resolve_cell_position(
 /// Finds the first available position after the initial index in the resolved
 /// grid of cells. Skips any non-absent positions (positions which already
 /// have cells specified by the user) as well as any headers and footers.
+///
+/// When `skip_rows` is true, one row is skipped on each iteration, preserving
+/// the column. That is used to find a position for a fixed column cell.
 #[inline]
-fn find_next_available_position<const SKIP_ROWS: bool>(
-    header: Option<&Header>,
+fn find_next_available_position(
+    headers: &[Repeatable<Header>],
     footer: Option<&(usize, Span, Footer)>,
     resolved_cells: &[Option<Entry<'_>>],
     columns: usize,
     initial_index: usize,
+    next_header: &mut usize,
+    skip_rows: bool,
 ) -> HintedStrResult<usize> {
     let mut resolved_index = initial_index;
 
@@ -2126,7 +2294,7 @@ fn find_next_available_position<const SKIP_ROWS: bool>(
             // determine where this cell will be placed. An out of
             // bounds position (thus `None`) is also a valid new
             // position (only requires expanding the vector).
-            if SKIP_ROWS {
+            if skip_rows {
                 // Skip one row at a time (cell chose its column, so we don't
                 // change it).
                 resolved_index =
@@ -2139,24 +2307,33 @@ fn find_next_available_position<const SKIP_ROWS: bool>(
                 // would become impractically large before this overflows.
                 resolved_index += 1;
             }
-        } else if let Some(header) =
-            header.filter(|header| resolved_index < header.end * columns)
+        } else if let Some(header) = headers
+            .get(*next_header)
+            .filter(|header| resolved_index >= header.range.start * columns)
         {
             // Skip header (can't place a cell inside it from outside it).
-            resolved_index = header.end * columns;
+            // No changes needed if we already passed this header (which
+            // also triggers this branch) - in that case, we only update the
+            // counter.
+            if resolved_index < header.range.end * columns {
+                resolved_index = header.range.end * columns;
 
-            if SKIP_ROWS {
-                // Ensure the cell's chosen column is kept after the
-                // header.
-                resolved_index += initial_index % columns;
+                if skip_rows {
+                    // Ensure the cell's chosen column is kept after the
+                    // header.
+                    resolved_index += initial_index % columns;
+                }
             }
+
+            // From now on, only check the headers afterwards.
+            *next_header += 1;
         } else if let Some((footer_end, _, _)) = footer.filter(|(end, _, footer)| {
             resolved_index >= footer.start * columns && resolved_index < *end * columns
         }) {
             // Skip footer, for the same reason.
             resolved_index = *footer_end * columns;
 
-            if SKIP_ROWS {
+            if skip_rows {
                 resolved_index += initial_index % columns;
             }
         } else {

crates/typst-library/src/loading/cbor.rs

@@ -23,8 +23,8 @@ pub fn cbor(
     /// A [path]($syntax/#paths) to a CBOR file or raw CBOR bytes.
     source: Spanned<DataSource>,
 ) -> SourceResult<Value> {
-    let data = source.load(engine.world)?;
-    ciborium::from_reader(data.as_slice())
+    let loaded = source.load(engine.world)?;
+    ciborium::from_reader(loaded.data.as_slice())
         .map_err(|err| eco_format!("failed to parse CBOR ({err})"))
         .at(source.span)
 }

crates/typst-library/src/loading/csv.rs

@@ -1,7 +1,7 @@
-use ecow::{eco_format, EcoString};
+use az::SaturatingAs;
 use typst_syntax::Spanned;
 
-use crate::diag::{bail, At, SourceResult};
+use crate::diag::{bail, LineCol, LoadError, LoadedWithin, ReportPos, SourceResult};
 use crate::engine::Engine;
 use crate::foundations::{cast, func, scope, Array, Dict, IntoValue, Type, Value};
 use crate::loading::{DataSource, Load, Readable};
@@ -44,7 +44,7 @@ pub fn csv(
     #[default(RowType::Array)]
     row_type: RowType,
 ) -> SourceResult<Array> {
-    let data = source.load(engine.world)?;
+    let loaded = source.load(engine.world)?;
 
     let mut builder = ::csv::ReaderBuilder::new();
     let has_headers = row_type == RowType::Dict;
@@ -53,7 +53,7 @@ pub fn csv(
 
     // Counting lines from 1 by default.
     let mut line_offset: usize = 1;
-    let mut reader = builder.from_reader(data.as_slice());
+    let mut reader = builder.from_reader(loaded.data.as_slice());
     let mut headers: Option<::csv::StringRecord> = None;
 
     if has_headers {
@@ -62,9 +62,9 @@ pub fn csv(
         headers = Some(
             reader
                 .headers()
+                .cloned()
                 .map_err(|err| format_csv_error(err, 1))
-                .at(source.span)?
-                .clone(),
+                .within(&loaded)?,
         );
     }
 
@@ -74,7 +74,7 @@ pub fn csv(
         // incorrect with `has_headers` set to `false`. See issue:
         // https://github.com/BurntSushi/rust-csv/issues/184
         let line = line + line_offset;
-        let row = result.map_err(|err| format_csv_error(err, line)).at(source.span)?;
+        let row = result.map_err(|err| format_csv_error(err, line)).within(&loaded)?;
         let item = if let Some(headers) = &headers {
             let mut dict = Dict::new();
             for (field, value) in headers.iter().zip(&row) {
@@ -164,15 +164,23 @@ cast! {
 }
 
 /// Format the user-facing CSV error message.
-fn format_csv_error(err: ::csv::Error, line: usize) -> EcoString {
+fn format_csv_error(err: ::csv::Error, line: usize) -> LoadError {
+    let msg = "failed to parse CSV";
+    let pos = (err.kind().position())
+        .map(|pos| {
+            let start = pos.byte().saturating_as();
+            ReportPos::from(start..start)
+        })
+        .unwrap_or(LineCol::one_based(line, 1).into());
     match err.kind() {
-        ::csv::ErrorKind::Utf8 { .. } => "file is not valid utf-8".into(),
-        ::csv::ErrorKind::UnequalLengths { expected_len, len, .. } => {
-            eco_format!(
-                "failed to parse CSV (found {len} instead of \
-                 {expected_len} fields in line {line})"
-            )
+        ::csv::ErrorKind::Utf8 { .. } => {
+            LoadError::new(pos, msg, "file is not valid utf-8")
         }
-        _ => eco_format!("failed to parse CSV ({err})"),
+        ::csv::ErrorKind::UnequalLengths { expected_len, len, .. } => {
+            let err =
+                format!("found {len} instead of {expected_len} fields in line {line}");
+            LoadError::new(pos, msg, err)
+        }
+        _ => LoadError::new(pos, "failed to parse CSV", err),
     }
 }

crates/typst-library/src/loading/json.rs

@@ -1,7 +1,7 @@
 use ecow::eco_format;
 use typst_syntax::Spanned;
 
-use crate::diag::{At, SourceResult};
+use crate::diag::{At, LineCol, LoadError, LoadedWithin, SourceResult};
 use crate::engine::Engine;
 use crate::foundations::{func, scope, Str, Value};
 use crate::loading::{DataSource, Load, Readable};
@@ -54,10 +54,13 @@ pub fn json(
     /// A [path]($syntax/#paths) to a JSON file or raw JSON bytes.
     source: Spanned<DataSource>,
 ) -> SourceResult<Value> {
-    let data = source.load(engine.world)?;
-    serde_json::from_slice(data.as_slice())
-        .map_err(|err| eco_format!("failed to parse JSON ({err})"))
-        .at(source.span)
+    let loaded = source.load(engine.world)?;
+    serde_json::from_slice(loaded.data.as_slice())
+        .map_err(|err| {
+            let pos = LineCol::one_based(err.line(), err.column());
+            LoadError::new(pos, "failed to parse JSON", err)
+        })
+        .within(&loaded)
 }
 
