engine: PivotTable filters test each record against a set of selected labels (#3279)

Every source record scanned the filter's whole selection, so a large
selection multiplied by many records was slow even when nothing passed.
Each filter's labels are lowercased into a set once per computation.

crates/engine/src/cmd/pivot/tests.rs

@@ -168,6 +168,19 @@ fn filters_and_report_filter_rows() {
     assert!(s.execute("pivot.filter", json!({"field": "Product", "selected": []})).is_err());
 }
 
+/// A selection of the most labels allowed (100,000) is looked up once per record, not scanned: the
+/// report still filters case-insensitively.
+#[test]
+fn a_huge_filter_selection_is_matched_through_a_set() {
+    let mut s = sales();
+    s.execute("insert.pivotTable", json!({"rows": ["Product"], "values": ["Sales"], "filters": ["Region"]})).unwrap();
+    let mut selected: Vec<String> = (0..99_999).map(|i| format!("Region {i}")).collect();
+    selected.push("EAST".into());
+    s.execute("pivot.filter", json!({"field": "Region", "selected": selected})).unwrap();
+    let gt = find_row(&s, "Grand Total").unwrap();
+    assert_eq!(val(&s, &CellRef::new(gt, 1).a1()), Value::Number(415.0));
+}
+
 #[test]
 fn show_values_as() {
     let mut s = sales();

crates/engine/src/cmd/print/tests.rs

@@ -695,6 +695,79 @@ fn bands_unit() {
     assert!(w.len() > 2);
 }
 
+/// The wrapper `wrap_lines` replaced, re-measuring strings, as the reference for its lines.
+fn wrap_lines_remeasuring(text: &str, font: Font, size: f32, width: f32) -> Vec<String> {
+    let mut out = Vec::new();
+    for para in text.split('\n') {
+        let mut line = String::new();
+        for word in para.split(' ') {
+            let cand = if line.is_empty() { word.to_string() } else { format!("{line} {word}") };
+            if line.is_empty() || gridcraft_pdf::text_width(font, size, &cand) <= width {
+                line = cand;
+            } else {
+                out.push(std::mem::take(&mut line));
+                line = word.to_string();
+            }
+            while gridcraft_pdf::text_width(font, size, &line) > width && line.chars().count() > 1 {
+                let chars: Vec<char> = line.chars().collect();
+                let (mut lo, mut hi) = (1usize, chars.len());
+                while lo < hi {
+                    let mid = lo + (hi - lo).div_ceil(2);
+                    let cut: String = chars[..mid].iter().collect();
+                    if gridcraft_pdf::text_width(font, size, &cut) <= width {
+                        lo = mid;
+                    } else {
+                        hi = mid - 1;
+                    }
+                }
+                if lo == chars.len() {
+                    break;
+                }
+                out.push(chars[..lo].iter().collect());
+                line = chars[lo..].iter().collect();
+            }
+            if out.len() > 2000 {
+                return out;
+            }
+        }
+        out.push(line);
+    }
+    out
+}
+
+#[test]
+fn wrap_lines_matches_the_remeasuring_wrapper() {
+    let texts = [
+        "the quick brown fox jumps",
+        "Supercalifragilistic",
+        "a  b   c\n\nverylongwordthatneedsbreaking short\nend ",
+        "Wide WWWWWWWWWW words, iiiiiiii narrow ones; mixed MMMMiiiiMMMM",
+        "€ ü ß — unicode Ünïcödé text",
+        "",
+        " ",
+    ];
+    for text in texts {
+        for font in [Font::Helvetica, Font::HelveticaBold, Font::Courier] {
+            for (size, width) in [(10.0, 50.0), (10.0, 20.0), (11.0, 1.0), (8.0, 300.0), (12.0, 0.0)] {
+                assert_eq!(wrap_lines(text, font, size, width), wrap_lines_remeasuring(text, font, size, width), "{text:?} {font:?} {size} {width}");
+            }
+        }
+    }
+}
+
+#[test]
+fn wrapping_a_long_word_in_a_narrow_box_stops_at_the_line_limit() {
+    // One character per line: the old wrapper re-measured and copied the rest of the word for
+    // each of the 100,000 lines before checking the limit.
+    let word = "w".repeat(100_000);
+    let lines = wrap_lines(&word, Font::Helvetica, 10.0, 1.0);
+    assert!(lines.len() <= 2002, "{}", lines.len());
+    assert!(lines.iter().all(|l| l == "w"));
+    // A long paragraph in a very wide box is one line.
+    let para = vec!["word"; 50_000].join(" ");
+    assert_eq!(wrap_lines(&para, Font::Helvetica, 10.0, 1.0e9), vec![para.clone()]);
+}
+
 #[test]
 fn center_across_spans_following_empty_center_across_cells() {
     let mut styles = StyleTable::default();

crates/engine/src/cmd/print/text.rs

@@ -3,42 +3,57 @@ use super::{CellRef, Color, Font, HAlign, MAX_COLS, Page, Point, RangeRef, Sheet
 
 const PAD_PX: f32 = 2.0;
 
