address review: bound merge expansion work; measure a normalized copy of an unnormalized table

crates/engine/src/cmd.rs

@@ -433,6 +433,17 @@ pub(crate) fn shape_of(s: &Session, id: ShapeId) -> Result<deckcraft_model::Shap
 pub fn xfrm_of(doc: &deckcraft_model::Presentation, sel: &crate::Selection, shape: &deckcraft_model::Shape) -> Xfrm {
     let mut x = shape.xfrm.unwrap_or_else(|| ctx_xfrm(doc, sel, shape));
     if let deckcraft_model::ShapeKind::Table(t) = &shape.kind {
+        // A loaded table may not be normalized yet: measure the bounded grid an edit will
+        // leave, not every cell the file stored.
+        let normalized;
+        let t = if t.is_normalized() {
+            t
+        } else {
+            let mut n = t.clone();
+            n.normalize();
+            normalized = n;
+            &normalized
+        };
         let drawn: f64 = table_drawn_heights(doc, sel, t).iter().sum();
         if drawn.is_finite() && drawn > x.h {
             x.h = drawn;

crates/engine/src/cmd/table.rs

@@ -285,6 +285,8 @@ fn with_table(s: &mut Session, p: &Value, cmd: &str, f: impl Fn(&mut Table) -> R
     };
     let r = edit_shapes(s, &json!({"id": id}), cmd, |sh| match &mut sh.kind {
         ShapeKind::Table(t) => {
+            // Edits walk the grid: bound it first (a loaded table need not be normalized yet).
+            t.normalize();
             f(t)?;
             t.normalize();
             let (w, h) = (t.width(), t.height());

crates/engine/src/fonts.rs

@@ -1,8 +1,6 @@
 //! Fonts a presentation uses that this machine doesn't have, so opening it can say so (its text
 //! draws in a stand-in; see `FontDb::document_face`).
 
-use std::collections::HashSet;
-
 use deckcraft_fonts::FontDb;
 use deckcraft_model::resolve::{self, Ctx};
 use deckcraft_model::text::{Bullet, TextBody};
@@ -14,13 +12,8 @@ use deckcraft_model::{Presentation, Shape, ShapeKind};
 /// font to load and are left for the round trip.
 pub fn install_embedded_fonts(p: &Presentation) {
     let db = FontDb::global();
-    // Faces can name one media item many times; each is copied and enumerated once.
-    let mut seen = HashSet::new();
     for font in &p.embedded_fonts {
         for id in font.faces().into_iter().flatten() {
-            if !seen.insert(id) {
-                continue;
-            }
             let Some(item) = p.media(id) else { continue };
             let Some(sfnt) = deckcraft_fonts::embedded_sfnt(&item.data) else { continue };
             db.add_font(sfnt.to_vec());
@@ -34,8 +27,12 @@ pub fn install_embedded_fonts(p: &Presentation) {
 /// and can't be installed, so they aren't reported.
 pub fn missing_fonts(p: &Presentation) -> Vec<String> {
     let db = FontDb::global();
-    // used_fonts is already free of case-insensitive duplicates.
-    let mut out: Vec<String> = used_fonts(p).into_iter().filter(|f| !f.starts_with('.') && !db.has_family(f)).collect();
+    let mut out: Vec<String> = Vec::new();
+    for family in used_fonts(p) {
+        if !family.starts_with('.') && !db.has_family(&family) && !out.iter().any(|f| f.eq_ignore_ascii_case(&family)) {
+            out.push(family);
+        }
+    }
     out.sort_by_key(|f| f.to_lowercase());
     out
 }
@@ -43,10 +40,9 @@ pub fn missing_fonts(p: &Presentation) -> Vec<String> {
 /// The families the slides' text runs and bullets are drawn in.
 pub fn used_fonts(p: &Presentation) -> Vec<String> {
     let mut out: Vec<String> = Vec::new();
-    let mut seen: HashSet<String> = HashSet::new();
     let mut add = |f: String| {
         let f = f.trim().to_string();
-        if !f.is_empty() && seen.insert(f.to_ascii_lowercase()) {
+        if !f.is_empty() && !out.iter().any(|o| o.eq_ignore_ascii_case(&f)) {
             out.push(f);
         }
     };

