Merge branch 'main' into fix/sec-model-table-cells

apps/deckcraft-cli/src/main.rs

@@ -148,6 +148,9 @@ fn warn_if_damaged(path: &str, opened: &serde_json::Value) {
     if slides > 0 {
         eprintln!("deckcraft-cli: warning: {path}: {slides} slide{} couldn't be read and left out", if slides == 1 { "" } else { "s" });
     }
+    for w in opened.get("importWarnings").and_then(|v| v.as_array()).into_iter().flatten().filter_map(|w| w.as_str()) {
+        eprintln!("deckcraft-cli: warning: {path}: {w}");
+    }
     if media > 0 {
         eprintln!("deckcraft-cli: warning: {path}: {media} picture or media file{} couldn't be read", if media == 1 { "" } else { "s" });
     }

crates/engine/src/cmd.rs

@@ -433,6 +433,15 @@ 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 {
+            normalized = t.normalized();
+            &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/file.rs

@@ -139,11 +139,13 @@ pub struct Damage {
     pub media: Vec<String>,
     /// Slides a `.pptx` lists but that couldn't be read, so they are not in the presentation.
     pub slides: Vec<String>,
+    /// Content a `.pptx` had that was cut to fit a limit; the file still opened.
+    pub warnings: Vec<String>,
 }
 
 impl Damage {
     pub fn is_empty(&self) -> bool {
-        self.media.is_empty() && self.slides.is_empty()
+        self.media.is_empty() && self.slides.is_empty() && self.warnings.is_empty()
     }
 }
 
@@ -151,17 +153,17 @@ impl Damage {
 pub fn open_checked(name: &str, bytes: &[u8]) -> Result<(Presentation, Damage)> {
     if deckcraft_format::sniff(bytes) {
         let (p, media) = deckcraft_format::load_checked(bytes).map_err(|e| EngineError::Other(e.to_string()))?;
-        return Ok((p, Damage { media, slides: vec![] }));
+        return Ok((p, Damage { media, ..Default::default() }));
     }
-    open_other(name, bytes).map(|(p, slides)| (p, Damage { media: vec![], slides }))
+    open_other(name, bytes).map(|(p, slides, warnings)| (p, Damage { slides, warnings, ..Default::default() }))
 }
 
-fn open_other(name: &str, bytes: &[u8]) -> Result<(Presentation, Vec<String>)> {
+fn open_other(name: &str, bytes: &[u8]) -> Result<(Presentation, Vec<String>, Vec<String>)> {
     if deckcraft_pptx::sniff(bytes) {
-        let (mut p, lost_slides) = deckcraft_pptx::import_checked(bytes).map_err(|e| EngineError::Other(e.to_string()))?;
+        let (mut p, lost_slides, warnings) = deckcraft_pptx::import_report(bytes).map_err(|e| EngineError::Other(e.to_string()))?;
         deckcraft_format::repair(&mut p);
         probe_clips(&mut p);
-        return Ok((p, lost_slides));
+        return Ok((p, lost_slides, warnings));
     }
     if deckcraft_pptx::is_encrypted(bytes) {
         return Err(EngineError::Other(format!("{name} is password-protected. DeckCraft can't open password-protected presentations yet.")));
@@ -180,7 +182,7 @@ fn open_other(name: &str, bytes: &[u8]) -> Result<(Presentation, Vec<String>)> {
         if p.slides.is_empty() {
             p = defaults::new_presentation(None);
         }
-        return Ok((p, vec![]));
+        return Ok((p, vec![], vec![]));
     }
     Err(EngineError::Other(format!("{name}: not a presentation DeckCraft can read")))
 }
@@ -276,7 +278,7 @@ fn adopt_loaded(s: &mut Session, f: LoadedFile) -> Result<Value> {
     let st = s.doc()?;
     Ok(json!({
         "document": i, "slides": st.doc.slides.len(), "missingFonts": st.missing_fonts,
-        "damagedMedia": damaged.media, "lostSlides": damaged.slides,
+        "damagedMedia": damaged.media, "lostSlides": damaged.slides, "importWarnings": damaged.warnings,
     }))
 }
 
