fix: bound spreadsheet width and skip empty spill scans

CHANGELOG.md

@@ -16,6 +16,9 @@ The release run heads these entries with the version and opens a fresh
 
 ## Unreleased
 
+- Spreadsheet rendering bounds columns as well as rows by the cell budget and
+  avoids rescanning blank spans for every empty cell.
+
 - Browser text editors delete complete supplementary Unicode characters when
   the browser supplies no deletion range, preserving valid text and undo.
 

src/odr/html.hpp

@@ -173,7 +173,7 @@ struct HtmlConfig {
   /// Largest sheet region written, per axis; cells past it are dropped.
   std::optional<TableDimensions> spreadsheet_limit{
       TableDimensions(100000, 500)};
-  /// Most cells written for one sheet; bounds the rows by the sheet's width.
+  /// Cell budget, at least one; caps columns first, then rows by that width.
   std::optional<std::uint64_t> spreadsheet_cell_limit{500000};
   /// Trim a sheet to the cells it uses before the limits above apply.
   bool spreadsheet_limit_by_content{true};

src/odr/internal/html/document_element.cpp

@@ -165,7 +165,11 @@ TableDimensions html::sheet_rendered_extent(const Sheet &sheet,
   }
   end_column = std::max(1u, end_column);
   if (config.spreadsheet_cell_limit) {
-    const std::uint64_t rows = *config.spreadsheet_cell_limit / end_column;
+    const std::uint64_t limit =
+        std::max<std::uint64_t>(1, *config.spreadsheet_cell_limit);
+    end_column =
+        static_cast<std::uint32_t>(std::min<std::uint64_t>(end_column, limit));
+    const std::uint64_t rows = limit / end_column;
     end_row = static_cast<std::uint32_t>(
         std::min<std::uint64_t>(end_row, std::max<std::uint64_t>(1, rows)));
   }
@@ -769,10 +773,9 @@ void html::translate_sheet(const Sheet &sheet, const WritingState &state) {
         // the bottom and would paint over the row below.
         cell_css += "overflow:hidden;";
       }
-      // Over the empty cells beside it, cut where the next one has something
-      // to show, unbounded where nothing follows. Each blank cell is walked by
-      // the one cell that may spill over it, so the row costs one pass.
-      if (!wraps && cuts_its_text && column_pixels[column_index].has_value()) {
+      // Only nonblank cells spill, so each blank span is scanned once.
+      if (!wraps && cuts_its_text && column_pixels[column_index].has_value() &&
+          !is_blank(cell)) {
         std::optional<double> spill(0);
         bool bounded = false;
         for (std::uint32_t ahead = next_column; ahead < end_column; ++ahead) {

test/src/html_test.cpp

@@ -1131,7 +1131,7 @@ TEST(html, an_editable_csv_states_that_it_takes_no_style) {
 }
 
 // The rows a sheet keeps follow how wide it turns out to be.
-TEST(html, the_cell_budget_bounds_the_rows_by_the_width) {
+TEST(html, the_cell_budget_bounds_columns_and_rows) {
   HtmlConfig config;
   config.spreadsheet_limit = TableDimensions(1000, 1000);
   config.spreadsheet_cell_limit = 60;
@@ -1153,6 +1153,12 @@ TEST(html, the_cell_budget_bounds_the_rows_by_the_width) {
   // the rectangle still caps a sheet the budget would let through
   config.spreadsheet_limit = TableDimensions(5, 1000);
   EXPECT_EQ(rendered(30, 3).rows, 5);
+  config.spreadsheet_limit.reset();
+  EXPECT_EQ(rendered(2, 80).columns, 60);
+  EXPECT_EQ(rendered(2, 80).rows, 1);
+  config.spreadsheet_cell_limit = 0;
+  EXPECT_EQ(rendered(2, 80).columns, 1);
+  EXPECT_EQ(rendered(2, 80).rows, 1);
 }
 
 // #822: an inline `style` is the one shape a browser cannot share across the
@@ -1622,6 +1628,7 @@ TEST(html, a_cell_spills_over_a_blank_neighbour_and_is_cut_by_a_full_one) {
   EXPECT_NE(spills.find("max-width:0;white-space:nowrap;"
                         "clip-path:inset(0 -96px 0 0)"),
             std::string::npos);
+  EXPECT_EQ(spills.find("nowrap;overflow:hidden"), std::string::npos);
 }
 
 // #822: and where nothing follows it at all, out onto the canvas — there is no