crates/engine/src/tests.rs

@@ -6458,24 +6458,3 @@ fn master_body_placeholder_format_keeps_per_level_formatting() {
     assert_eq!((level(&s, 0).run.size, level(&s, 1).run.size), (Some(28.0), Some(24.0)));
     assert_eq!(shown(&s), vec![Some(28.0), Some(24.0)]);
 }
-
-#[test]
-fn many_distinct_fonts_are_listed_once_each_without_a_quadratic_scan() {
-    // 30,000 runs in 15,000 families, each written twice in different case.
-    let mut p = deckcraft_model::Presentation::default();
-    let slide = std::sync::Arc::make_mut(&mut p.slides[0]);
-    let mut body = deckcraft_model::text::TextBody::from_text("x");
-    let run = body.paragraphs[0].runs[0].clone();
-    body.paragraphs[0].runs = (0..30_000)
-        .map(|i| {
-            let mut r = run.clone();
-            let name = format!("Qqq Font {}", i / 2);
-            r.props.font = Some(if i % 2 == 0 { name } else { name.to_uppercase() });
-            r
-        })
-        .collect();
-    slide.shapes[0].text = Some(body);
-    let used = crate::fonts::used_fonts(&p);
-    assert_eq!(used.iter().filter(|f| f.starts_with("Qqq Font ")).count(), 15_000);
-    assert_eq!(crate::fonts::missing_fonts(&p).iter().filter(|f| f.to_ascii_lowercase().starts_with("qqq font ")).count(), 15_000);
-}

crates/model/src/table.rs

@@ -77,6 +77,12 @@ impl Default for Table {
     }
 }
 
+/// Cells a table may hold after [`Table::normalize`] (PowerPoint's own limit is 75 x 75).
+pub const MAX_CELLS: usize = 100_000;
+/// Cell visits [`Table::merge`] may spend growing its rectangle over existing merges: a few
+/// passes over the largest grid, where a real table needs one or two.
+const MAX_MERGE_WORK: usize = 16 * MAX_CELLS;
+
 impl Table {
     pub fn new(rows: usize, cols: usize, width: f64, row_height: f64) -> Self {
         let rows = rows.clamp(1, 1000);
@@ -164,8 +170,12 @@ impl Table {
         true
     }
     /// The merged areas: every uncovered cell spanning more than one cell, clipped to the grid.
+    /// Merges of a well-formed table do not overlap, so together they cover at most the grid; a
+    /// file's overlapping spans past that are dropped, so rebuilding them cannot revisit cells
+    /// without bound.
     fn merges(&self) -> Vec<Merge> {
         let (nr, nc) = (self.rows.len(), self.cols.len());
+        let mut area = nr.saturating_mul(nc);
         let mut out = vec![];
         for (r, row) in self.rows.iter().enumerate() {
             for (c, cell) in row.cells.iter().enumerate().take(nc) {
@@ -174,7 +184,10 @@ impl Table {
                 }
                 let rs = (cell.row_span.max(1) as usize).min(nr - r);
                 let gs = (cell.grid_span.max(1) as usize).min(nc - c);
-                if rs > 1 || gs > 1 {
+                if (rs > 1 || gs > 1)
+                    && let Some(left) = area.checked_sub(rs.saturating_mul(gs))
+                {
+                    area = left;
                     out.push(Merge { r, c, rs, gs });
                 }
             }
@@ -227,9 +240,17 @@ impl Table {
         if r1 >= self.rows.len() || c1 >= self.cols.len() || (r0 == r1 && c0 == c1) {
             return false;
         }
-        // Grow the rectangle until no existing merge sticks out of it.
+        // Grow the rectangle until no existing merge sticks out of it. Each pass reads the whole
+        // grid and may grow it by one row, so the passes are charged against
+        // [`MAX_MERGE_WORK`]: past it the merge is refused, before anything changed.
         let (mut r0, mut c0, mut r1, mut c1) = (r0, c0, r1, c1);
+        let pass = self.rows.len().saturating_mul(self.cols.len()).max(1);
+        let mut work = 0usize;
         loop {
+            work = work.saturating_add(pass);
+            if work > MAX_MERGE_WORK {
+                return false;
+            }
             let before = (r0, c0, r1, c1);
             for (r, row) in self.rows.iter().enumerate() {
                 for (c, cell) in row.cells.iter().enumerate().take(self.cols.len()) {
@@ -283,21 +304,31 @@ impl Table {
             return false;
         }
         // Spans come from the file and may reach past the table; only walk the cells that exist.
-        for rr in r..r.saturating_add(rs).min(self.rows.len()) {
-            for cc in c..c.saturating_add(gs).min(self.cols.len()) {
-                if let Some(x) = self.cell_mut(rr, cc) {
-                    x.h_merge = false;
-                    x.v_merge = false;
-                    x.grid_span = 1;
-                    x.row_span = 1;
-                }
+        let end = r.saturating_add(rs).min(self.rows.len());
+        let gs = gs.min(self.cols.len().saturating_sub(c));
+        for row in self.rows.get_mut(r..end).unwrap_or_default() {
+            for x in row.cells.iter_mut().skip(c).take(gs) {
+                x.h_merge = false;
+                x.v_merge = false;
+                x.grid_span = 1;
+                x.row_span = 1;
             }
         }
         true
     }
-    /// Make every row have exactly `cols.len()` cells (repair after loading).
-    pub fn normalize(&mut self) {
+    /// Whether the grid is as [`Table::normalize`] leaves it: within [`MAX_CELLS`], every row
+    /// exactly `cols.len()` cells wide.
+    pub fn is_normalized(&self) -> bool {
         let n = self.cols.len();
+        n <= MAX_CELLS && self.rows.len() <= (MAX_CELLS / n.max(1)).max(1) && self.rows.iter().all(|r| r.cells.len() == n)
+    }
+    /// Make every row have exactly `cols.len()` cells (repair after loading). A grid over
+    /// [`MAX_CELLS`] loses its last rows (and columns) first, so a file naming many rows and many
+    /// columns but few cells cannot make this allocate their product.
+    pub fn normalize(&mut self) {
+        self.cols.truncate(MAX_CELLS);
+        let n = self.cols.len();
+        self.rows.truncate((MAX_CELLS / n.max(1)).max(1));
         for r in &mut self.rows {
             r.cells.resize_with(n, || Cell::new(""));
             for c in &mut r.cells {
@@ -513,6 +544,61 @@ mod tests {
         }
     }
 