 #[scope]

crates/typst-library/src/loading/mod.rs

@@ -17,7 +17,7 @@ mod yaml_;
 
 use comemo::Tracked;
 use ecow::EcoString;
-use typst_syntax::Spanned;
+use typst_syntax::{FileId, Spanned};
 
 pub use self::cbor_::*;
 pub use self::csv_::*;
@@ -74,39 +74,44 @@ pub trait Load {
 }
 
 impl Load for Spanned<DataSource> {
-    type Output = Bytes;
+    type Output = Loaded;
 
-    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Bytes> {
+    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Self::Output> {
         self.as_ref().load(world)
     }
 }
 
 impl Load for Spanned<&DataSource> {
-    type Output = Bytes;
+    type Output = Loaded;
 
-    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Bytes> {
+    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Self::Output> {
         match &self.v {
             DataSource::Path(path) => {
                 let file_id = self.span.resolve_path(path).at(self.span)?;
-                world.file(file_id).at(self.span)
+                let data = world.file(file_id).at(self.span)?;
+                let source = Spanned::new(LoadSource::Path(file_id), self.span);
+                Ok(Loaded::new(source, data))
+            }
+            DataSource::Bytes(data) => {
+                let source = Spanned::new(LoadSource::Bytes, self.span);
+                Ok(Loaded::new(source, data.clone()))
             }
-            DataSource::Bytes(bytes) => Ok(bytes.clone()),
         }
     }
 }
 
 impl Load for Spanned<OneOrMultiple<DataSource>> {
-    type Output = Vec<Bytes>;
+    type Output = Vec<Loaded>;
 
-    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Vec<Bytes>> {
+    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Self::Output> {
         self.as_ref().load(world)
     }
 }
 
 impl Load for Spanned<&OneOrMultiple<DataSource>> {
-    type Output = Vec<Bytes>;
+    type Output = Vec<Loaded>;
 
-    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Vec<Bytes>> {
+    fn load(&self, world: Tracked<dyn World + '_>) -> SourceResult<Self::Output> {
         self.v
             .0
             .iter()
@@ -115,6 +120,28 @@ impl Load for Spanned<&OneOrMultiple<DataSource>> {
     }
 }
 
+/// Data loaded from a [`DataSource`].
+#[derive(Clone, Debug, PartialEq, Eq, Hash)]
+pub struct Loaded {
+    /// Details about where `data` was loaded from.
+    pub source: Spanned<LoadSource>,
+    /// The loaded data.
+    pub data: Bytes,
+}
+
+impl Loaded {
+    pub fn new(source: Spanned<LoadSource>, bytes: Bytes) -> Self {
+        Self { source, data: bytes }
+    }
+}
+
+/// A loaded [`DataSource`].
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum LoadSource {
+    Path(FileId),
+    Bytes,
+}
+
 /// A value that can be read from a file.
 #[derive(Debug, Clone, PartialEq, Hash)]
 pub enum Readable {

crates/typst-library/src/loading/read.rs

@@ -1,11 +1,10 @@
 use ecow::EcoString;
 use typst_syntax::Spanned;
 
-use crate::diag::{At, FileError, SourceResult};
+use crate::diag::{LoadedWithin, SourceResult};
 use crate::engine::Engine;
 use crate::foundations::{func, Cast};
-use crate::loading::Readable;
-use crate::World;
+use crate::loading::{DataSource, Load, Readable};
 
 /// Reads plain text or data from a file.
 ///
@@ -36,14 +35,10 @@ pub fn read(
     #[default(Some(Encoding::Utf8))]
     encoding: Option<Encoding>,
 ) -> SourceResult<Readable> {
-    let Spanned { v: path, span } = path;
-    let id = span.resolve_path(&path).at(span)?;
-    let data = engine.world.file(id).at(span)?;
+    let loaded = path.map(DataSource::Path).load(engine.world)?;
     Ok(match encoding {
-        None => Readable::Bytes(data),
-        Some(Encoding::Utf8) => {
-            Readable::Str(data.to_str().map_err(FileError::from).at(span)?)
-        }
+        None => Readable::Bytes(loaded.data),
+        Some(Encoding::Utf8) => Readable::Str(loaded.data.to_str().within(&loaded)?),
     })
 }
 

crates/typst-library/src/loading/toml.rs

@@ -1,7 +1,7 @@
-use ecow::{eco_format, EcoString};
-use typst_syntax::{is_newline, Spanned};
+use ecow::eco_format;
+use typst_syntax::Spanned;
 
-use crate::diag::{At, FileError, SourceResult};
+use crate::diag::{At, LoadError, LoadedWithin, ReportPos, SourceResult};
 use crate::engine::Engine;
 use crate::foundations::{func, scope, Str, Value};
 use crate::loading::{DataSource, Load, Readable};
@@ -32,11 +32,9 @@ pub fn toml(
     /// A [path]($syntax/#paths) to a TOML file or raw TOML bytes.
     source: Spanned<DataSource>,
 ) -> SourceResult<Value> {
-    let data = source.load(engine.world)?;
-    let raw = data.as_str().map_err(FileError::from).at(source.span)?;
-    ::toml::from_str(raw)
-        .map_err(|err| format_toml_error(err, raw))
-        .at(source.span)
+    let loaded = source.load(engine.world)?;
+    let raw = loaded.data.as_str().within(&loaded)?;
+    ::toml::from_str(raw).map_err(format_toml_error).within(&loaded)
 }
 
 #[scope]
@@ -71,15 +69,7 @@ impl toml {
 }
 
 /// Format the user-facing TOML error message.
-fn format_toml_error(error: ::toml::de::Error, raw: &str) -> EcoString {
-    if let Some(head) = error.span().and_then(|range| raw.get(..range.start)) {
-        let line = head.lines().count();
-        let column = 1 + head.chars().rev().take_while(|&c| !is_newline(c)).count();
-        eco_format!(
-            "failed to parse TOML ({} at line {line} column {column})",
-            error.message(),
-        )
-    } else {
-        eco_format!("failed to parse TOML ({})", error.message())
-    }
+fn format_toml_error(error: ::toml::de::Error) -> LoadError {
+    let pos = error.span().map(ReportPos::from).unwrap_or_default();
+    LoadError::new(pos, "failed to parse TOML", error.message())
 }

crates/typst-library/src/loading/xml.rs

@@ -1,8 +1,7 @@
-use ecow::EcoString;
 use roxmltree::ParsingOptions;
 use typst_syntax::Spanned;
 
-use crate::diag::{format_xml_like_error, At, FileError, SourceResult};
+use crate::diag::{format_xml_like_error, LoadError, LoadedWithin, SourceResult};
 use crate::engine::Engine;
 use crate::foundations::{dict, func, scope, Array, Dict, IntoValue, Str, Value};
 use crate::loading::{DataSource, Load, Readable};
@@ -61,14 +60,14 @@ pub fn xml(
     /// A [path]($syntax/#paths) to an XML file or raw XML bytes.
     source: Spanned<DataSource>,
 ) -> SourceResult<Value> {
-    let data = source.load(engine.world)?;
-    let text = data.as_str().map_err(FileError::from).at(source.span)?;
+    let loaded = source.load(engine.world)?;
+    let text = loaded.data.as_str().within(&loaded)?;
     let document = roxmltree::Document::parse_with_options(
         text,
         ParsingOptions { allow_dtd: true, ..Default::default() },
     )
     .map_err(format_xml_error)
-    .at(source.span)?;
+    .within(&loaded)?;
     Ok(convert_xml(document.root()))
 }
 
@@ -111,6 +110,6 @@ fn convert_xml(node: roxmltree::Node) -> Value {
 }
 
 /// Format the user-facing XML error message.
-fn format_xml_error(error: roxmltree::Error) -> EcoString {
+fn format_xml_error(error: roxmltree::Error) -> LoadError {
     format_xml_like_error("XML", error)
 }

crates/typst-library/src/loading/yaml.rs

@@ -1,7 +1,7 @@
 use ecow::eco_format;
 use typst_syntax::Spanned;
 