@@ -311,7 +313,9 @@ fn open_bytes(s: &mut Session, p: &Value) -> Result<Value> {
     let (doc, damaged) = open_checked(name, &data)?;
     let i = add_opened(s, name, None, doc);
     report_damage(s, &damaged)?;
-    Ok(json!({"document": i, "missingFonts": s.doc()?.missing_fonts, "damagedMedia": damaged.media, "lostSlides": damaged.slides}))
+    Ok(
+        json!({"document": i, "missingFonts": s.doc()?.missing_fonts, "damagedMedia": damaged.media, "lostSlides": damaged.slides, "importWarnings": damaged.warnings}),
+    )
 }
 
 /// Tell the user which pictures, media and slides a damaged file lost, and mark the copy repaired.
@@ -339,6 +343,10 @@ fn report_damage(s: &mut Session, damaged: &Damage) -> Result<()> {
             if n == 1 { "was" } else { "were" },
         ));
     }
+    for w in &damaged.warnings {
+        text.push_str(w);
+        text.push_str(". ");
+    }
     text.push_str("It was opened as a repaired copy; save it under a new name to keep the original file.");
     s.ui_requests.push(crate::UiRequest::Message { text });
     Ok(())

crates/engine/src/cmd/table.rs

@@ -286,6 +286,12 @@ 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).
+            // An over-cap table is refused, not trimmed: a trim would delete rows while reporting success.
+            if t.over_cap() {
+                return Err(bad(cmd, format!("the table is larger than {} cells; edit refused", deckcraft_model::table::MAX_CELLS)));
+            }
+            t.normalize();
             f(t)?;
             t.normalize();
             let (w, h) = (t.width(), t.height());
@@ -319,10 +325,17 @@ fn rows_cols(s: &mut Session, p: &Value, op: &str, name: &'static str) -> Result
     let op = op.to_string();
     with_table(s, p, name, move |t| {
         match op.as_str() {
+            "rowAbove" | "rowBelow" | "colLeft" | "colRight" => {
+                let ok = match op.as_str() {
                     "rowAbove" => t.insert_row(r),
                     "rowBelow" => t.insert_row(r.saturating_add(1)),
                     "colLeft" => t.insert_col(c),
-            "colRight" => t.insert_col(c.saturating_add(1)),
+                    _ => t.insert_col(c.saturating_add(1)),
+                };
+                if !ok {
+                    return Err(bad(name, format!("the table is at its {}-cell limit", deckcraft_model::table::MAX_CELLS)));
+                }
+            }
             "delRow" => {
                 if !t.delete_row(r) {
                     return Err(bad(name, "can't delete the last row"));

crates/engine/src/tests.rs

@@ -6921,6 +6921,45 @@ fn master_body_placeholder_format_keeps_per_level_formatting() {
     assert_eq!(shown(&s), vec![Some(28.0), Some(24.0)]);
 }
 
