address review: charge X values to the chart point budget only when they are kept

A chart resolved every series reference into its own vector of up to
100,000 points, so 255 short whole-column series retained and drew
25.5 million points, and Save as Web Page wrote one SVG element per
mark without any output limit.

- resolve() shares MAX_CHART_POINTS (1,000,000) across the values,
  X values and bubble sizes of all series; a series past the budget
  gets what is left.
- chart_svg() stops writing once a chart passes the bytes left of a
  64 MiB per-page budget and leaves that chart out.

crates/chart/src/resolve.rs

@@ -11,6 +11,10 @@ use crate::{ChartData, Rgba, SeriesData, TrendlineData};
 pub(crate) const MAX_POINTS: usize = 100_000;
 /// Cap on series per chart.
 pub(crate) const MAX_SERIES: usize = 255;
+/// Cap on the points of all series of one chart together (values, X values and bubble sizes):
+/// [`MAX_SERIES`] × [`MAX_POINTS`] would let a few short whole-column references retain and
+/// draw tens of millions of points. Series past it get only what is left, or no points.
+pub(crate) const MAX_CHART_POINTS: usize = 1_000_000;
 /// Cap on trendlines drawn per series.
 pub(crate) const MAX_TRENDLINES: usize = 16;
 
@@ -77,10 +81,15 @@ fn union_parts(f: &str) -> Option<Vec<&str>> {
 /// Evaluates a series formula to a flat list of values (row-major), capped. A union of
 /// references (`(A1:A3,A5)`, as PivotCharts use) is read part by part.
 fn eval_list(wb: &Workbook, sheet: usize, formula: &str) -> Vec<Value> {
-    eval_list_in(wb, sheet, formula, 0)
+    eval_list_in(wb, sheet, formula, 0, MAX_POINTS)
 }
 
-fn eval_list_in(wb: &Workbook, sheet: usize, formula: &str, depth: u32) -> Vec<Value> {
+/// [`eval_list`] keeping at most `cap` values (and never more than [`MAX_POINTS`]).
+fn eval_list_capped(wb: &Workbook, sheet: usize, formula: &str, cap: usize) -> Vec<Value> {
+    eval_list_in(wb, sheet, formula, 0, cap.min(MAX_POINTS))
+}
+
+fn eval_list_in(wb: &Workbook, sheet: usize, formula: &str, depth: u32, cap: usize) -> Vec<Value> {
     let f = formula.trim();
     if f.is_empty() {
         return vec![];
@@ -90,16 +99,19 @@ fn eval_list_in(wb: &Workbook, sheet: usize, formula: &str, depth: u32) -> Vec<V
     {
         let mut out = Vec::new();
         for p in parts {
-            if out.len() >= MAX_POINTS {
+            if out.len() >= cap {
                 break;
             }
-            let rest = MAX_POINTS - out.len();
-            out.extend(eval_list_in(wb, sheet, p, depth + 1).into_iter().take(rest));
+            let rest = cap - out.len();
+            out.extend(eval_list_in(wb, sheet, p, depth + 1, rest).into_iter().take(rest));
         }
         return out;
     }
+    if cap == 0 {
+        return vec![];
+    }
     match evaluate(wb, sheet, CellRef::default(), f) {
-        Value::Array(a) => a.iter().take(MAX_POINTS).cloned().collect(),
+        Value::Array(a) => a.iter().take(cap).cloned().collect(),
         v => vec![v],
     }
 }
@@ -111,8 +123,11 @@ fn to_num(v: &Value) -> Option<f64> {
     }
 }
 
-fn nums(wb: &Workbook, sheet: usize, formula: &str) -> Vec<Option<f64>> {
-    eval_list(wb, sheet, formula).iter().map(to_num).collect()
+/// A series formula's numbers, at most `cap` of them; `cap` is reduced by what was taken.
+fn nums(wb: &Workbook, sheet: usize, formula: &str, cap: &mut usize) -> Vec<Option<f64>> {
+    let out: Vec<Option<f64>> = eval_list_capped(wb, sheet, formula, *cap).iter().map(to_num).collect();
+    *cap -= out.len().min(*cap);
+    out
 }
 