-use crate::diag::{At, SourceResult};
+use crate::diag::{At, LineCol, LoadError, LoadedWithin, ReportPos, SourceResult};
 use crate::engine::Engine;
 use crate::foundations::{func, scope, Str, Value};
 use crate::loading::{DataSource, Load, Readable};
@@ -44,10 +44,10 @@ pub fn yaml(
     /// A [path]($syntax/#paths) to a YAML file or raw YAML bytes.
     source: Spanned<DataSource>,
 ) -> SourceResult<Value> {
-    let data = source.load(engine.world)?;
-    serde_yaml::from_slice(data.as_slice())
-        .map_err(|err| eco_format!("failed to parse YAML ({err})"))
-        .at(source.span)
+    let loaded = source.load(engine.world)?;
+    serde_yaml::from_slice(loaded.data.as_slice())
+        .map_err(format_yaml_error)
+        .within(&loaded)
 }
 
 #[scope]
@@ -76,3 +76,16 @@ impl yaml {
             .at(span)
     }
 }
+
+/// Format the user-facing YAML error message.
+pub fn format_yaml_error(error: serde_yaml::Error) -> LoadError {
+    let pos = error
+        .location()
+        .map(|loc| {
+            let line_col = LineCol::one_based(loc.line(), loc.column());
+            let range = loc.index()..loc.index();
+            ReportPos::full(range, line_col)
+        })
+        .unwrap_or_default();
+    LoadError::new(pos, "failed to parse YAML", error)
+}

crates/typst-library/src/model/bibliography.rs

@@ -16,10 +16,13 @@ use hayagriva::{
 };
 use indexmap::IndexMap;
 use smallvec::{smallvec, SmallVec};
-use typst_syntax::{Span, Spanned};
+use typst_syntax::{Span, Spanned, SyntaxMode};
 use typst_utils::{Get, ManuallyHash, NonZeroExt, PicoStr};
 
-use crate::diag::{bail, error, At, FileError, HintedStrResult, SourceResult, StrResult};
+use crate::diag::{
+    bail, error, At, HintedStrResult, LoadError, LoadResult, LoadedWithin, ReportPos,
+    SourceResult, StrResult,
+};
 use crate::engine::{Engine, Sink};
 use crate::foundations::{
     elem, Bytes, CastInfo, Content, Derived, FromValue, IntoValue, Label, NativeElement,
@@ -31,12 +34,12 @@ use crate::layout::{
     BlockBody, BlockElem, Em, GridCell, GridChild, GridElem, GridItem, HElem, PadElem,
     Sides, Sizing, TrackSizings,
 };
-use crate::loading::{DataSource, Load};
+use crate::loading::{format_yaml_error, DataSource, Load, LoadSource, Loaded};
 use crate::model::{
     CitationForm, CiteGroup, Destination, FootnoteElem, HeadingElem, LinkElem, ParElem,
     Url,
 };
-use crate::routines::{EvalMode, Routines};
+use crate::routines::Routines;
 use crate::text::{
     FontStyle, Lang, LocalName, Region, Smallcaps, SubElem, SuperElem, TextElem,
     WeightDelta,
@@ -48,8 +51,8 @@ use crate::World;
 /// You can create a new bibliography by calling this function with a path
 /// to a bibliography file in either one of two formats:
 ///
-/// - A Hayagriva `.yml` file. Hayagriva is a new bibliography file format
-///   designed for use with Typst. Visit its
+/// - A Hayagriva `.yaml`/`.yml` file. Hayagriva is a new bibliography
+///   file format designed for use with Typst. Visit its
 ///   [documentation](https://github.com/typst/hayagriva/blob/main/docs/file-format.md)
 ///   for more details.
 /// - A BibLaTeX `.bib` file.
@@ -224,7 +227,15 @@ impl Show for Packed<BibliographyElem> {
         let references = works
             .references
             .as_ref()
-            .ok_or("CSL style is not suitable for bibliographies")
+            .ok_or_else(|| match self.style(styles).source {
+                CslSource::Named(style) => eco_format!(
+                    "CSL style \"{}\" is not suitable for bibliographies",
+                    style.display_name()
+                ),
+                CslSource::Normal(..) => {
+                    "CSL style is not suitable for bibliographies".into()
+                }
+            })
             .at(span)?;
 
         if references.iter().any(|(prefix, _)| prefix.is_some()) {
@@ -294,24 +305,21 @@ impl Bibliography {
         world: Tracked<dyn World + '_>,
         sources: Spanned<OneOrMultiple<DataSource>>,
     ) -> SourceResult<Derived<OneOrMultiple<DataSource>, Self>> {
-        let data = sources.load(world)?;
-        let bibliography = Self::decode(&sources.v, &data).at(sources.span)?;
+        let loaded = sources.load(world)?;
+        let bibliography = Self::decode(&loaded)?;
         Ok(Derived::new(sources.v, bibliography))
     }
 
     /// Decode a bibliography from loaded data sources.
     #[comemo::memoize]
     #[typst_macros::time(name = "load bibliography")]
-    fn decode(
-        sources: &OneOrMultiple<DataSource>,
-        data: &[Bytes],
-    ) -> StrResult<Bibliography> {
+    fn decode(data: &[Loaded]) -> SourceResult<Bibliography> {
         let mut map = IndexMap::new();
         let mut duplicates = Vec::<EcoString>::new();
 
         // We might have multiple bib/yaml files
-        for (source, data) in sources.0.iter().zip(data) {
-            let library = decode_library(source, data)?;
+        for d in data.iter() {
+            let library = decode_library(d)?;
             for entry in library {
                 match map.entry(Label::new(PicoStr::intern(entry.key()))) {
                     indexmap::map::Entry::Vacant(vacant) => {
@@ -325,7 +333,11 @@ impl Bibliography {
         }
 
         if !duplicates.is_empty() {
-            bail!("duplicate bibliography keys: {}", duplicates.join(", "));
+            // TODO: Store spans of entries for duplicate key error messages.
+            // Requires hayagriva entries to store their location, which should
+            // be fine, since they are 1kb anyway.
+            let span = data.first().unwrap().source.span;
+            bail!(span, "duplicate bibliography keys: {}", duplicates.join(", "));
         }
 
         Ok(Bibliography(Arc::new(ManuallyHash::new(map, typst_utils::hash128(data)))))
@@ -351,36 +363,47 @@ impl Debug for Bibliography {
 }
 
 /// Decode on library from one data source.
-fn decode_library(source: &DataSource, data: &Bytes) -> StrResult<Library> {
-    let src = data.as_str().map_err(FileError::from)?;
+fn decode_library(loaded: &Loaded) -> SourceResult<Library> {
+    let data = loaded.data.as_str().within(loaded)?;
 
-    if let DataSource::Path(path) = source {
+    if let LoadSource::Path(file_id) = loaded.source.v {
         // If we got a path, use the extension to determine whether it is
         // YAML or BibLaTeX.
-        let ext = Path::new(path.as_str())
+        let ext = file_id
+            .vpath()
+            .as_rooted_path()
             .extension()
             .and_then(OsStr::to_str)
             .unwrap_or_default();
 
         match ext.to_lowercase().as_str() {
-            "yml" | "yaml" => hayagriva::io::from_yaml_str(src)
-                .map_err(|err| eco_format!("failed to parse YAML ({err})")),
-            "bib" => hayagriva::io::from_biblatex_str(src)
-                .map_err(|errors| format_biblatex_error(src, Some(path), errors)),
-            _ => bail!("unknown bibliography format (must be .yml/.yaml or .bib)"),
+            "yml" | "yaml" => hayagriva::io::from_yaml_str(data)
+                .map_err(format_yaml_error)
+                .within(loaded),
+            "bib" => hayagriva::io::from_biblatex_str(data)
+                .map_err(format_biblatex_error)
+                .within(loaded),
+            _ => bail!(
+                loaded.source.span,
+                "unknown bibliography format (must be .yml/.yaml or .bib)"
+            ),
         }
     } else {
         // If we just got bytes, we need to guess. If it can be decoded as
         // hayagriva YAML, we'll use that.
-        let haya_err = match hayagriva::io::from_yaml_str(src) {
+        let haya_err = match hayagriva::io::from_yaml_str(data) {
             Ok(library) => return Ok(library),
             Err(err) => err,
         };
 