+#[test]
+fn table_edits_past_the_cell_cap_are_refused_not_trimmed() {
+    use deckcraft_model::table::{MAX_CELLS, Row};
+    let mut s = session();
+    let id = table_at(&mut s, &[100.0, 100.0], json!([["a", "b"], ["c", "d"]]));
+    let cols = 200;
+    let rows = MAX_CELLS / cols;
+    let big = deckcraft_model::Table {
+        cols: vec![1.0; cols],
+        rows: vec![Row { height: 1.0, cells: vec![Default::default(); cols] }; rows],
+        ..Default::default()
+    };
+    for sl in std::sync::Arc::make_mut(&mut s.doc_mut().unwrap().doc).slides.iter_mut() {
+        for sh in std::sync::Arc::make_mut(sl).shapes.iter_mut() {
+            if sh.id == id {
+                sh.kind = deckcraft_model::ShapeKind::Table(big.clone());
+            }
+        }
+    }
+    for cmd in ["table.insertRowAbove", "table.insertRowBelow"] {
+        let e = s.execute(cmd, &json!({"id": id.0, "row": 0})).unwrap_err().to_string();
+        assert!(e.contains("cell limit"), "{cmd}: {e}");
+    }
+    let e = s.execute("table.insertColumnLeft", &json!({"id": id.0, "col": 0})).unwrap_err().to_string();
+    assert!(e.contains("cell limit"), "{e}");
+    assert_eq!(table_of(&s, id), big, "a refused insert leaves every row in place");
+    // A table already over the cap is refused too, not trimmed behind a success.
+    let huge = deckcraft_model::Table { rows: vec![Row { height: 1.0, cells: vec![Default::default(); cols] }; rows + 1], ..big.clone() };
+    for sl in std::sync::Arc::make_mut(&mut s.doc_mut().unwrap().doc).slides.iter_mut() {
+        for sh in std::sync::Arc::make_mut(sl).shapes.iter_mut() {
+            if sh.id == id {
+                sh.kind = deckcraft_model::ShapeKind::Table(huge.clone());
+            }
+        }
+    }
+    assert!(s.execute("table.deleteRow", &json!({"id": id.0, "row": 0})).is_err());
+    assert_eq!(table_of(&s, id).n_rows(), rows + 1);
+}
+
 /// A shape named many times in one edit is edited once and not copied per mention.
 #[test]
 fn repeated_target_ids_edit_the_shape_once() {

crates/model/src/table.rs

@@ -83,6 +83,12 @@ 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);
@@ -113,15 +119,24 @@ impl Table {
         self.rows.iter().map(|r| r.height).sum()
     }
     /// Insert an empty row before `at`. A merge the new row falls inside grows to cover it.
-    pub fn insert_row(&mut self, at: usize) {
+    /// Returns false, leaving the table as it was, when the row would take it past [`MAX_CELLS`].
+    pub fn insert_row(&mut self, at: usize) -> bool {
+        if self.cols.len().saturating_mul(self.rows.len().saturating_add(1)) > MAX_CELLS {
+            return false;
+        }
         let at = at.min(self.rows.len());
         let merges = self.merges();
         let h = self.rows.get(at.saturating_sub(1)).map(|r| r.height).unwrap_or(30.0);
         self.rows.insert(at, Row { height: h, cells: (0..self.cols.len()).map(|_| Cell::new("")).collect() });
         self.set_merges(merges.into_iter().map(|m| m.insert_row(at)));
+        true
     }
     /// Insert an empty column before `at`. A merge the new column falls inside grows to cover it.
-    pub fn insert_col(&mut self, at: usize) {
+    /// Returns false, leaving the table as it was, when the column would take it past [`MAX_CELLS`].
+    pub fn insert_col(&mut self, at: usize) -> bool {
+        if self.cols.len().saturating_add(1).saturating_mul(self.rows.len()) > MAX_CELLS {
+            return false;
+        }
         let at = at.min(self.cols.len());
         let merges = self.merges();
         let w = self.cols.get(at.saturating_sub(1)).copied().unwrap_or(80.0);
@@ -131,6 +146,7 @@ impl Table {
             r.cells.insert(i, Cell::new(""));
         }
         self.set_merges(merges.into_iter().map(|m| m.insert_col(at)));
+        true
     }
     /// Delete row `at`. Merges through it shrink; a merge anchored on it moves its anchor (and
     /// text) to the next row it covers.
@@ -170,8 +186,12 @@ 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) {
@@ -180,7 +200,10 @@ 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 });
                 }
             }
@@ -233,9 +256,17 @@ 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()) {
@@ -289,21 +320,50 @@ 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()) {
+        let end = r.saturating_add(rs).min(self.rows.len());
-            for cc in c..c.saturating_add(gs).min(self.cols.len()) {
+        let gs = gs.min(self.cols.len().saturating_sub(c));
-                if let Some(x) = self.cell_mut(rr, cc) {
+        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).
+    /// Whether the grid is as [`Table::normalize`] leaves it: within [`MAX_CELLS`], every row
-    pub fn normalize(&mut self) {
+    /// 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)
+    }
+    /// Whether [`Table::normalize`] would drop rows or columns to fit [`MAX_CELLS`].
+    pub fn over_cap(&self) -> bool {
+        let n = self.cols.len();
+        n > MAX_CELLS || self.rows.len() > (MAX_CELLS / n.max(1)).max(1)
+    }
+    /// What [`Table::normalize`] would leave of this table, cloning only the rows and cells that
+    /// survive (an oversized table is not copied whole first).
+    pub fn normalized(&self) -> Table {
+        let n = self.cols.len().min(MAX_CELLS);
+        let keep = (MAX_CELLS / n.max(1)).max(1);
+        let rows = self.rows.iter().take(keep).map(|r| Row { height: r.height, cells: r.cells.iter().take(n).cloned().collect() }).collect();
+        let mut t = Table { cols: self.cols[..n].to_vec(), rows, background: self.background.clone(), style: self.style.clone(), ..*self };
+        t.normalize();
+        if self.over_cap() {
+            t.clip_spans();
+        }
+        t
+    }
+    /// 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) {
+        let truncated = self.over_cap();
+        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 {
@@ -311,6 +371,20 @@ impl Table {
                 c.row_span = c.row_span.max(1);
             }
         }
+        if truncated {
+            self.clip_spans();
+        }
+    }
+    /// Cut every merge span to the rows and columns that remain, so no merge reaches past a
+    /// table that lost rows or columns to the cell cap.
+    fn clip_spans(&mut self) {
+        let (nr, nc) = (self.rows.len(), self.cols.len());
+        for (ri, r) in self.rows.iter_mut().enumerate() {
+            for (ci, c) in r.cells.iter_mut().enumerate() {
+                c.row_span = c.row_span.min((nr - ri) as u32).max(1);
+                c.grid_span = c.grid_span.min((nc - ci) as u32).max(1);
+            }
+        }
     }
 }
 