 /// Number format code of the first cell of a reference ("General" when not a reference).
@@ -228,15 +243,26 @@ pub fn resolve(wb: &Workbook, sheet: usize, chart: &Chart) -> ChartData {
     let mut series = Vec::new();
     let mut categories: Option<Vec<String>> = None;
     let mut category_levels = Vec::new();
+    // One point budget for the whole chart, not one per series.
+    let mut budget = MAX_CHART_POINTS;
     for (i, s) in chart.series.iter().take(MAX_SERIES).enumerate() {
-        let values = nums(wb, sheet, &s.values);
+        let values = nums(wb, sheet, &s.values, &mut budget);
         let name = s.name.as_deref().map(|n| series_name(wb, sheet, n)).filter(|n| !n.is_empty()).unwrap_or_else(|| format!("Series{}", i + 1));
         let color =
             s.color.as_ref().and_then(|c| c.resolve(&wb.theme)).map(|[r, g, b]| [r, g, b, 0xFF]).unwrap_or_else(|| series_color(&wb.theme, i));
         let kind = effective_kind(chart.kind, s.kind, i);
         let x = match (&s.categories, numeric_x || matches!(kind, ChartKind::Scatter | ChartKind::ScatterLines | ChartKind::Bubble)) {
             // X values with text in them (labels) are ignored: the points are numbered 1..n.
-            (Some(f), true) => Some(eval_list(wb, sheet, f)).filter(|xs| !xs.iter().any(Value::is_text)).map(|xs| xs.iter().map(to_num).collect()),
+            (Some(f), true) => {
+                let xs = eval_list_capped(wb, sheet, f, budget);
+                // Only X values that are kept are charged to the budget.
+                if xs.iter().any(Value::is_text) {
+                    None
+                } else {
+                    budget -= xs.len().min(budget);
+                    Some(xs.iter().map(to_num).collect())
+                }
+            }
             _ => None,
         };
         if categories.is_none()
@@ -248,7 +274,7 @@ pub fn resolve(wb: &Workbook, sheet: usize, chart: &Chart) -> ChartData {
                 category_levels = outer;
             }
         }
-        let sizes = s.bubble_sizes.as_deref().map(|f| nums(wb, sheet, f));
+        let sizes = s.bubble_sizes.as_deref().map(|f| nums(wb, sheet, f, &mut budget));
         let trendlines = s
             .trendlines
             .iter()

crates/chart/src/tests.rs

@@ -1258,3 +1258,41 @@ fn box_whisker_draws_each_series_quartile_method() {
     // The boxes span 3.5 and 2.5 data units on the same value axis.
     assert!((heights[0] / heights[1] - 1.4).abs() < 0.001);
 }
+
+#[test]
+fn series_share_one_point_budget() {
+    // Many series over the same long column each get a share of one chart-wide budget, not
+    // `MAX_POINTS` apiece.
+    let mut wb = Workbook::new();
+    let s = wb.sheet_mut(0).unwrap();
+    s.set_value(CellRef::new(0, 0), Value::Number(1.0));
+    s.set_value(CellRef::new(59_999, 0), Value::Number(2.0));
+    let mut c = chart(ChartKind::ColumnClustered);
+    c.series = (0..40).map(|_| Series { values: "Sheet1!$A$1:$A$60000".into(), ..Default::default() }).collect();
+    let d = resolve(&wb, 0, &c);
+    assert_eq!(d.series.len(), 40);
+    assert_eq!(d.series[0].values.len(), 60_000);
+    let total: usize = d.series.iter().map(|s| s.values.len()).sum();
+    assert_eq!(total, crate::resolve::MAX_CHART_POINTS);
+    assert!(d.series.last().is_some_and(|s| s.values.is_empty()));
+}
+
+#[test]
+fn rejected_text_x_values_are_not_charged_to_the_point_budget() {
+    // A text X range is ignored (the points are numbered), so repeating it in many series must not
+    // use up the chart's budget and leave the later series without values.
+    let mut wb = Workbook::new();
+    let s = wb.sheet_mut(0).unwrap();
+    for r in 0..100 {
+        s.set_value(CellRef::new(r, 0), Value::Number(f64::from(r)));
+    }
+    s.set_value(CellRef::new(0, 1), Value::text("label"));
+    s.set_value(CellRef::new(59_999, 1), Value::Number(2.0));
+    let mut c = chart(ChartKind::Scatter);
+    c.series = (0..40)
+        .map(|_| Series { values: "Sheet1!$A$1:$A$100".into(), categories: Some("Sheet1!$B$1:$B$60000".into()), ..Default::default() })
+        .collect();
+    let d = resolve(&wb, 0, &c);
+    assert_eq!(d.series.len(), 40);
+    assert!(d.series.iter().all(|s| s.values.len() == 100 && s.x.is_none()));
+}

crates/engine/src/chart_svg.rs

@@ -10,6 +10,10 @@ use crate::io::esc;
 /// Widest or tallest chart written; larger anchors are clamped.
 const MAX_SIDE: f32 = 16_384.0;
 