         // If it can be decoded as BibLaTeX, we use that isntead.
-        let bib_errs = match hayagriva::io::from_biblatex_str(src) {
-            Ok(library) => return Ok(library),
-            Err(err) => err,
+        let bib_errs = match hayagriva::io::from_biblatex_str(data) {
+            // If the file is almost valid yaml, but contains no `@` character
+            // it will be successfully parsed as an empty BibLaTeX library,
+            // since BibLaTeX does support arbitrary text outside of entries.
+            Ok(library) if !library.is_empty() => return Ok(library),
+            Ok(_) => None,
+            Err(err) => Some(err),
         };
 
         // If neither decoded correctly, check whether `:` or `{` appears
@@ -388,7 +411,7 @@ fn decode_library(source: &DataSource, data: &Bytes) -> StrResult<Library> {
         // and emit the more appropriate error.
         let mut yaml = 0;
         let mut biblatex = 0;
-        for c in src.chars() {
+        for c in data.chars() {
             match c {
                 ':' => yaml += 1,
                 '{' => biblatex += 1,
@@ -396,37 +419,33 @@ fn decode_library(source: &DataSource, data: &Bytes) -> StrResult<Library> {
             }
         }
 
-        if yaml > biblatex {
-            bail!("failed to parse YAML ({haya_err})")
-        } else {
-            Err(format_biblatex_error(src, None, bib_errs))
+        match bib_errs {
+            Some(bib_errs) if biblatex >= yaml => {
+                Err(format_biblatex_error(bib_errs)).within(loaded)
+            }
+            _ => Err(format_yaml_error(haya_err)).within(loaded),
         }
     }
 }
 
 /// Format a BibLaTeX loading error.
-fn format_biblatex_error(
-    src: &str,
-    path: Option<&str>,
-    errors: Vec<BibLaTeXError>,
-) -> EcoString {
-    let Some(error) = errors.first() else {
-        return match path {
-            Some(path) => eco_format!("failed to parse BibLaTeX file ({path})"),
-            None => eco_format!("failed to parse BibLaTeX"),
-        };
+fn format_biblatex_error(errors: Vec<BibLaTeXError>) -> LoadError {
+    // TODO: return multiple errors?
+    let Some(error) = errors.into_iter().next() else {
+        // TODO: can this even happen, should we just unwrap?
+        return LoadError::new(
+            ReportPos::None,
+            "failed to parse BibLaTeX",
+            "something went wrong",
+        );
     };
 
-    let (span, msg) = match error {
-        BibLaTeXError::Parse(error) => (&error.span, error.kind.to_string()),
-        BibLaTeXError::Type(error) => (&error.span, error.kind.to_string()),
+    let (range, msg) = match error {
+        BibLaTeXError::Parse(error) => (error.span, error.kind.to_string()),
+        BibLaTeXError::Type(error) => (error.span, error.kind.to_string()),
     };
 
-    let line = src.get(..span.start).unwrap_or_default().lines().count();
-    match path {
-        Some(path) => eco_format!("failed to parse BibLaTeX file ({path}:{line}: {msg})"),
-        None => eco_format!("failed to parse BibLaTeX ({line}: {msg})"),
-    }
+    LoadError::new(range, "failed to parse BibLaTeX", msg)
 }
 
 /// A loaded CSL style.
@@ -442,8 +461,8 @@ impl CslStyle {
         let style = match &source {
             CslSource::Named(style) => Self::from_archived(*style),
             CslSource::Normal(source) => {
-                let data = Spanned::new(source, span).load(world)?;
-                Self::from_data(data).at(span)?
+                let loaded = Spanned::new(source, span).load(world)?;
+                Self::from_data(&loaded.data).within(&loaded)?
             }
         };
         Ok(Derived::new(source, style))
@@ -464,16 +483,18 @@ impl CslStyle {
 
     /// Load a CSL style from file contents.
     #[comemo::memoize]
-    pub fn from_data(data: Bytes) -> StrResult<CslStyle> {
-        let text = data.as_str().map_err(FileError::from)?;
+    pub fn from_data(bytes: &Bytes) -> LoadResult<CslStyle> {
+        let text = bytes.as_str()?;
         citationberg::IndependentStyle::from_xml(text)
             .map(|style| {
                 Self(Arc::new(ManuallyHash::new(
                     style,
-                    typst_utils::hash128(&(TypeId::of::<Bytes>(), data)),
+                    typst_utils::hash128(&(TypeId::of::<Bytes>(), bytes)),
                 )))
             })
-            .map_err(|err| eco_format!("failed to load CSL style ({err})"))
+            .map_err(|err| {
+                LoadError::new(ReportPos::None, "failed to load CSL style", err)
+            })
     }
 
     /// Get the underlying independent style.
@@ -1003,7 +1024,7 @@ impl ElemRenderer<'_> {
             Sink::new().track_mut(),
             math,
             self.span,
-            EvalMode::Math,
+            SyntaxMode::Math,
             Scope::new(),
         )
         .map(Value::display)

crates/typst-library/src/model/table.rs

@@ -1,4 +1,4 @@
-use std::num::NonZeroUsize;
+use std::num::{NonZeroU32, NonZeroUsize};
 use std::sync::Arc;
 
 use typst_utils::NonZeroExt;
@@ -292,16 +292,61 @@ fn show_cellgrid_html(grid: CellGrid, styles: StyleChain) -> Content {
         elem(tag::tr, Content::sequence(row))
     };
 
+    // TODO(subfooters): similarly to headers, take consecutive footers from
+    // the end for 'tfoot'.
     let footer = grid.footer.map(|ft| {
-        let rows = rows.drain(ft.unwrap().start..);
+        let rows = rows.drain(ft.start..);
         elem(tag::tfoot, Content::sequence(rows.map(|row| tr(tag::td, row))))
     });
-    let header = grid.header.map(|hd| {
-        let rows = rows.drain(..hd.unwrap().end);
-        elem(tag::thead, Content::sequence(rows.map(|row| tr(tag::th, row))))
-    });
 
-    let mut body = Content::sequence(rows.into_iter().map(|row| tr(tag::td, row)));
+    // Store all consecutive headers at the start in 'thead'. All remaining
+    // headers are just 'th' rows across the table body.
+    let mut consecutive_header_end = 0;
+    let first_mid_table_header = grid
+        .headers
+        .iter()
+        .take_while(|hd| {
+            let is_consecutive = hd.range.start == consecutive_header_end;
+            consecutive_header_end = hd.range.end;
+
+            is_consecutive
+        })
+        .count();
+
+    let (y_offset, header) = if first_mid_table_header > 0 {
+        let removed_header_rows =
+            grid.headers.get(first_mid_table_header - 1).unwrap().range.end;
+        let rows = rows.drain(..removed_header_rows);
+
+        (
+            removed_header_rows,
+            Some(elem(tag::thead, Content::sequence(rows.map(|row| tr(tag::th, row))))),
+        )
+    } else {
+        (0, None)
+    };
+
+    // TODO: Consider improving accessibility properties of multi-level headers
+    // inside tables in the future, e.g. indicating which columns they are
+    // relative to and so on. See also:
+    // https://www.w3.org/WAI/tutorials/tables/multi-level/
+    let mut next_header = first_mid_table_header;
+    let mut body =
+        Content::sequence(rows.into_iter().enumerate().map(|(relative_y, row)| {
+            let y = relative_y + y_offset;
+            if let Some(current_header) =
+                grid.headers.get(next_header).filter(|h| h.range.contains(&y))
+            {
+                if y + 1 == current_header.range.end {
+                    next_header += 1;
+                }
+
+                tr(tag::th, row)
+            } else {
+                tr(tag::td, row)
+            }
+        }));
+
     if header.is_some() || footer.is_some() {
         body = elem(tag::tbody, body);
     }
@@ -492,6 +537,17 @@ pub struct TableHeader {
     #[default(true)]
     pub repeat: bool,
 
