fix(pptx): warn when custom geometry paths are left out for the guide-cache limits

- crates/pptx/src/custgeom.rs

crates/pptx/src/custgeom.rs

@@ -15,6 +15,14 @@ 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-name bytes all cached spaces may hold between them; the first space counts towards it but is always kept.
+/// Every space copies the names, which the XML limits bound only once.
+const MAX_SPACE_NAME_BYTES: usize = 4 << 20;
+
 /// Guide variables for one shape.
 pub struct Guides {
     vars: HashMap<String, f64>,
@@ -66,6 +74,10 @@ impl Guides {
         set("7cd8", 18_900_000.0);
         Guides { vars: v }
     }
+    /// Bytes of the guide names held.
+    fn name_bytes(&self) -> usize {
+        self.vars.keys().map(String::len).sum()
+    }
     pub fn get(&self, name: &str) -> f64 {
         let name = name.trim();
         if let Some(v) = self.vars.get(name) {
@@ -149,16 +161,51 @@ fn pt(g: &Guides, e: &El) -> (f64, f64) {
 /// Convert `a:custGeom` to custom paths. `w`/`h` is the shape size in EMU (used for guides and for
 /// paths that don't give their own coordinate space).
 pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
+    read_cust_geom_counting(cg, w_emu, h_emu).0
+}
+
+/// [`read_cust_geom`], and how many paths were left out for lack of room in the guide caches
+/// (`(too many coordinate spaces, too many guide-name bytes)`); each cause is also logged once.
+fn read_cust_geom_counting(cg: &El, w_emu: f64, h_emu: f64) -> (Vec<CustomPath>, (usize, usize)) {
     let mut out = vec![];
-    let Some(pl) = cg.child("pathLst") else { return out };
+    let mut dropped = (0usize, 0usize);
+    let Some(pl) = cg.child("pathLst") else { return (out, dropped) };
+    let mut spaces: HashMap<(u64, u64), Guides> = HashMap::new();
+    let mut name_bytes = 0usize;
+    let mut budget_full = false;
+    // 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) };
-        let mut g = Guides::new(gw, gh);
-        g.eval_list(cg.child("avLst"));
-        g.eval_list(cg.child("gdLst"));
+        // 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 && (budget_full || spaces.len() >= MAX_SPACES) {
+                if budget_full {
+                    dropped.1 += 1;
+                } else {
+                    dropped.0 += 1;
+                }
+                continue;
+            }
+            let mut g = Guides::new(gw, gh);
+            g.eval_list(cg.child("avLst"));
+            g.eval_list(cg.child("gdLst"));
+            // The first space is always kept; later ones only while the cache stays within budget.
+            let bytes = g.name_bytes();
+            if guided && !spaces.is_empty() && name_bytes.saturating_add(bytes) > MAX_SPACE_NAME_BYTES {
+                // Every space holds the same names, so no later space fits either.
+                budget_full = true;
+                dropped.1 += 1;
+                continue;
+            }
+            name_bytes = name_bytes.saturating_add(bytes);
+            spaces.insert(key, g);
+        }
+        let Some(g) = spaces.get(&key) else { continue };
         let mut d = String::new();
         let mut cur = (0.0, 0.0);
         let mut start = (0.0, 0.0);
@@ -166,7 +213,7 @@ 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);
@@ -177,7 +224,7 @@ 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(' ');
@@ -186,7 +233,7 @@ 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] {
@@ -199,7 +246,7 @@ 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] {
@@ -236,7 +283,16 @@ pub fn read_cust_geom(cg: &El, w_emu: f64, h_emu: f64) -> Vec<CustomPath> {
         let (ow, oh) = if pw > 0.0 || ph > 0.0 { (pw, ph) } else { (w_emu.max(1.0), h_emu.max(1.0)) };
         out.push(CustomPath { w: ow, h: oh, d: d.trim_end().to_string(), fill, stroke: p.bool("stroke").unwrap_or(true) });
     }
-    out
+    if dropped.0 > 0 {
+        log::warn!("pptx: a custom geometry uses more than {MAX_SPACES} distinct path coordinate spaces; {} later path(s) are not kept", dropped.0);
+    }
+    if dropped.1 > 0 {
+        log::warn!(
+            "pptx: a custom geometry's guide names pass {MAX_SPACE_NAME_BYTES} bytes across its coordinate spaces; {} later path(s) are not kept",
+            dropped.1
+        );
+    }
+    (out, dropped)
 }
 
 /// Ellipse parametric angle for a visual angle `a` (radians).
@@ -470,6 +526,46 @@ 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, dropped) = read_cust_geom_counting(&d.root, 1.0, 1.0);
+        assert_eq!(out.len(), MAX_SPACES + 1, "16 spaces plus the repeat of the first");
+        assert_eq!(dropped, (40 - MAX_SPACES, 0), "the paths of later spaces are counted, to be warned about");
+        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_name_bytes_across_spaces_are_budgeted() {
+        let name = "n".repeat(3 << 20);
+        let path = |w: usize| format!(r#"<a:path w="{w}" h="10"><a:moveTo><a:pt x="{name}" y="1"/></a:moveTo></a:path>"#);
+        let x = format!(
+            r#"<a:custGeom xmlns:a="x"><a:gdLst><a:gd name="{name}" fmla="*/ w 1 2"/></a:gdLst><a:pathLst>{}{}{}</a:pathLst></a:custGeom>"#,
+            path(10),
+            path(20),
+            path(10)
+        );
+        let d = parse(x.as_bytes()).unwrap();
+        let (out, dropped) = read_cust_geom_counting(&d.root, 1.0, 1.0);
+        assert_eq!(out.len(), 2, "the second 3 MiB space is over budget and left out; the first space's repeat stays");
+        assert_eq!(dropped, (0, 1), "the path left out is counted, to be warned about");
+    }
+
+    #[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 };