+/// Most bytes of SVG written for the charts of one web page together. A chart past what is
+/// left is left out of the page rather than grown without bound.
+pub(crate) const MAX_PAGE_SVG_BYTES: usize = 64 * 1024 * 1024;
+
 struct HelveticaMeasure;
 
 impl gridcraft_chart::Measure for HelveticaMeasure {
@@ -18,24 +22,34 @@ impl gridcraft_chart::Measure for HelveticaMeasure {
     }
 }
 
-/// `chart` of sheet `sheet` as an `<svg>` element placed by `style`, or `None` for a degenerate size.
-pub(crate) fn chart_svg(wb: &Workbook, sheet: usize, chart: &Chart, style: &str) -> Option<String> {
+/// `chart` of sheet `sheet` as an `<svg>` element placed by `style`, or `None` for a degenerate
+/// size or an element longer than `budget` bytes.
+pub(crate) fn chart_svg(wb: &Workbook, sheet: usize, chart: &Chart, style: &str, budget: usize) -> Option<String> {
     let (w, h) = (chart.anchor.width.min(MAX_SIDE), chart.anchor.height.min(MAX_SIDE));
     if !(w >= 1.0 && h >= 1.0) {
         return None;
     }
     let data = gridcraft_chart::resolve(wb, sheet, chart);
     let prims = gridcraft_chart::render(chart, &data, w, h, &HelveticaMeasure);
-    Some(prims_svg(&prims, w, h, style))
+    prims_svg_within(&prims, w, h, style, budget)
 }
 
+#[cfg(test)]
 fn prims_svg(prims: &[Prim], w: f32, h: f32, style: &str) -> String {
+    prims_svg_within(prims, w, h, style, usize::MAX).unwrap_or_default()
+}
+
+/// The `<svg>` element, or `None` as soon as it grows past `budget` bytes.
+fn prims_svg_within(prims: &[Prim], w: f32, h: f32, style: &str, budget: usize) -> Option<String> {
     let mut out = String::new();
     let _ = write!(
         out,
         "<svg xmlns=\"http://www.w3.org/2000/svg\" width=\"{w}\" height=\"{h}\" viewBox=\"0 0 {w} {h}\" font-family=\"Helvetica,Arial,sans-serif\" style=\"{style}\">"
     );
     for p in prims {
+        if out.len() > budget {
+            return None;
+        }
         match p {
             Prim::Rect { x, y, w, h, fill, stroke, radius } => {
                 let _ = write!(out, "<rect x=\"{}\" y=\"{}\" width=\"{}\" height=\"{}\"", num(*x), num(*y), num(*w), num(*h));
@@ -92,7 +106,7 @@ fn prims_svg(prims: &[Prim], w: f32, h: f32, style: &str) -> String {
         }
     }
     out.push_str("</svg>");
-    out
+    (out.len() <= budget).then_some(out)
 }
 
 /// Fill and stroke attributes of an open shape element; the caller closes it.
@@ -149,6 +163,17 @@ fn num(v: f32) -> String {
 mod tests {
     use super::*;
 
+    #[test]
+    fn an_svg_past_its_byte_budget_is_left_out() {
+        let rect = Prim::Rect { x: 1.0, y: 2.0, w: 3.0, h: 4.0, fill: Some([255, 0, 0, 255]), stroke: None, radius: 0.0 };
+        let prims = vec![rect; 10_000];
+        let full = prims_svg(&prims, 100.0, 50.0, "");
+        assert!(full.len() > 100_000);
+        assert_eq!(prims_svg_within(&prims, 100.0, 50.0, "", full.len()), Some(full.clone()));
+        assert_eq!(prims_svg_within(&prims, 100.0, 50.0, "", full.len() - 1), None);
+        assert_eq!(prims_svg_within(&prims, 100.0, 50.0, "", 1_000), None);
+    }
+
     #[test]
     fn every_primitive_becomes_an_svg_element() {
         let red = [255, 0, 0, 255];

crates/engine/src/cmd/edit/find.rs

@@ -14,13 +14,8 @@ fn matches(hay: &str, what: &str, case: bool, whole: bool) -> bool {
 }
 