@@ -519,6 +593,102 @@ 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(1_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(998, 0, 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));
+        // `normalized` leaves what `normalize` does.
+        let wide = Table { cols: vec![10.0; 1000], rows: vec![Row { height: 1.0, cells: vec![Cell::new("x"); 1500] }; 150], ..Default::default() };
+        let mut n = wide.clone();
+        n.normalize();
+        assert_eq!(wide.normalized(), n);
+        // 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");
+        assert!(t.insert_row(n));
+        assert_eq!(t.n_rows(), n + 1);
+        assert!(t.split(0, 0));
+    }
+
+    /// A table at the cell cap refuses another row or column instead of losing rows to
+    /// normalization, and the cap sits above PowerPoint's own 75 x 75.
+    #[test]
+    fn insert_past_the_cap_is_refused() {
+        const { assert!(MAX_CELLS >= 75 * 75) };
+        let mut t =
+            Table { cols: vec![1.0; 200], rows: vec![Row { height: 1.0, cells: vec![Cell::new(""); 200] }; MAX_CELLS / 200], ..Default::default() };
+        assert!(!t.over_cap());
+        let before = t.clone();
+        assert!(!t.insert_row(0));
+        assert!(!t.insert_col(0));
+        assert_eq!(t, before);
+        assert!(Table::new(75, 75, 100.0, 10.0).insert_row(0));
+    }
+
+    /// A merge that crosses the truncation boundary is clipped to the rows and columns that remain.
+    #[test]
+    fn merges_are_clipped_when_the_cap_drops_rows() {
+        let cols = 1000;
+        let rows = MAX_CELLS / cols + 1;
+        let mut t = Table { cols: vec![1.0; cols], rows: vec![Row { height: 1.0, cells: vec![Cell::new(""); cols] }; rows], ..Default::default() };
+        t.rows[0].cells[0].row_span = rows as u32;
+        t.rows[0].cells[cols - 1].grid_span = 5;
+        let n = t.normalized();
+        assert_eq!(n.rows.len(), rows - 1);
+        assert_eq!(n.rows[0].cells[0].row_span as usize, rows - 1);
+        assert_eq!(n.rows[0].cells[cols - 1].grid_span, 1);
+        t.normalize();
+        assert_eq!(t.rows[0].cells[0].row_span as usize, rows - 1);
+        // Within the cap, spans are left as the file wrote them.
+        let mut small = Table::new(3, 3, 30.0, 10.0);
+        small.rows[2].cells[2].grid_span = 4;
+        small.normalize();
+        assert_eq!(small.rows[2].cells[2].grid_span, 4);
+    }
+
     fn merged() -> Table {
         // 4x4 with a 2x2 merge at (1,1) holding "M".
         let mut t = Table::new(4, 4, 400.0, 30.0);
@@ -552,10 +722,10 @@ mod tests {
     fn row_and_column_edits_keep_merges_consistent() {
         for at in 0..=4 {
             let mut t = merged();
-            t.insert_row(at);
+            assert!(t.insert_row(at));
             assert_topology(&t);
             let mut t = merged();
-            t.insert_col(at);
+            assert!(t.insert_col(at));
             assert_topology(&t);
         }
         for at in 0..4 {

