address review: keep the original code in the public number_format for overlong formats

The display cache hashed the full format code on every lookup, so one
huge code shared by many cells cost its length for every cell drawn.
A code with more than 4 × 255 bytes can't be within Excel's 255
characters and is now formatted as General straight away.

crates/engine/src/display.rs

@@ -8,8 +8,13 @@ use gridcraft_model::{Sheet, Workbook};
 use gridcraft_numfmt::{Formatted, NumberFormat, format_value};
 
 /// Parsed number formats are cached by code.
-fn parsed(code: &str) -> NumberFormat {
+pub(crate) fn parsed(code: &str) -> NumberFormat {
     static CACHE: std::sync::LazyLock<Mutex<HashMap<String, NumberFormat>>> = std::sync::LazyLock::new(|| Mutex::new(HashMap::new()));
+    // A code of more bytes than 4 × the character limit can't be within it: it shows as General
+    // without being hashed or scanned again for every cell that shares it.
+    if code.len() > 4 * gridcraft_numfmt::MAX_FORMAT_CODE_CHARS {
+        return NumberFormat::parse("General");
+    }
     let lock = || CACHE.lock().unwrap_or_else(std::sync::PoisonError::into_inner);
     if let Some(f) = lock().get(code) {
         return f.clone();
@@ -24,7 +29,12 @@ fn parsed(code: &str) -> NumberFormat {
     f
 }
 
+/// The parsed format of `code`; one longer than the character limit is a General format that keeps
+/// the code it was made from. The engine itself formats through `parsed`, which doesn't copy it.
 pub fn number_format(code: &str) -> NumberFormat {
+    if code.len() > 4 * gridcraft_numfmt::MAX_FORMAT_CODE_CHARS {
+        return NumberFormat::parse(code);
+    }
     parsed(code)
 }
 
@@ -226,3 +236,31 @@ pub fn stats(sheet: &Sheet, ranges: &[RangeRef]) -> Stats {
     }
     s
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// A format code far past Excel's 255 characters shows its values as General, for every cell
+    /// that shares it, without being hashed into the cache each time.
+    #[test]
+    fn an_overlong_code_formats_as_general() {
+        let wb = Workbook::new();
+        let code = format!("0.00{}", "\"x\"".repeat(400_000));
+        for _ in 0..1_000 {
+            assert_eq!(format(&Value::Number(1.5), &code, &wb).text, "1.5");
+        }
+        assert_eq!(format(&Value::Number(1.5), "0.00", &wb).text, "1.50");
+    }
+
+    /// The public `number_format` keeps the code of an overlong format; it is General, but not the
+    /// same format as a real `General`.
+    #[test]
+    fn number_format_keeps_an_overlong_code() {
+        let code = format!("0.00{}", "\"x\"".repeat(2_000));
+        let f = number_format(&code);
+        assert!(f.is_general());
+        assert_eq!(f.code(), code);
+        assert_ne!(f, number_format("General"));
+    }
+}

crates/engine/src/io.rs

@@ -3,7 +3,7 @@
 use std::fmt::Write as _;
 
 use gridcraft_core::{CellRef, RangeRef};
-use gridcraft_model::{LineIndex, Sheet, Workbook};
+use gridcraft_model::{Sheet, Workbook};
 
 use crate::{EngineError, Result};
 
@@ -443,7 +443,7 @@ pub fn csv_values_bytes(wb: &Workbook) -> Result<Vec<u8>> {
     let Some(sh) = wb.sheet(wb.active_sheet()) else { return Ok(Vec::new()) };
     gridcraft_xlsx::write_csv(sh, wb.csv_delimiter.unwrap_or(b','), |c| match sh.value(c) {
         gridcraft_core::Value::Number(n) if n.is_finite() => {
-            let shape = crate::display::number_format(wb.styles.get(sh.style_id(c)).num_fmt.as_str()).date_shape(n);
+            let shape = crate::display::parsed(wb.styles.get(sh.style_id(c)).num_fmt.as_str()).date_shape(n);
             match shape.zip(gridcraft_core::date::datetime_from_serial(wb.date_system, n)) {
                 Some(((date, time), d)) => iso(&d, date, time),
                 None => n.to_string(),
@@ -548,23 +548,6 @@ fn table_dir(sh: &Sheet) -> &'static str {
 const CLIPBOARD_HTML_MAX_BYTES: usize = 4 * 1024 * 1024;
 const CLIPBOARD_HTML_MAX_CELLS: u64 = 10_000;
 