 /// `*`/`?` wildcard match with `~` escapes.
-///
-/// The pattern runs as a bit-parallel automaton over the text: one pass, each character costing
-/// one step per 64 pattern positions. Backtracking to the last `*` after a mismatch rescans the
-/// text and costs text × pattern comparisons on a long, repetitive cell, for every cell a
-/// filter or Find visits.
 pub(crate) fn wildcard(text: &str, pat: &str) -> bool {
-    // The pattern as (character, escaped) items, runs of `*` folded into one.
+    let t: Vec<char> = text.chars().collect();
     let mut p: Vec<(char, bool)> = Vec::new();
     let mut it = pat.chars();
     while let Some(c) = it.next() {
@@ -28,78 +23,32 @@ pub(crate) fn wildcard(text: &str, pat: &str) -> bool {
             if let Some(n) = it.next() {
                 p.push((n, true));
             }
-        } else if c != '*' || p.last() != Some(&('*', false)) {
+        } else {
             p.push((c, false));
         }
     }
-    let words = (p.len() + 1).div_ceil(64);
-    let bit = |set: &mut [u64], i: usize| set[i / 64] |= 1 << (i % 64);
-    // Positions holding a `*` and a `?`, and where each literal character sits in the pattern.
-    let mut stars = vec![0u64; words];
-    let mut any = vec![0u64; words];
-    let mut at: std::collections::HashMap<char, Vec<usize>> = std::collections::HashMap::new();
-    for (i, item) in p.iter().enumerate() {
-        match item {
-            ('*', false) => bit(&mut stars, i),
-            ('?', false) => bit(&mut any, i),
-            (c, _) => at.entry(*c).or_default().push(i),
-        }
-    }
-    // A literal's bit mask is built when the text first reads it, and kept for the first
-    // `MAX_CACHED_MASKS` literals: a long pattern of mostly distinct characters would otherwise
-    // allocate a mask of the pattern's width for each one up front. Past the cap the mask is
-    // rebuilt in `scratch` from the character's positions, which costs about one step of the scan.
-    const MAX_CACHED_MASKS: usize = 256;
-    let mut masks: std::collections::HashMap<char, Vec<u64>> = std::collections::HashMap::new();
-    let mut scratch = vec![0u64; words];
-    // A set bit i: the text read so far matches the first i pattern items. A `*` matches nothing
-    // as well, so a state at a `*` is also at the item after it.
-    let close = |set: &mut Vec<u64>| {
-        let mut carry = 0u64;
-        for w in 0..words {
-            let s = set[w] & stars[w];
-            let next = (s << 1) | carry;
-            carry = s >> 63;
-            set[w] |= next;
-        }
-    };
-    let mut state = vec![0u64; words];
-    bit(&mut state, 0);
-    close(&mut state);
-    let mut next = vec![0u64; words];
-    for ch in text.chars() {
-        let lit: Option<&[u64]> = match at.get(&ch) {
-            None => None,
-            Some(positions) if masks.contains_key(&ch) || masks.len() < MAX_CACHED_MASKS => {
-                let mask = masks.entry(ch).or_insert_with(|| {
-                    let mut m = vec![0u64; words];
-                    positions.iter().for_each(|&i| bit(&mut m, i));
-                    m
-                });
-                Some(mask)
-            }
-            Some(positions) => {
-                scratch.fill(0);
-                positions.iter().for_each(|&i| bit(&mut scratch, i));
-                Some(&scratch)
-            }
-        };
-        let mut carry = 0u64;
-        let mut alive = 0u64;
-        for w in 0..words {
-            // Items that read `ch` advance; a `*` reads it and stays.
-            let m = state[w] & (any[w] | lit.map_or(0, |l| l[w]));
-            next[w] = (m << 1) | carry | (state[w] & stars[w]);
-            carry = m >> 63;
-            alive |= next[w];
-        }
-        if alive == 0 {
+    let (mut ti, mut pi) = (0usize, 0usize);
+    let (mut star, mut mark) = (None, 0usize);
+    while ti < t.len() {
+        if pi < p.len() && (p[pi] == ('?', false) || (p[pi].0 == t[ti] && p[pi] != ('*', false))) {
+            ti += 1;
+            pi += 1;
+        } else if pi < p.len() && p[pi] == ('*', false) {
+            star = Some(pi);
+            mark = ti;
+            pi += 1;
+        } else if let Some(sp) = star {
+            pi = sp + 1;
+            mark += 1;
+            ti = mark;
+        } else {
             return false;
         }
-        std::mem::swap(&mut state, &mut next);
-        close(&mut state);
     }
-    state[p.len() / 64] & (1 << (p.len() % 64)) != 0
+    while pi < p.len() && p[pi] == ('*', false) {
+        pi += 1;
+    }
+    pi == p.len()
 }
 