crates/pptx/src/lib.rs

@@ -70,6 +70,12 @@ pub fn import_checked(bytes: &[u8]) -> Result<(Presentation, Vec<String>), PptxE
     read::import_checked(bytes)
 }
 
+/// [`import_checked`], plus warnings for content cut to fit a limit (a table past the cell cap
+/// keeps its leading rows); the file still opens.
+pub fn import_report(bytes: &[u8]) -> Result<(Presentation, Vec<String>, Vec<String>), PptxError> {
+    read::import_report(bytes)
+}
+
 /// [`import`] with a lower presentation-wide chart point budget, so tests reach the over-budget
 /// path without a multi-million-point chart.
 #[doc(hidden)]

crates/pptx/src/read.rs

@@ -55,12 +55,24 @@ pub struct Imp<'a> {
     pub tag_bytes: usize,
     /// The tag budget ran out and the warning was given.
     pub tags_capped: bool,
+    /// What the file lost that the reader kept going past (see [`MAX_IMPORT_WARNINGS`]).
+    pub warnings: Vec<String>,
     /// Whether a part has relationships of its own, by normalized part name: looked up once
     /// however many embedded objects point at the part.
     pub has_rels: HashMap<String, bool>,
 }
 
+/// Import warnings kept; a hostile file with many oversized tables adds no more than this.
+const MAX_IMPORT_WARNINGS: usize = 32;
+
 impl Imp<'_> {
+    /// Record something the file lost, and log it.
+    pub fn warn(&mut self, msg: String) {
+        log::warn!("pptx: {msg}");
+        if self.warnings.len() < MAX_IMPORT_WARNINGS {
+            self.warnings.push(msg);
+        }
+    }
     pub fn alloc(&mut self) -> u32 {
         self.next = self.next.saturating_add(1);
         self.next
@@ -462,6 +474,7 @@ pub fn import_theme(bytes: &[u8]) -> Result<(deckcraft_model::theme::Theme, Vec<
         tags_by_part: HashMap::new(),
         tag_bytes: 0,
         tags_capped: false,
+        warnings: vec![],
         has_rels: HashMap::new(),
     };
     let part = Part::new(&pkg, &theme_path, &doc);
@@ -476,11 +489,22 @@ pub fn import(bytes: &[u8]) -> Result<Presentation, PptxError> {
 /// [`import`], plus the parts of the slide list the file names but that couldn't be read (a
 /// missing or truncated slide): those slides are left out, so a caller can say so (#203).
 pub fn import_checked(bytes: &[u8]) -> Result<(Presentation, Vec<String>), PptxError> {
-    import_with_chart_points(bytes, chart::MAX_PRESENTATION_CHART_POINTS)
+    import_report(bytes).map(|(p, lost, _)| (p, lost))
+}
+
+/// [`import_checked`], plus warnings for content the file had that was cut to fit a limit (a
+/// table past the cell cap keeps its leading rows).
+pub fn import_report(bytes: &[u8]) -> Result<(Presentation, Vec<String>, Vec<String>), PptxError> {
+    import_report_with_chart_points(bytes, chart::MAX_PRESENTATION_CHART_POINTS)
 }
 