+    /// The level of the header. Must not be zero.
+    ///
+    /// This allows repeating multiple headers at once. Headers with different
+    /// levels can repeat together, as long as they have ascending levels.
+    ///
+    /// Notably, when a header with a lower level starts repeating, all higher
+    /// or equal level headers stop repeating (they are "replaced" by the new
+    /// header).
+    #[default(NonZeroU32::ONE)]
+    pub level: NonZeroU32,
+
     /// The cells and lines within the header.
     #[variadic]
     pub children: Vec<TableItem>,
@@ -770,7 +826,14 @@ impl Show for Packed<TableCell> {
 
 impl Default for Packed<TableCell> {
     fn default() -> Self {
-        Packed::new(TableCell::new(Content::default()))
+        Packed::new(
+            // Explicitly set colspan and rowspan to ensure they won't be
+            // overridden by set rules (default cells are created after
+            // colspans and rowspans are processed in the resolver)
+            TableCell::new(Content::default())
+                .with_colspan(NonZeroUsize::ONE)
+                .with_rowspan(NonZeroUsize::ONE),
+        )
     }
 }
 

crates/typst-library/src/pdf/embed.rs

@@ -59,7 +59,7 @@ pub struct EmbedElem {
     // We can't distinguish between the two at the moment.
     #[required]
     #[parse(
-        match args.find::<Bytes>()? {
+        match args.eat::<Bytes>()? {
             Some(data) => data,
             None => engine.world.file(id).at(span)?,
         }

crates/typst-library/src/routines.rs

@@ -4,7 +4,7 @@ use std::hash::{Hash, Hasher};
 use std::num::NonZeroUsize;
 
 use comemo::{Tracked, TrackedMut};
-use typst_syntax::Span;
+use typst_syntax::{Span, SyntaxMode};
 use typst_utils::LazyHash;
 
 use crate::diag::SourceResult;
@@ -58,7 +58,7 @@ routines! {
         sink: TrackedMut<Sink>,
         string: &str,
         span: Span,
-        mode: EvalMode,
+        mode: SyntaxMode,
         scope: Scope,
     ) -> SourceResult<Value>
 
@@ -312,17 +312,6 @@ routines! {
     ) -> SourceResult<Fragment>
 }
 
-/// In which mode to evaluate a string.
-#[derive(Debug, Copy, Clone, Eq, PartialEq, Hash, Cast)]
-pub enum EvalMode {
-    /// Evaluate as code, as after a hash.
-    Code,
-    /// Evaluate as markup, like in a Typst file.
-    Markup,
-    /// Evaluate as math, as in an equation.
-    Math,
-}
-
 /// Defines what kind of realization we are performing.
 pub enum RealizationKind<'a> {
     /// This the root realization for layout. Requires a mutable reference

crates/typst-library/src/text/font/book.rs

@@ -196,6 +196,8 @@ bitflags::bitflags! {
         const SERIF = 1 << 1;
         /// Font face has a MATH table
         const MATH = 1 << 2;
+        /// Font face has an fvar table
+        const VARIABLE = 1 << 3;
     }
 }
 
@@ -275,6 +277,7 @@ impl FontInfo {
         let mut flags = FontFlags::empty();
         flags.set(FontFlags::MONOSPACE, ttf.is_monospaced());
         flags.set(FontFlags::MATH, ttf.tables().math.is_some());
+        flags.set(FontFlags::VARIABLE, ttf.is_variable());
 
         // Determine whether this is a serif or sans-serif font.
         if let Some(panose) = ttf

crates/typst-library/src/text/font/mod.rs

@@ -106,13 +106,26 @@ impl Font {
     }
 
     /// Look up the horizontal advance width of a glyph.
-    pub fn advance(&self, glyph: u16) -> Option<Em> {
+    pub fn x_advance(&self, glyph: u16) -> Option<Em> {
         self.0
             .ttf
             .glyph_hor_advance(GlyphId(glyph))
             .map(|units| self.to_em(units))
     }
 
+    /// Look up the vertical advance width of a glyph.
+    pub fn y_advance(&self, glyph: u16) -> Option<Em> {
+        self.0
+            .ttf
+            .glyph_ver_advance(GlyphId(glyph))
+            .map(|units| self.to_em(units))
+    }
+
+    /// Look up the width of a space.
+    pub fn space_width(&self) -> Option<Em> {
+        self.0.ttf.glyph_index(' ').and_then(|id| self.x_advance(id.0))
+    }
+
     /// Lookup a name by id.
     pub fn find_name(&self, id: u16) -> Option<String> {
         find_name(&self.0.ttf, id)

crates/typst-library/src/text/item.rs

@@ -35,6 +35,11 @@ impl TextItem {
     pub fn width(&self) -> Abs {
         self.glyphs.iter().map(|g| g.x_advance).sum::<Em>().at(self.size)
     }
+
+    /// The height of the text run.
+    pub fn height(&self) -> Abs {
+        self.glyphs.iter().map(|g| g.y_advance).sum::<Em>().at(self.size)
+    }
 }
 
 impl Debug for TextItem {
@@ -54,6 +59,10 @@ pub struct Glyph {
     pub x_advance: Em,
     /// The horizontal offset of the glyph.
     pub x_offset: Em,
+    /// The advance height (Y-up) of the glyph.
+    pub y_advance: Em,
+    /// The vertical offset (Y-up) of the glyph.
+    pub y_offset: Em,
     /// The range of the glyph in its item's text. The range's length may
     /// be more than one due to multi-byte UTF-8 encoding or ligatures.
     pub range: Range<u16>,
@@ -115,4 +124,13 @@ impl<'a> TextItemView<'a> {
             .sum::<Em>()
             .at(self.item.size)
     }
+
+    /// The total height of this text slice
+    pub fn height(&self) -> Abs {
+        self.glyphs()
+            .iter()
+            .map(|g| g.y_advance)
+            .sum::<Em>()
+            .at(self.item.size)
+    }
 }

crates/typst-library/src/text/lang.rs

@@ -14,13 +14,14 @@ macro_rules! translation {
     };
 }
 
-const TRANSLATIONS: [(&str, &str); 41] = [
+const TRANSLATIONS: &[(&str, &str)] = &[
     translation!("ar"),
     translation!("bg"),
     translation!("ca"),
     translation!("cs"),
     translation!("da"),
     translation!("de"),
+    translation!("el"),
     translation!("en"),
     translation!("es"),
     translation!("et"),
@@ -28,7 +29,6 @@ const TRANSLATIONS: [(&str, &str); 41] = [
     translation!("fi"),
     translation!("fr"),
     translation!("gl"),
-    translation!("el"),
     translation!("he"),
     translation!("hr"),
     translation!("hu"),
@@ -42,8 +42,8 @@ const TRANSLATIONS: [(&str, &str); 41] = [
     translation!("nl"),
     translation!("nn"),
     translation!("pl"),
-    translation!("pt-PT"),
     translation!("pt"),
+    translation!("pt-PT"),
     translation!("ro"),
     translation!("ru"),
     translation!("sl"),
@@ -54,8 +54,8 @@ const TRANSLATIONS: [(&str, &str); 41] = [
     translation!("tr"),
     translation!("uk"),
     translation!("vi"),
-    translation!("zh-TW"),
     translation!("zh"),
+    translation!("zh-TW"),
 ];
 
 /// An identifier for a natural language.
@@ -313,14 +313,74 @@ fn lang_str(lang: Lang, region: Option<Region>) -> EcoString {
 
 #[cfg(test)]
 mod tests {
+    use std::collections::HashSet;
+    use std::path::PathBuf;
+
     use typst_utils::option_eq;
 
     use super::*;
 