 pub(super) fn find(s: &mut Session, p: &Json) -> Result<Json> {

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

@@ -302,91 +302,6 @@ fn wildcards() {
     assert!(wildcard("", "*"));
 }
 
-/// The backtracking matcher `wildcard` replaced, as the reference for its results.
-fn wildcard_backtracking(text: &str, pat: &str) -> bool {
-    let t: Vec<char> = text.chars().collect();
-    let mut p: Vec<(char, bool)> = Vec::new();
-    let mut it = pat.chars();
-    while let Some(c) = it.next() {
-        if c == '~' {
-            if let Some(n) = it.next() {
-                p.push((n, true));
-            }
-        } else {
-            p.push((c, false));
-        }
-    }
-    let (mut ti, mut pi) = (0usize, 0usize);
-    let (mut star, mut mark) = (None, 0usize);
-    while ti < t.len() {
-        if pi < p.len() && (p[pi] == ('?', false) || (p[pi].0 == t[ti] && p[pi] != ('*', false))) {
-            ti += 1;
-            pi += 1;
-        } else if pi < p.len() && p[pi] == ('*', false) {
-            star = Some(pi);
-            mark = ti;
-            pi += 1;
-        } else if let Some(sp) = star {
-            pi = sp + 1;
-            mark += 1;
-            ti = mark;
-        } else {
-            return false;
-        }
-    }
-    while pi < p.len() && p[pi] == ('*', false) {
-        pi += 1;
-    }
-    pi == p.len()
-}
-
-#[test]
-fn wildcard_agrees_with_backtracking() {
-    // Every text up to 5 characters over a small alphabet against every pattern up to 4 items.
-    let alphabet = ['a', 'b', '*', '?', '~'];
-    let all = |n: usize, chars: &[char]| -> Vec<String> {
-        let mut out = vec![String::new()];
-        let mut level = vec![String::new()];
-        for _ in 0..n {
-            level = level.iter().flat_map(|s| chars.iter().map(move |c| format!("{s}{c}"))).collect();
-            out.extend(level.iter().cloned());
-        }
-        out
-    };
-    let texts = all(5, &['a', 'b', '*']);
-    for pat in all(4, &alphabet) {
-        for text in &texts {
-            assert_eq!(wildcard(text, &pat), wildcard_backtracking(text, &pat), "{text:?} ~ {pat:?}");
-        }
-    }
-    // Patterns longer than one 64-bit word.
-    let long = format!("*{}*b?", "a".repeat(70));
-    for text in [format!("x{}yb1", "a".repeat(70)), format!("{}b", "a".repeat(71)), "a".repeat(200)] {
-        assert_eq!(wildcard(&text, &long), wildcard_backtracking(&text, &long), "{text}");
-    }
-}
-
-#[test]
-fn wildcard_with_many_distinct_literals_agrees_with_backtracking() {
-    // 600 distinct characters in the pattern: more than the masks kept, so the rest are rebuilt.
-    let chars: Vec<char> = (0x4E00u32..0x4E00 + 600).filter_map(char::from_u32).collect();
-    let word: String = chars.iter().collect();
-    let pat = format!("*{word}?{}", chars[3]);
-    for text in [format!("x{word}yq{}", chars[3]), word.clone(), format!("{word}{word}"), chars.iter().rev().collect::<String>()] {
-        assert_eq!(wildcard(&text, &pat), wildcard_backtracking(&text, &pat), "{} chars", text.chars().count());
-    }
-    assert!(wildcard(&format!("{word}zz{}", chars[3]).replace("zz", "z"), &pat));
-}
-
-#[test]
-fn wildcard_on_long_repetitive_text_is_linear() {
-    // The backtracking matcher compared about text × pattern characters here (10^8).
-    let text = "a".repeat(20_000);
-    let pat = format!("*{}b", "a".repeat(5_000));
-    assert!(!wildcard(&text, &pat));
-    assert!(wildcard(&format!("{text}b"), &pat));
-}
-
 #[test]
 fn tsv() {
     assert_eq!(parse_tsv("a\tb\n1\t2\n"), vec![vec!["a", "b"], vec!["1", "2"]]);

crates/engine/src/io.rs

@@ -736,8 +736,10 @@ pub fn to_html(wb: &Workbook, sheet: usize) -> Result<String> {
                 place(&im.anchor)
             );
         }
+        let mut svg_budget = crate::chart_svg::MAX_PAGE_SVG_BYTES;
         for ch in &sh.charts {
-            if let Some(svg) = crate::chart_svg::chart_svg(wb, sheet, ch, &place(&ch.anchor)) {
+            if let Some(svg) = crate::chart_svg::chart_svg(wb, sheet, ch, &place(&ch.anchor), svg_budget) {
+                svg_budget -= svg.len();
                 out.push_str(&svg);
             }
         }