+    /// Each cell spans two rows, uncovered: merging the last two rows would grow the rectangle a
+    /// row per pass over the whole grid, 10,000 passes. The merge is refused at its work bound
+    /// instead, leaving the table as it was; a small merge still works.
+    #[test]
+    fn merge_expansion_stops_at_its_work_bound() {
+        let mut t = Table::new(10_000, 10, 400.0, 30.0);
+        for row in &mut t.rows {
+            for c in &mut row.cells {
+                c.row_span = 2;
+            }
+        }
+        let before = t.clone();
+        assert!(!t.merge(9_998, 0, 9_999, 9));
+        assert_eq!(t, before);
+        let mut small = Table::new(4, 4, 400.0, 30.0);
+        assert!(small.merge(0, 0, 1, 1));
+        assert!(small.is_normalized() && !Table { cols: vec![1.0; 3], rows: vec![Row::default(); 2], ..Default::default() }.is_normalized());
+    }
+
+    #[test]
+    fn normalize_caps_a_sparse_grid_at_max_cells() {
+        // 1,000 columns and 1,000 empty rows would be a million cells.
+        let mut t = Table { cols: vec![10.0; 1000], rows: vec![Row::default(); 1000], ..Default::default() };
+        t.normalize();
+        assert_eq!(t.cols.len(), 1000);
+        assert_eq!(t.rows.len(), MAX_CELLS / 1000);
+        assert!(t.rows.iter().all(|r| r.cells.len() == 1000));
+        // Columns alone past the cap keep one row of MAX_CELLS cells.
+        let mut t = Table { cols: vec![1.0; MAX_CELLS + 5], rows: vec![Row::default(); 3], ..Default::default() };
+        t.normalize();
+        assert_eq!((t.cols.len(), t.rows.len()), (MAX_CELLS, 1));
+        // A normal table is untouched.
+        let mut t = Table::new(4, 4, 400.0, 30.0);
+        t.normalize();
+        assert_eq!((t.n_rows(), t.n_cols()), (4, 4));
+    }
+
+    #[test]
+    fn overlapping_file_spans_rebuild_in_bounded_work() {
+        // Every cell an uncovered anchor spanning to the far corner: their areas add up to about
+        // n^4 / 4, but only what fits the grid is kept, so an edit's rebuild stays near n^2.
+        let n = 200;
+        let mut t = Table::new(n, n, 400.0, 30.0);
+        for row in &mut t.rows {
+            for c in &mut row.cells {
+                c.grid_span = u32::MAX;
+                c.row_span = u32::MAX;
+            }
+        }
+        assert_eq!(t.merges().len(), 1, "the first anchor already covers the whole grid");
+        t.insert_row(n);
+        assert_eq!(t.n_rows(), n + 1);
+        assert!(t.split(0, 0));
+    }
+
     fn merged() -> Table {
         // 4x4 with a 2x2 merge at (1,1) holding "M".
         let mut t = Table::new(4, 4, 400.0, 30.0);