-/// The merge covering each cell of `rect`, as `Sheet::merge_at` would find it (the first in the
-/// sheet's list), built once so a table lookup is one probe instead of a scan of every merge.
-/// `None` when overlapping merges would cost more than a few visits per cell of `rect`: a
-/// hostile file's merge list can't multiply the work of a bounded export.
-fn merge_index(sh: &Sheet, rect: RangeRef) -> Option<std::collections::HashMap<CellRef, RangeRef>> {
-    let mut map = std::collections::HashMap::new();
-    let mut left = rect.count().saturating_mul(4);
-    for m in &sh.merges {
-        let Some(part) = m.intersection(&rect) else { continue };
-        left = left.checked_sub(part.count())?;
-        for c in part.iter() {
-            map.entry(c).or_insert(*m);
-        }
-    }
-    Some(map)
-}
-
 /// An HTML fragment for the same range as `plain_text`, previously produced by
 /// `display::range_text`. Large copies keep plain text only; never return a partial table.
 pub(crate) fn range_html(wb: &Workbook, sheet: usize, r: RangeRef, plain_text: &str) -> Option<String> {
@@ -573,7 +556,6 @@ pub(crate) fn range_html(wb: &Workbook, sheet: usize, r: RangeRef, plain_text: &
         return None;
     }
     let sh = wb.sheet(sheet)?;
-    let merges = merge_index(sh, r)?;
     let mut out = ClipboardHtml(String::new());
     write!(out, "<table{} style=\"border-collapse:collapse;font-family:Calibri,Carlito,Arial,sans-serif;font-size:11pt\">", table_dir(sh)).ok()?;
     for row in r.start.row..=r.end.row {
@@ -583,7 +565,7 @@ pub(crate) fn range_html(wb: &Workbook, sheet: usize, r: RangeRef, plain_text: &
         out.write_str("<tr>").ok()?;
         for col in r.start.col..=r.end.col {
             let c = CellRef::new(row, col);
-            let span = if let Some(merged) = merges.get(&c).and_then(|m| m.intersection(&r)) {
+            let span = if let Some(merged) = sh.merges.iter().find(|m| m.contains(c)).and_then(|m| m.intersection(&r)) {
                 let first_row = (merged.start.row..=merged.end.row).find(|&row| !sh.is_row_hidden(row))?;
                 if c != CellRef::new(first_row, merged.start.col) {
                     continue;
@@ -646,14 +628,6 @@ const HTML_EXPORT_MAX_ROWS: u32 = 100_001;
 /// tall, wide one) would otherwise be gigabytes of `<td></td>` before finishing (#2387).
 const HTML_EXPORT_MAX_CELLS: u64 = 4_000_000;
 