+    fn translation_files_iter() -> impl Iterator<Item = PathBuf> {
+        std::fs::read_dir("translations")
+            .unwrap()
+            .map(|e| e.unwrap().path())
+            .filter(|e| e.is_file() && e.extension().is_some_and(|e| e == "txt"))
+    }
+
     #[test]
     fn test_region_option_eq() {
         let region = Some(Region([b'U', b'S']));
         assert!(option_eq(region, "US"));
         assert!(!option_eq(region, "AB"));
     }
+
+    #[test]
+    fn test_all_translations_included() {
+        let defined_keys =
+            HashSet::<&str>::from_iter(TRANSLATIONS.iter().map(|(lang, _)| *lang));
+        let mut checked = 0;
+        for file in translation_files_iter() {
+            assert!(
+                defined_keys.contains(
+                    file.file_stem()
+                        .expect("translation file should have basename")
+                        .to_str()
+                        .expect("translation file name should be utf-8 encoded")
+                ),
+                "translation from {:?} should be registered in TRANSLATIONS in {}",
+                file.file_name().unwrap(),
+                file!(),
+            );
+            checked += 1;
+        }
+        assert_eq!(TRANSLATIONS.len(), checked);
+    }
+
+    #[test]
+    fn test_all_translation_files_formatted() {
+        for file in translation_files_iter() {
+            let content = std::fs::read_to_string(&file)
+                .expect("translation file should be in utf-8 encoding");
+            let filename = file.file_name().unwrap();
+            assert!(
+                content.ends_with('\n'),
+                "translation file {filename:?} should end with linebreak",
+            );
+            for line in content.lines() {
+                assert_eq!(
+                    line.trim(),
+                    line,
+                    "line {line:?} in {filename:?} should not have extra whitespaces"
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn test_translations_sorted() {
+        assert!(
+            TRANSLATIONS.is_sorted_by_key(|(lang, _)| lang),
+            "TRANSLATIONS should be sorted"
+        );
+    }
 }

crates/typst-library/src/text/mod.rs

@@ -30,6 +30,7 @@ pub use self::space::*;
 
 use std::fmt::{self, Debug, Formatter};
 use std::hash::Hash;
+use std::str::FromStr;
 use std::sync::LazyLock;
 
 use ecow::{eco_format, EcoString};
@@ -1283,6 +1284,12 @@ pub fn features(styles: StyleChain) -> Vec<Feature> {
         feat(b"frac", 1);
     }
 
+    match EquationElem::size_in(styles) {
+        MathSize::Script => feat(b"ssty", 1),
+        MathSize::ScriptScript => feat(b"ssty", 2),
+        _ => {}
+    }
+
     for (tag, value) in TextElem::features_in(styles).0 {
         tags.push(Feature::new(tag, value, ..))
     }
@@ -1290,6 +1297,17 @@ pub fn features(styles: StyleChain) -> Vec<Feature> {
     tags
 }
 
+/// Process the language and region of a style chain into a
+/// rustybuzz-compatible BCP 47 language.
+pub fn language(styles: StyleChain) -> rustybuzz::Language {
+    let mut bcp: EcoString = TextElem::lang_in(styles).as_str().into();
+    if let Some(region) = TextElem::region_in(styles) {
+        bcp.push('-');
+        bcp.push_str(region.as_str());
+    }
+    rustybuzz::Language::from_str(&bcp).unwrap()
+}
+
 /// A toggle that turns on and off alternatingly if folded.
 #[derive(Debug, Copy, Clone, Eq, PartialEq, Hash)]
 pub struct ItalicToggle(pub bool);
@@ -1394,12 +1412,24 @@ pub fn is_default_ignorable(c: char) -> bool {
 fn check_font_list(engine: &mut Engine, list: &Spanned<FontList>) {
     let book = engine.world.book();
     for family in &list.v {
-        if !book.contains_family(family.as_str()) {
-            engine.sink.warn(warning!(
+        match book.select_family(family.as_str()).next() {
+            Some(index) => {
+                if book
+                    .info(index)
+                    .is_some_and(|x| x.flags.contains(FontFlags::VARIABLE))
+                {
+                    engine.sink.warn(warning!(
+                        list.span,
+                        "variable fonts are not currently supported and may render incorrectly";
+                        hint: "try installing a static version of \"{}\" instead", family.as_str()
+                    ))
+                }
+            }
+            None => engine.sink.warn(warning!(
                 list.span,
                 "unknown font family: {}",
                 family.as_str(),
-            ));
+            )),
         }
     }
 }

crates/typst-library/src/text/raw.rs

@@ -3,15 +3,17 @@ use std::ops::Range;
 use std::sync::{Arc, LazyLock};
 
 use comemo::Tracked;
-use ecow::{eco_format, EcoString, EcoVec};
-use syntect::highlighting as synt;
-use syntect::parsing::{SyntaxDefinition, SyntaxSet, SyntaxSetBuilder};
+use ecow::{EcoString, EcoVec};
+use syntect::highlighting::{self as synt};
+use syntect::parsing::{ParseSyntaxError, SyntaxDefinition, SyntaxSet, SyntaxSetBuilder};
 use typst_syntax::{split_newlines, LinkedNode, Span, Spanned};
 use typst_utils::ManuallyHash;
 use unicode_segmentation::UnicodeSegmentation;
 
 use super::Lang;
-use crate::diag::{At, FileError, SourceResult, StrResult};
+use crate::diag::{
+    LineCol, LoadError, LoadResult, LoadedWithin, ReportPos, SourceResult,
+};
 use crate::engine::Engine;
 use crate::foundations::{
     cast, elem, scope, Bytes, Content, Derived, NativeElement, OneOrMultiple, Packed,
@@ -539,40 +541,29 @@ impl RawSyntax {
         world: Tracked<dyn World + '_>,
         sources: Spanned<OneOrMultiple<DataSource>>,
     ) -> SourceResult<Derived<OneOrMultiple<DataSource>, Vec<RawSyntax>>> {
-        let data = sources.load(world)?;
-        let list = sources
-            .v
-            .0
+        let loaded = sources.load(world)?;
+        let list = loaded
             .iter()
-            .zip(&data)
-            .map(|(source, data)| Self::decode(source, data))
-            .collect::<StrResult<_>>()
-            .at(sources.span)?;
+            .map(|data| Self::decode(&data.data).within(data))
+            .collect::<SourceResult<_>>()?;
         Ok(Derived::new(sources.v, list))
     }
 
     /// Decode a syntax from a loaded source.
     #[comemo::memoize]
     #[typst_macros::time(name = "load syntaxes")]
-    fn decode(source: &DataSource, data: &Bytes) -> StrResult<RawSyntax> {
-        let src = data.as_str().map_err(FileError::from)?;
-        let syntax = SyntaxDefinition::load_from_str(src, false, None).map_err(
-            |err| match source {
-                DataSource::Path(path) => {
-                    eco_format!("failed to parse syntax file `{path}` ({err})")
-                }
-                DataSource::Bytes(_) => {
-                    eco_format!("failed to parse syntax ({err})")
-                }
-            },
-        )?;
+    fn decode(bytes: &Bytes) -> LoadResult<RawSyntax> {
+        let str = bytes.as_str()?;
+
+        let syntax = SyntaxDefinition::load_from_str(str, false, None)
+            .map_err(format_syntax_error)?;
 
         let mut builder = SyntaxSetBuilder::new();
         builder.add(syntax);
 
         Ok(RawSyntax(Arc::new(ManuallyHash::new(
             builder.build(),
-            typst_utils::hash128(data),
+            typst_utils::hash128(bytes),
         ))))
     }
 
@@ -582,6 +573,24 @@ impl RawSyntax {
     }
 }
 
+fn format_syntax_error(error: ParseSyntaxError) -> LoadError {
+    let pos = syntax_error_pos(&error);
+    LoadError::new(pos, "failed to parse syntax", error)
+}
+
+fn syntax_error_pos(error: &ParseSyntaxError) -> ReportPos {
+    match error {
+        ParseSyntaxError::InvalidYaml(scan_error) => {
+            let m = scan_error.marker();
+            ReportPos::full(
+                m.index()..m.index(),
+                LineCol::one_based(m.line(), m.col() + 1),
+            )
+        }
+        _ => ReportPos::None,
+    }
+}
+
 /// A loaded syntect theme.
 #[derive(Debug, Clone, PartialEq, Hash)]
 pub struct RawTheme(Arc<ManuallyHash<synt::Theme>>);
@@ -592,18 +601,18 @@ impl RawTheme {
         world: Tracked<dyn World + '_>,
         source: Spanned<DataSource>,
     ) -> SourceResult<Derived<DataSource, Self>> {
-        let data = source.load(world)?;
-        let theme = Self::decode(&data).at(source.span)?;
+        let loaded = source.load(world)?;
+        let theme = Self::decode(&loaded.data).within(&loaded)?;
         Ok(Derived::new(source.v, theme))
     }
 