 /// [`import_checked`] with the presentation's chart point budget set (tests lower it).
 pub(crate) fn import_with_chart_points(bytes: &[u8], chart_points_cap: usize) -> Result<(Presentation, Vec<String>), PptxError> {
+    import_report_with_chart_points(bytes, chart_points_cap).map(|(p, lost, _)| (p, lost))
+}
+
+/// [`import_report`] with the presentation's chart point budget set.
+fn import_report_with_chart_points(bytes: &[u8], chart_points_cap: usize) -> Result<(Presentation, Vec<String>, Vec<String>), PptxError> {
     let mut lost_slides: Vec<String> = vec![];
     let pkg = Package::open(bytes)?;
     let root_rels = pkg.rels("");
@@ -513,6 +537,7 @@ pub(crate) fn import_with_chart_points(bytes: &[u8], chart_points_cap: usize) ->
         tags_by_part: HashMap::new(),
         tag_bytes: 0,
         tags_capped: false,
+        warnings: vec![],
         has_rels: HashMap::new(),
     };
     // Slide ids first so hyperlinks anywhere can point at slides.
@@ -544,7 +569,7 @@ pub(crate) fn import_with_chart_points(bytes: &[u8], chart_points_cap: usize) ->
     p.media = std::mem::take(&mut imp.media);
     p.next_id = imp.next;
     p.fix_next_id();
-    Ok((p, lost_slides))
+    Ok((p, lost_slides, std::mem::take(&mut imp.warnings)))
 }
 
 /// Masters, layouts and slides to read, in file order and without duplicates: (master paths,

crates/pptx/src/read/shapes.rs

@@ -785,6 +785,9 @@ fn table(imp: &mut Imp, part: &Part, tbl: &El) -> Table {
         let n = t.rows.iter().map(|r| r.cells.len()).max().unwrap_or(0);
         t.cols = vec![72.0; n];
     }
+    if t.over_cap() {
+        imp.warn(format!("a table in {} is over {} cells; its last rows were left out", part.name, deckcraft_model::table::MAX_CELLS));
+    }
     t.normalize();
     t
 }

crates/pptx/src/write.rs

@@ -65,6 +65,8 @@ pub struct Exp<'a> {
     n_kept: u32,
     pub slide_parts: HashMap<SlideId, String>,
     pub table_styles: Vec<String>,
+    /// A table past the cell cap was met; export fails rather than write a trimmed one.
+    pub table_over_cap: bool,
     /// The ids in `table_styles`, for checking membership without a scan.
     pub table_style_set: std::collections::HashSet<String>,
     placeholder: Option<String>,
@@ -558,6 +560,7 @@ pub fn export(p: &Presentation, main_ct: &str) -> Result<Vec<u8>, PptxError> {
         n_kept: 0,
         slide_parts: HashMap::new(),
         table_styles: vec![],
+        table_over_cap: false,
         table_style_set: Default::default(),
         placeholder: None,
     };
@@ -766,6 +769,12 @@ pub fn export(p: &Presentation, main_ct: &str) -> Result<Vec<u8>, PptxError> {
         root.add_new(&k.rel_type, &k.name, false);
     }
     x.pkg.rels("", &root);
+    if x.table_over_cap {
+        return Err(PptxError::Unsupported(format!(
+            "a table has more than {} cells; export refused rather than drop rows",
+            deckcraft_model::table::MAX_CELLS
+        )));
+    }
     x.pkg.finish()
 }
 