-/// Most bytes of one web page. Every cell may hold long text (shared across many cells in the
-/// file), so the cell count alone doesn't bound the page; past this the export fails instead.
-const HTML_EXPORT_MAX_BYTES: usize = 256 * 1024 * 1024;
-
-fn html_too_big() -> EngineError {
-    EngineError::Other(format!("the web page would be over {} MiB. Save as .xlsx or CSV instead.", HTML_EXPORT_MAX_BYTES / (1024 * 1024)))
-}
-
 /// A sheet as a standalone HTML table (Save as Web Page). A sheet with pictures gets the sheet's
 /// own column widths and row heights so each picture and chart (inline SVG) can sit at its anchor.
 pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
@@ -696,23 +670,6 @@ pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
                 r.width()
             )));
         }
-        let merges = merge_index(sh, r).ok_or_else(|| EngineError::Other("the sheet's merged cells overlap too much for a web page.".into()))?;
-        // Visible rows and columns of the table before each line, so a merge's span is counted
-        // in constant time whatever the file says its size is.
-        let visible_before = |hidden: &dyn Fn(u32) -> bool, from: u32, to: u32| {
-            let mut v = Vec::with_capacity((to - from) as usize + 2);
-            v.push(0u32);
-            for line in from..=to {
-                v.push(v.last().copied().unwrap_or(0) + u32::from(!hidden(line)));
-            }
-            v
-        };
-        let rows_before = visible_before(&|y| sh.is_row_hidden(y), r.start.row, r.end.row);
-        let cols_before = visible_before(&|x| sh.is_col_hidden(x), r.start.col, r.end.col);
-        let visible = |before: &[u32], from: u32, a: u32, b: u32| {
-            let at = |line: u32| before.get((line - from) as usize).copied().unwrap_or(0);
-            at(b + 1) - at(a)
-        };
         if fixed {
             out.push_str("<colgroup>");
             for col in (r.start.col..=r.end.col).filter(|&c| !sh.is_col_hidden(c)) {
@@ -731,19 +688,20 @@ pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
             }
             for col in r.start.col..=r.end.col {
                 let c = CellRef::new(row, col);
-                let merge = merges.get(&c);
-                if merge.is_some_and(|m| m.start != c) || (fixed && sh.is_col_hidden(col)) {
+                if sh.merges.iter().any(|m| m.contains(c) && m.start != c) || (fixed && sh.is_col_hidden(col)) {
                     continue;
                 }
                 let text = crate::display::cell_text(wb, sh, c);
                 let dir = CellDir::of(sh, c, &text);
                 let css = html_cell_css(wb, sh, c, dir);
-                let span = merge
+                let span = sh
+                    .merges
+                    .iter()
+                    .find(|m| m.start == c)
                     .map(|m| {
                         if fixed {
-                            // Spans count the table's visible lines: the merge is clipped to the table.
-                            let rows = visible(&rows_before, r.start.row, m.start.row, m.end.row.min(r.end.row));
-                            let cols = visible(&cols_before, r.start.col, m.start.col, m.end.col.min(r.end.col));
+                            let rows = (m.start.row..=m.end.row).filter(|&y| !sh.is_row_hidden(y)).count();
+                            let cols = (m.start.col..=m.end.col).filter(|&x| !sh.is_col_hidden(x)).count();
                             format!(" rowspan=\"{rows}\" colspan=\"{cols}\"")
                         } else {
                             format!(" rowspan=\"{}\" colspan=\"{}\"", m.height(), m.width())
@@ -751,9 +709,6 @@ pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
                     })
                     .unwrap_or_default();
                 let _ = write!(out, "<td style=\"{css}\"{}{span}>{}</td>", dir.attr(), esc(&text));
-                if out.len() > HTML_EXPORT_MAX_BYTES {
-                    return Err(html_too_big());
-                }
             }
             out.push_str("</tr>");
         }
@@ -761,15 +716,12 @@ pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
     out.push_str("</table>");
     if fixed {
         // Sheet units are CSS pixels; the table starts at the used range, not at A1.
-        // Indexed positions: each anchor would otherwise sum every custom row and column before it.
-        let ri = LineIndex::new(&sh.rows, sh.default_row_height, gridcraft_core::MAX_ROWS);
-        let ci = LineIndex::new(&sh.cols, sh.default_col_width, gridcraft_core::MAX_COLS);
-        let (left, top) = used.map_or((0.0, 0.0), |r| (ci.start(r.start.col), ri.start(r.start.row)));
+        let (left, top) = used.map_or((0.0, 0.0), |r| (sh.col_left(r.start.col), sh.row_top(r.start.row)));
         let place = |a: &gridcraft_model::Anchor| {
             format!(
                 "position:absolute;left:{}px;top:{}px;width:{}px;height:{}px",
-                ci.start(a.cell.col) - left + f64::from(a.dx),
-                ri.start(a.cell.row) - top + f64::from(a.dy),
+                sh.col_left(a.cell.col) - left + f64::from(a.dx),
+                sh.row_top(a.cell.row) - top + f64::from(a.dy),
                 a.width.max(0.0),
                 a.height.max(0.0)
             )
@@ -783,24 +735,15 @@ pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
                 esc(&im.alt),
                 place(&im.anchor)
             );
-            if out.len() > HTML_EXPORT_MAX_BYTES {
-                return Err(html_too_big());
-            }
         }
         for ch in &sh.charts {
             if let Some(svg) = crate::chart_svg::chart_svg(wb, sheet, ch, &place(&ch.anchor)) {
                 out.push_str(&svg);
-                if out.len() > HTML_EXPORT_MAX_BYTES {
-                    return Err(html_too_big());
-                }
             }
         }
         out.push_str("</div>");
     }
     out.push_str("</body></html>");
-    if out.len() > HTML_EXPORT_MAX_BYTES {
-        return Err(html_too_big());
-    }
     Ok(out)
 }
 