     /// Decode a theme from bytes.
     #[comemo::memoize]
-    fn decode(data: &Bytes) -> StrResult<RawTheme> {
-        let mut cursor = std::io::Cursor::new(data.as_slice());
-        let theme = synt::ThemeSet::load_from_reader(&mut cursor)
-            .map_err(|err| eco_format!("failed to parse theme ({err})"))?;
-        Ok(RawTheme(Arc::new(ManuallyHash::new(theme, typst_utils::hash128(data)))))
+    fn decode(bytes: &Bytes) -> LoadResult<RawTheme> {
+        let mut cursor = std::io::Cursor::new(bytes.as_slice());
+        let theme =
+            synt::ThemeSet::load_from_reader(&mut cursor).map_err(format_theme_error)?;
+        Ok(RawTheme(Arc::new(ManuallyHash::new(theme, typst_utils::hash128(bytes)))))
     }
 
     /// Get the underlying syntect theme.
@@ -612,6 +621,14 @@ impl RawTheme {
     }
 }
 
+fn format_theme_error(error: syntect::LoadingError) -> LoadError {
+    let pos = match &error {
+        syntect::LoadingError::ParseSyntax(err, _) => syntax_error_pos(err),
+        _ => ReportPos::None,
+    };
+    LoadError::new(pos, "failed to parse theme", error)
+}
+
 /// A highlighted line of raw text.
 ///
 /// This is a helper element that is synthesized by [`raw`] elements.

crates/typst-library/src/text/shift.rs

@@ -2,7 +2,10 @@ use ecow::EcoString;
 
 use crate::diag::SourceResult;
 use crate::engine::Engine;
-use crate::foundations::{elem, Content, Packed, SequenceElem, Show, StyleChain};
+use crate::foundations::{
+    elem, Content, NativeElement, Packed, SequenceElem, Show, StyleChain, TargetElem,
+};
+use crate::html::{tag, HtmlElem};
 use crate::layout::{Em, Length};
 use crate::text::{variant, SpaceElem, TextElem, TextSize};
 use crate::World;
@@ -52,6 +55,13 @@ impl Show for Packed<SubElem> {
     fn show(&self, engine: &mut Engine, styles: StyleChain) -> SourceResult<Content> {
         let body = self.body.clone();
 
+        if TargetElem::target_in(styles).is_html() {
+            return Ok(HtmlElem::new(tag::sub)
+                .with_body(Some(body))
+                .pack()
+                .spanned(self.span()));
+        }
+
         if self.typographic(styles) {
             if let Some(text) = convert_script(&body, true) {
                 if is_shapable(engine, &text, styles) {
@@ -111,6 +121,13 @@ impl Show for Packed<SuperElem> {
     fn show(&self, engine: &mut Engine, styles: StyleChain) -> SourceResult<Content> {
         let body = self.body.clone();
 
+        if TargetElem::target_in(styles).is_html() {
+            return Ok(HtmlElem::new(tag::sup)
+                .with_body(Some(body))
+                .pack()
+                .spanned(self.span()));
+        }
+
         if self.typographic(styles) {
             if let Some(text) = convert_script(&body, false) {
                 if is_shapable(engine, &text, styles) {

crates/typst-library/src/visualize/gradient.rs

@@ -549,7 +549,7 @@ impl Gradient {
     }
 
     /// Repeats this gradient a given number of times, optionally mirroring it
-    /// at each repetition.
+    /// at every second repetition.
     ///
     /// ```example
     /// #circle(
@@ -564,7 +564,17 @@ impl Gradient {
         &self,
         /// The number of times to repeat the gradient.
         repetitions: Spanned<usize>,
-        /// Whether to mirror the gradient at each repetition.
+        /// Whether to mirror the gradient at every second repetition, i.e.,
+        /// the first instance (and all odd ones) stays unchanged.
+        ///
+        /// ```example
+        /// #circle(
+        ///   radius: 40pt,
+        ///   fill: gradient
+        ///     .conic(green, black)
+        ///     .repeat(2, mirror: true)
+        /// )
+        /// ```
         #[named]
         #[default(false)]
         mirror: bool,

crates/typst-library/src/visualize/image/mod.rs

@@ -22,7 +22,7 @@ use crate::foundations::{
     Smart, StyleChain,
 };
 use crate::layout::{BlockElem, Length, Rel, Sizing};
-use crate::loading::{DataSource, Load, Readable};
+use crate::loading::{DataSource, Load, LoadSource, Loaded, Readable};
 use crate::model::Figurable;
 use crate::text::LocalName;
 
@@ -65,10 +65,10 @@ pub struct ImageElem {
     #[required]
     #[parse(
         let source = args.expect::<Spanned<DataSource>>("source")?;
-        let data = source.load(engine.world)?;
-        Derived::new(source.v, data)
+        let loaded = source.load(engine.world)?;
+        Derived::new(source.v, loaded)
     )]
-    pub source: Derived<DataSource, Bytes>,
+    pub source: Derived<DataSource, Loaded>,
 
     /// The image's format.
     ///
@@ -154,8 +154,8 @@ pub struct ImageElem {
     /// to `{auto}`, Typst will try to extract an ICC profile from the image.
     #[parse(match args.named::<Spanned<Smart<DataSource>>>("icc")? {
         Some(Spanned { v: Smart::Custom(source), span }) => Some(Smart::Custom({
-            let data = Spanned::new(&source, span).load(engine.world)?;
-            Derived::new(source, data)
+            let loaded = Spanned::new(&source, span).load(engine.world)?;
+            Derived::new(source, loaded.data)
         })),
         Some(Spanned { v: Smart::Auto, .. }) => Some(Smart::Auto),
         None => None,
@@ -173,7 +173,7 @@ impl ImageElem {
     pub fn decode(
         span: Span,
         /// The data to decode as an image. Can be a string for SVGs.
-        data: Readable,
+        data: Spanned<Readable>,
         /// The image's format. Detected automatically by default.
         #[named]
         format: Option<Smart<ImageFormat>>,
@@ -193,8 +193,10 @@ impl ImageElem {
         #[named]
         scaling: Option<Smart<ImageScaling>>,
     ) -> StrResult<Content> {
-        let bytes = data.into_bytes();
-        let source = Derived::new(DataSource::Bytes(bytes.clone()), bytes);
+        let bytes = data.v.into_bytes();
+        let loaded =
+            Loaded::new(Spanned::new(LoadSource::Bytes, data.span), bytes.clone());
+        let source = Derived::new(DataSource::Bytes(bytes), loaded);
         let mut elem = ImageElem::new(source);
         if let Some(format) = format {
             elem.push_format(format);

crates/typst-library/src/visualize/image/svg.rs

@@ -3,10 +3,9 @@ use std::hash::{Hash, Hasher};
 use std::sync::{Arc, Mutex};
 
 use comemo::Tracked;
-use ecow::EcoString;
 use siphasher::sip128::{Hasher128, SipHasher13};
 
-use crate::diag::{format_xml_like_error, StrResult};
+use crate::diag::{format_xml_like_error, LoadError, LoadResult, ReportPos};
 use crate::foundations::Bytes;
 use crate::layout::Axes;
 use crate::text::{
@@ -30,7 +29,7 @@ impl SvgImage {
     /// Decode an SVG image without fonts.
     #[comemo::memoize]
     #[typst_macros::time(name = "load svg")]
-    pub fn new(data: Bytes) -> StrResult<SvgImage> {
+    pub fn new(data: Bytes) -> LoadResult<SvgImage> {
         let tree =
             usvg::Tree::from_data(&data, &base_options()).map_err(format_usvg_error)?;
         Ok(Self(Arc::new(Repr { data, size: tree_size(&tree), font_hash: 0, tree })))
@@ -43,7 +42,7 @@ impl SvgImage {
         data: Bytes,
         world: Tracked<dyn World + '_>,
         families: &[&str],
-    ) -> StrResult<SvgImage> {
+    ) -> LoadResult<SvgImage> {
         let book = world.book();
         let resolver = Mutex::new(FontResolver::new(world, book, families));
         let tree = usvg::Tree::from_data(
@@ -125,16 +124,15 @@ fn tree_size(tree: &usvg::Tree) -> Axes<f64> {
 }
 