+/// Most lines [`wrap_lines`] returns (it may return one more, the rest of a paragraph).
+const MAX_WRAP_LINES: usize = 2000;
+
 /// Word-wraps `text` to `width` points.
+///
+/// Widths are kept as running sums of glyph widths (the sum `text_width` makes), so each
+/// character is measured once: re-measuring the line for every word, or the rest of an
+/// over-long word for every piece cut from it, made a long paragraph or word quadratic.
 pub(super) fn wrap_lines(text: &str, font: Font, size: f32, width: f32) -> Vec<String> {
+    let units_of = |s: &str| -> u64 { s.chars().map(|c| u64::from(font.char_width(c))).sum() };
+    let fits = |units: u64| units as f32 * size / 1000.0 <= width;
+    let space = units_of(" ");
     let mut out = Vec::new();
     for para in text.split('\n') {
         let mut line = String::new();
+        let mut line_units: u64 = 0;
         for word in para.split(' ') {
-            let cand = if line.is_empty() { word.to_string() } else { format!("{line} {word}") };
-            if line.is_empty() || gridcraft_pdf::text_width(font, size, &cand) <= width {
-                line = cand;
+            let word_units = units_of(word);
+            if line.is_empty() {
+                line.push_str(word);
+                line_units = word_units;
+            } else if fits(line_units + space + word_units) {
+                line.push(' ');
+                line.push_str(word);
+                line_units += space + word_units;
             } else {
                 out.push(std::mem::take(&mut line));
-                line = word.to_string();
+                line.push_str(word);
+                line_units = word_units;
             }
-            // Break a single over-long word by characters. The cut that fits is found by binary
-            // search (each try measures one candidate prefix), so an over-long word costs
-            // O(n log n) measurements instead of re-measuring the whole cut per character for
-            // every emitted line (`a` × 100k pasted text would be cubic worst case).
-            while gridcraft_pdf::text_width(font, size, &line) > width && line.chars().count() > 1 {
-                let chars: Vec<char> = line.chars().collect();
-                let (mut lo, mut hi) = (1usize, chars.len());
-                while lo < hi {
-                    let mid = lo + (hi - lo).div_ceil(2);
-                    let cut: String = chars[..mid].iter().collect();
-                    if gridcraft_pdf::text_width(font, size, &cut) <= width {
-                        lo = mid;
-                    } else {
-                        hi = mid - 1;
+            // Break a single over-long word by characters: each piece is the longest that fits
+            // (at least one character).
+            if !fits(line_units) && line.chars().nth(1).is_some() {
+                let mut piece = String::new();
+                let mut units: u64 = 0;
+                for c in line.chars() {
+                    let w = u64::from(font.char_width(c));
+                    if !piece.is_empty() && !fits(units + w) {
+                        out.push(std::mem::take(&mut piece));
+                        units = 0;
+                        if out.len() > MAX_WRAP_LINES {
+                            return out;
+                        }
                     }
+                    piece.push(c);
+                    units += w;
                 }
-                if lo == chars.len() {
-                    break;
-                }
-                out.push(chars[..lo].iter().collect());
-                line = chars[lo..].iter().collect();
+                line = piece;
+                line_units = units;
             }
-            if out.len() > 2000 {
+            if out.len() > MAX_WRAP_LINES {
                 return out;
             }
         }

crates/engine/src/pivot/compute.rs

@@ -77,6 +77,8 @@ struct AggregateInput<'a> {
     extra: &'a [usize],
     filters: &'a [(usize, &'a PivotFilter, PivotDateGroup, bool)],
     filter_fmts: &'a [String],
+    /// Each filter's selected labels, ASCII-lowercased, so a record's test is one probe.
+    filter_sets: &'a [Option<HashSet<String>>],
     row_fmts: &'a [String],
     col_fmts: &'a [String],
     empty: bool,
@@ -99,10 +101,10 @@ fn aggregate_record(
     ri: usize,
     rec: &[Value],
 ) -> Result<(), String> {
-    for ((c, field, group, _), fmt) in input.filters.iter().zip(input.filter_fmts) {
-        if let Some(selected) = &field.selected {
+    for (((c, _, group, _), fmt), selected) in input.filters.iter().zip(input.filter_fmts).zip(input.filter_sets) {
+        if let Some(selected) = selected {
             let label = item_label(input.wb, rec.get(*c).unwrap_or(&Value::Empty), *group, fmt);
-            if !selected.iter().any(|value| value.eq_ignore_ascii_case(&label)) {
+            if !selected.contains(&label.to_ascii_lowercase()) {
                 return Ok(());
             }
         }
@@ -475,6 +477,8 @@ pub(super) fn compute_from(wb: &Workbook, pt: &PivotTable, src: &Source, mut pro
 
     // ---- aggregate
     let filter_fmts: Vec<String> = filters.iter().map(|f| fmt_of(f.0)).collect();
+    let filter_sets: Vec<Option<HashSet<String>>> =
+        filters.iter().map(|(_, field, _, _)| field.selected.as_ref().map(|s| s.iter().map(|v| v.to_ascii_lowercase()).collect())).collect();
     let row_fmts: Vec<String> = rows.iter().map(|(c, _)| fmt_of(*c)).collect();
     let col_fmts: Vec<String> = cols.iter().map(|(c, _)| fmt_of(*c)).collect();
     let aggregate = aggregate_source(
@@ -488,6 +492,7 @@ pub(super) fn compute_from(wb: &Workbook, pt: &PivotTable, src: &Source, mut pro
             extra: &extra,
             filters: &filters,
             filter_fmts: &filter_fmts,
+            filter_sets: &filter_sets,
             row_fmts: &row_fmts,
             col_fmts: &col_fmts,
             empty,