crates/engine/src/io/tests.rs

@@ -604,57 +604,3 @@ fn opening_google_sheets_export_unwraps_only_formulas_gridcraft_calculates() {
     assert_eq!(text("A2").as_deref(), Some("IFERROR(__XLUDF.DUMMYFUNCTION(\"GOOGLETRANSLATE(\"\"hi\"\",\"\"en\"\",\"\"fr\"\")\"),\"salut\")"));
     assert_eq!(opened.sheets[0].cells.get(CellRef::parse("A2").unwrap()).map(|c| c.value.clone()), Some(Value::text("salut")));
 }
-
-/// Merges far from what is copied or exported, in a hostile number, are indexed once instead of
-/// being scanned for every cell; a merge running past the table is clipped to it.
-#[test]
-fn html_merge_lookups_and_spans_stay_bounded() {
-    let mut wb = book_with_picture("A1", (0, 0));
-    let sh = wb.sheet_mut(0).unwrap();
-    for i in 0..200_000u32 {
-        sh.merges.push(RangeRef::new(CellRef::new(500_000 + i * 2, 5), CellRef::new(500_000 + i * 2, 6)));
-    }
-    sh.merges.push(RangeRef::new(CellRef::new(0, 0), CellRef::new(gridcraft_core::MAX_ROWS - 1, 1)));
-    sh.set_value(CellRef::new(2, 1), gridcraft_core::Value::Number(2.0));
-    let html = copied(&wb, "A1:C50");
-    assert!(html.contains("rowspan=\"50\" colspan=\"2\""), "{html}");
-    let page = to_html(&wb, 0).unwrap();
-    // The table is A1:B3 (the used range): the whole-height merge spans its three rows only.
-    assert!(page.contains("rowspan=\"3\" colspan=\"2\""), "{page}");
-    // Merges stacked over one another past a few per cell are refused, not walked.
-    let sh = wb.sheet_mut(0).unwrap();
-    sh.merges = vec![RangeRef::new(CellRef::new(0, 0), CellRef::new(2, 1)); 100];
-    assert!(to_html(&wb, 0).is_err());
-    assert!(range_html(&wb, 0, RangeRef::parse("A1:B3").unwrap(), "").is_none());
-}
-
-/// Thousands of pictures over thousands of custom rows: each is placed through a row index.
-#[test]
-fn html_export_places_many_pictures_over_many_custom_rows() {
-    let mut wb = book_with_picture("A1", (0, 0));
-    let sh = wb.sheet_mut(0).unwrap();
-    for row in 0..50_000 {
-        sh.rows.insert(row, gridcraft_model::LineInfo { size: Some(30.0), ..Default::default() });
-    }
-    let image = sh.images[0].clone();
-    for i in 0..5_000u32 {
-        let mut im = image.clone();
-        im.anchor.cell = CellRef::new(40_000 + i, 0);
-        sh.images.push(im);
-    }
-    let page = to_html(&wb, 0).unwrap();
-    assert!(page.contains(&format!("top:{}px", 30 * 44_999)), "last picture placed by the index");
-}
-
-/// One long text shown by many cells: the page fails at its byte limit instead of growing on.
-#[test]
-fn html_export_refuses_a_page_past_its_byte_limit() {
-    let mut wb = Workbook::new();
-    let sh = wb.sheet_mut(0).unwrap();
-    let text = gridcraft_core::Value::text("x".repeat(32_000));
-    for row in 0..10_000 {
-        sh.set_value(CellRef::new(row, 0), text.clone());
-    }
-    let err = to_html(&wb, 0).unwrap_err();
-    assert!(err.to_string().contains("MiB"), "{err}");
-}

crates/engine/src/macros/recorder.rs

@@ -440,7 +440,7 @@ pub fn infer_input(wb: &Workbook, sheet: usize, r: RangeRef) -> NamedInput {
         Some(match v {
             Value::Empty => return None,
             Value::Text(t) if t.is_empty() => return None,
-            Value::Number(_) if crate::display::number_format(wb.styles.get(sh.style_id(c)).num_fmt.as_str()).is_date() => ColType::Date,
+            Value::Number(_) if crate::display::parsed(wb.styles.get(sh.style_id(c)).num_fmt.as_str()).is_date() => ColType::Date,
             Value::Number(_) => ColType::Number,
             Value::Text(_) => ColType::Text,
             Value::Bool(_) => ColType::Bool,

crates/engine/src/macros/run.rs

@@ -335,7 +335,7 @@ fn check_input(wb: &Workbook, input: &gridcraft_model::NamedInput, place: Place)
                 gridcraft_model::ColType::Number => v.is_number(),
                 gridcraft_model::ColType::Text => v.is_text(),
                 gridcraft_model::ColType::Bool => matches!(v, Value::Bool(_)),
-                gridcraft_model::ColType::Date => v.is_number() && crate::display::number_format(&date()).is_date(),
+                gridcraft_model::ColType::Date => v.is_number() && crate::display::parsed(&date()).is_date(),
             };
             if !ok {
                 wrong.push((c, holds(&v, ty)));

crates/engine/src/pivot.rs

@@ -182,7 +182,7 @@ pub fn read_source(wb: &Workbook, text: &str, default_sheet: usize) -> Result<So
     for (n, a) in numeric.iter_mut().zip(any.iter()) {
         *n = *n && *a;
     }
-    let dates = numeric.iter().zip(formats.iter()).map(|(n, f)| *n && crate::display::number_format(f).is_date()).collect();
+    let dates = numeric.iter().zip(formats.iter()).map(|(n, f)| *n && crate::display::parsed(f).is_date()).collect();
     Ok(Source { sheet: si, range: r, headers, formats, numeric, dates, rows })
 }