crates/pptx/src/write/shapes.rs

@@ -721,8 +721,8 @@ fn frame_close(w: &mut W) {
 }
 
 fn table(w: &mut W, x: &mut Exp, o: &mut Out, t: &Table) {
-    let mut t = t.clone();
+    x.table_over_cap |= t.over_cap();
-    t.normalize();
+    let t = t.normalized();
     w.open0("a:tbl");
     let (guid, _) = crate::tables::guid_for_style(&t.style);
     if x.table_style_set.insert(t.style.clone()) {

crates/pptx/tests/it/malformed.rs

@@ -288,6 +288,44 @@ fn unreadable_slide_is_reported() {
     assert_eq!(deckcraft_pptx::import(&broken).expect("import").slides.len(), 1);
 }
 
+fn table_deck(rows: usize, cols: usize) -> deckcraft_model::Presentation {
+    use deckcraft_model::table::Row;
+    use deckcraft_model::{Shape, ShapeId, ShapeKind, Table, Xfrm};
+    let mut p = defaults::new_presentation(None);
+    let id = ShapeId(p.alloc_id().expect("ids"));
+    let t = Table { cols: vec![1.0; cols], rows: vec![Row { height: 1.0, cells: vec![Default::default(); cols] }; rows], ..Default::default() };
+    let slide = std::sync::Arc::make_mut(&mut p.slides[0]);
+    slide.shapes.push(Shape { id, name: "T".into(), xfrm: Some(Xfrm::new(0.0, 0.0, 10.0, 10.0)), kind: ShapeKind::Table(t), ..Default::default() });
+    p
+}
+
+/// A table past the cell cap opens with its leading rows and an import warning; saving one that
+/// is over the cap is an error, never a silent trim.
+#[test]
+fn oversized_table_warns_on_import_and_refuses_export() {
+    let max = deckcraft_model::table::MAX_CELLS;
+    let cols = 1000;
+    let rows = max / cols;
+    // A table exactly at the cap round-trips with no warning.
+    let bytes = deckcraft_pptx::export(&table_deck(rows, cols)).expect("export at the cap");
+    let (_, _, warnings) = deckcraft_pptx::import_report(&bytes).expect("import");
+    assert!(warnings.is_empty(), "{warnings:?}");
+    // One more row in the file: kept truncated, and reported.
+    let mut es = entries(&bytes);
+    let slide = es.iter_mut().find(|(n, _)| n == "ppt/slides/slide1.xml").expect("slide1");
+    let xml = String::from_utf8(std::mem::take(&mut slide.1)).expect("utf8");
+    let start = xml.rfind("<a:tr ").expect("row");
+    let end = xml.rfind("</a:tr>").expect("row end") + "</a:tr>".len();
+    slide.1 = format!("{}{}{}", &xml[..end], &xml[start..end], &xml[end..]).into_bytes();
+    let (q, _, warnings) = deckcraft_pptx::import_report(&rezip(&es)).expect("import");
+    assert_eq!(warnings.len(), 1, "{warnings:?}");
+    let t = q.slides[0].shapes.iter().find_map(|s| if let deckcraft_model::ShapeKind::Table(t) = &s.kind { Some(t) } else { None }).expect("table");
+    assert_eq!((t.rows.len(), t.cols.len()), (rows, cols));
+    // A model table past the cap is refused on export.
+    let err = deckcraft_pptx::export(&table_deck(rows + 1, cols)).expect_err("over the cap");
+    assert!(err.to_string().contains("table"), "{err}");
+}
+
 /// Many slides naming one notes part: it is read for the first slide only, not copied onto every
 /// slide that names it.
 #[test]