 /// Format the user-facing SVG decoding error message.
-fn format_usvg_error(error: usvg::Error) -> EcoString {
-    match error {
-        usvg::Error::NotAnUtf8Str => "file is not valid utf-8".into(),
-        usvg::Error::MalformedGZip => "file is not compressed correctly".into(),
-        usvg::Error::ElementsLimitReached => "file is too large".into(),
-        usvg::Error::InvalidSize => {
-            "failed to parse SVG (width, height, or viewbox is invalid)".into()
-        }
-        usvg::Error::ParsingFailed(error) => format_xml_like_error("SVG", error),
-    }
+fn format_usvg_error(error: usvg::Error) -> LoadError {
+    let error = match error {
+        usvg::Error::NotAnUtf8Str => "file is not valid utf-8",
+        usvg::Error::MalformedGZip => "file is not compressed correctly",
+        usvg::Error::ElementsLimitReached => "file is too large",
+        usvg::Error::InvalidSize => "width, height, or viewbox is invalid",
+        usvg::Error::ParsingFailed(error) => return format_xml_like_error("SVG", error),
+    };
+    LoadError::new(ReportPos::None, "failed to parse SVG", error)
 }
 
 /// Provides Typst's fonts to usvg.

crates/typst-library/translations/ar.txt

@@ -5,4 +5,4 @@ bibliography = المراجع
 heading = الفصل
 outline = المحتويات
 raw = قائمة
-page = صفحة
+page = صفحة

crates/typst-library/translations/bg.txt

@@ -5,4 +5,4 @@ bibliography = Библиография
 heading = Раздел
 outline = Съдържание
 raw = Приложение
-page = стр.
+page = стр.

crates/typst-library/translations/ca.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografia
 heading = Secció
 outline = Índex
 raw = Llistat
-page = pàgina
+page = pàgina

crates/typst-library/translations/cs.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografie
 heading = Kapitola
 outline = Obsah
 raw = Výpis
-page = strana
+page = strana

crates/typst-library/translations/da.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografi
 heading = Afsnit
 outline = Indhold
 raw = Liste
-page = side
+page = side

crates/typst-library/translations/de.txt

@@ -5,4 +5,4 @@ bibliography = Bibliographie
 heading = Abschnitt
 outline = Inhaltsverzeichnis
 raw = Listing
-page = Seite
+page = Seite

crates/typst-library/translations/el.txt

@@ -4,4 +4,4 @@ equation = Εξίσωση
 bibliography = Βιβλιογραφία
 heading = Κεφάλαιο
 outline = Περιεχόμενα
-raw = Παράθεση
+raw = Παράθεση

crates/typst-library/translations/en.txt

@@ -5,4 +5,4 @@ bibliography = Bibliography
 heading = Section
 outline = Contents
 raw = Listing
-page = page
+page = page

crates/typst-library/translations/es.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografía
 heading = Sección
 outline = Índice
 raw = Listado
-page = página
+page = página

crates/typst-library/translations/et.txt

@@ -5,4 +5,4 @@ bibliography = Viited
 heading = Peatükk
 outline = Sisukord
 raw = List
-page = lk.
+page = lk.

crates/typst-library/translations/fi.txt

@@ -5,4 +5,4 @@ bibliography = Viitteet
 heading = Osio
 outline = Sisällys
 raw = Esimerkki
-page = sivu
+page = sivu

crates/typst-library/translations/fr.txt

@@ -5,4 +5,4 @@ bibliography = Bibliographie
 heading = Chapitre
 outline = Table des matières
 raw = Liste
-page = page
+page = page

crates/typst-library/translations/gl.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografía
 heading = Sección
 outline = Índice
 raw = Listado
-page = páxina
+page = páxina

crates/typst-library/translations/he.txt

@@ -5,4 +5,4 @@ bibliography = רשימת מקורות
 heading = חלק
 outline = תוכן עניינים
 raw = קטע מקור
-page = עמוד
+page = עמוד

crates/typst-library/translations/hu.txt

@@ -4,5 +4,5 @@ equation = Egyenlet
 bibliography = Irodalomjegyzék
 heading = Fejezet
 outline = Tartalomjegyzék
-# raw = 
-page = oldal
+# raw =
+page = oldal

crates/typst-library/translations/is.txt

@@ -5,4 +5,4 @@ bibliography = Heimildaskrá
 heading = Kafli
 outline = Efnisyfirlit
 raw = Sýnishorn
-page = blaðsíða
+page = blaðsíða

crates/typst-library/translations/it.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografia
 heading = Sezione
 outline = Indice
 raw = Codice
-page = pag.
+page = pag.

crates/typst-library/translations/ja.txt

@@ -5,4 +5,4 @@ bibliography = 参考文献
 heading = 節
 outline = 目次
 raw = リスト
-page = ページ
+page = ページ

crates/typst-library/translations/la.txt

@@ -5,4 +5,4 @@ bibliography = Conspectus librorum
 heading = Caput
 outline = Index capitum
 raw = Exemplum
-page = charta
+page = charta

crates/typst-library/translations/nb.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografi
 heading = Kapittel
 outline = Innhold
 raw = Utskrift
-page = side
+page = side

crates/typst-library/translations/nl.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografie
 heading = Hoofdstuk
 outline = Inhoudsopgave
 raw = Listing
-page = pagina
+page = pagina

crates/typst-library/translations/nn.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografi
 heading = Kapittel
 outline = Innhald
 raw = Utskrift
-page = side
+page = side

crates/typst-library/translations/pl.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografia
 heading = Sekcja
 outline = Spis treści
 raw = Program
-page = strona
+page = strona

crates/typst-library/translations/pt-PT.txt

@@ -1,8 +1,8 @@
-# figure = 
-# table = 
-# equation = 
-# bibliography = 
+# figure =
+# table =
+# equation =
+# bibliography =
 heading = Secção
 outline = Índice
-# raw = 
-page = página
+# raw =
+page = página

crates/typst-library/translations/pt.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografia
 heading = Seção
 outline = Sumário
 raw = Listagem
-page = página
+page = página

crates/typst-library/translations/ro.txt

@@ -6,4 +6,4 @@ heading = Secțiunea
 outline = Cuprins
 # may be wrong
 raw = Listă
-page = pagina
+page = pagina

crates/typst-library/translations/ru.txt

@@ -5,4 +5,4 @@ bibliography = Библиография
 heading = Раздел
 outline = Содержание
 raw = Листинг
-page = с.
+page = с.

crates/typst-library/translations/sl.txt

@@ -5,4 +5,4 @@ bibliography = Literatura
 heading = Poglavje
 outline = Kazalo
 raw = Program
-page = stran
+page = stran

crates/typst-library/translations/sq.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografi
 heading = Kapitull
 outline = Përmbajtja
 raw = List
-page = faqe
+page = faqe

crates/typst-library/translations/sr.txt

@@ -5,4 +5,4 @@ bibliography = Литература
 heading = Поглавље
 outline = Садржај
 raw = Програм
-page = страна
+page = страна

crates/typst-library/translations/sv.txt

@@ -5,4 +5,4 @@ bibliography = Bibliografi
 heading = Kapitel
 outline = Innehåll
 raw = Listing
-page = sida
+page = sida

crates/typst-library/translations/tl.txt

@@ -5,4 +5,4 @@ bibliography = Bibliograpiya
 heading = Seksyon
 outline = Talaan ng mga Nilalaman
 raw = Listahan
-# page = 
+# page =

crates/typst-library/translations/tr.txt

@@ -5,4 +5,4 @@ bibliography = Kaynakça
 heading = Bölüm
 outline = İçindekiler
 raw = Liste
-page = sayfa
+page = sayfa

crates/typst-library/translations/uk.txt

@@ -5,4 +5,4 @@ bibliography = Бібліографія
 heading = Розділ
 outline = Зміст
 raw = Лістинг
-page = c.
+page = c.

crates/typst-library/translations/vi.txt

@@ -6,4 +6,4 @@ heading = Phần
 outline = Mục lục
 # may be wrong
 raw = Chương trình
-page = trang
+page = trang

crates/typst-library/translations/zh-TW.txt

@@ -1,8 +1,8 @@
 figure = 圖
-# table = 
+# table =
 equation = 式
 bibliography = 書目
 heading = 小節
 outline = 目錄
 raw = 程式
-# page = 
+# page =