PPTX export: stop adding chart series once a group's caches pass a point budget

crates/pptx/src/custgeom.rs

@@ -15,10 +15,6 @@ use crate::xml::{A, El, W};
 
 const DEG: f64 = 60_000.0;
 
-/// Distinct path coordinate spaces whose guides are evaluated for one custom geometry; paths in
-/// further spaces are left out.
-const MAX_SPACES: usize = 16;
-
 /// Guide variables for one shape.
 pub struct Guides {
     vars: HashMap<String, f64>,
@@ -155,26 +151,14 @@ fn pt(g: &Guides, e: &El) -> (f64, f64) {
 pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
     let mut out = vec![];
     let Some(pl) = cg.child("pathLst") else { return out };
-    let mut spaces: HashMap<(u64, u64), Guides> = HashMap::new();
-    // The cap on cached spaces saves guide evaluation: a geometry without guides has none to save.
-    let guided = ["avLst", "gdLst"].iter().any(|n| cg.child(n).is_some_and(|l| l.elements().next().is_some()));
     for p in pl.children_named("path").take(1000) {
         let pw = p.f64("w").unwrap_or(0.0);
         let ph = p.f64("h").unwrap_or(0.0);
         // Guides are evaluated in the path's own space when it has one.
         let (gw, gh) = if pw > 0.0 || ph > 0.0 { (pw, ph) } else { (w_emu, h_emu) };
-        // The guide lists are evaluated once per coordinate space, not once per path.
-        let key = (gw.to_bits(), gh.to_bits());
-        if !spaces.contains_key(&key) {
-            if guided && spaces.len() >= MAX_SPACES {
-                continue;
-            }
-            let mut g = Guides::new(gw, gh);
-            g.eval_list(cg.child("avLst"));
-            g.eval_list(cg.child("gdLst"));
-            spaces.insert(key, g);
-        }
-        let Some(g) = spaces.get(&key) else { continue };
+        let mut g = Guides::new(gw, gh);
+        g.eval_list(cg.child("avLst"));
+        g.eval_list(cg.child("gdLst"));
         let mut d = String::new();
         let mut cur = (0.0, 0.0);
         let mut start = (0.0, 0.0);
@@ -182,7 +166,7 @@ pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
             match c.local() {
                 "moveTo" => {
                     if let Some(q) = c.child("pt") {
-                        cur = pt(g, q);
+                        cur = pt(&g, q);
                         start = cur;
                         d.push_str("M ");
                         num(&mut d, cur.0);
@@ -193,7 +177,7 @@ pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
                 }
                 "lnTo" => {
                     if let Some(q) = c.child("pt") {
-                        cur = pt(g, q);
+                        cur = pt(&g, q);
                         d.push_str("L ");
                         num(&mut d, cur.0);
                         d.push(' ');
@@ -202,7 +186,7 @@ pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
                     }
                 }
                 "cubicBezTo" => {
-                    let pts: Vec<(f64, f64)> = c.children_named("pt").map(|q| pt(g, q)).collect();
+                    let pts: Vec<(f64, f64)> = c.children_named("pt").map(|q| pt(&g, q)).collect();
                     if let [a, b, e] = pts.as_slice() {
                         d.push_str("C ");
                         for (x, y) in [a, b, e] {
@@ -215,7 +199,7 @@ pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
                     }
                 }
                 "quadBezTo" => {
-                    let pts: Vec<(f64, f64)> = c.children_named("pt").map(|q| pt(g, q)).collect();
+                    let pts: Vec<(f64, f64)> = c.children_named("pt").map(|q| pt(&g, q)).collect();
                     if let [a, e] = pts.as_slice() {
                         d.push_str("Q ");
                         for (x, y) in [a, e] {
@@ -486,29 +470,6 @@ mod tests {
         assert_eq!(paths[1].d, "M 1 1 Q 2 2 3 1");
     }
 
-    #[test]
-    fn paths_share_guides_and_distinct_spaces_are_capped() {
-        let path = |w: usize| format!(r#"<a:path w="{w}" h="10"><a:moveTo><a:pt x="mid" y="1"/></a:moveTo></a:path>"#);
-        let paths: String = (1..=40).map(|i| path(10 * i)).collect();
-        let x = format!(
-            r#"<a:custGeom xmlns:a="x"><a:gdLst><a:gd name="mid" fmla="*/ w 1 2"/></a:gdLst><a:pathLst>{paths}{}</a:pathLst></a:custGeom>"#,
-            path(10)
-        );
-        let d = parse(x.as_bytes()).unwrap();
-        let out = read_cust_geom(&d.root, 1.0, 1.0);
-        assert_eq!(out.len(), MAX_SPACES + 1, "16 spaces plus the repeat of the first");
-        assert!(out[0].d.starts_with("M 5 1"), "{}", out[0].d);
-        assert!(out[MAX_SPACES].d.starts_with("M 5 1"), "{}", out[MAX_SPACES].d);
-    }
-
-    #[test]
-    fn guide_free_geometries_are_not_capped_by_space() {
-        let paths: String = (1..=40).map(|i| format!(r#"<a:path w="{}" h="10"><a:moveTo><a:pt x="1" y="1"/></a:moveTo></a:path>"#, 10 * i)).collect();
-        let x = format!(r#"<a:custGeom xmlns:a="x"><a:avLst/><a:gdLst/><a:pathLst>{paths}</a:pathLst></a:custGeom>"#);
-        let d = parse(x.as_bytes()).unwrap();
-        assert_eq!(read_cust_geom(&d.root, 1.0, 1.0).len(), 40);
-    }
-
     #[test]
     fn write_then_read_path() {
         let p = CustomPath { w: 100.0, h: 50.0, d: "M 0 0 L 100 0 C 100 10 90 50 50 50 Q 0 50 0 0 Z".into(), fill: FillMode::Norm, stroke: true };

crates/pptx/src/write/chart.rs

@@ -246,6 +246,9 @@ fn point_fills(w: &mut W, s: &Series, g: Group) {
     }
 }
 
+/// Data points (categories and values counted per series) one chart group writes to its caches.
+const MAX_EXPORT_POINTS: usize = 4_000_000;
+
 fn series(w: &mut W, c: &Chart, s: &Series, i: usize, kind: ChartType, g: Group) {
     w.open0("c:ser");
     w.val("c:idx", i);
@@ -529,7 +532,16 @@ fn plot_group(w: &mut W, c: &Chart, kind: ChartType, list: &[(usize, &Series)])
     if g != Group::Stock {
         w.val("c:varyColors", vary);
     }
-    for (i, s) in list {
+    // Each series repeats the categories in its own cache, so the output grows with series times
+    // categories: stop adding series once a group's caches pass the budget (the first is always written).
+    let mut left = MAX_EXPORT_POINTS;
+    for (n, (i, s)) in list.iter().enumerate() {
+        let cost = c.categories.len().max(s.values.len()).saturating_mul(2).saturating_add(s.x_values.len() + s.sizes.len());
+        if n > 0 && cost > left {
+            log::warn!("pptx: chart export stopped after {n} series; the data caches passed {MAX_EXPORT_POINTS} points");
+            break;
+        }
+        left = left.saturating_sub(cost);
         series(w, c, s, *i, kind, g);
     }
     if c.data_labels {