From 8732c5649903fba8935a2ef909e1ac1c691071f2 Mon Sep 17 00:00:00 2001
From: jpirnay <jens@pirnay.com>
Date: Mon, 18 May 2026 08:18:13 +0200
Subject: [PATCH] Properly deal with svg elements

---
 .../Epub/parsers/ChapterHtmlSlimParser.cpp    | 26 +++++++++++++++++++
 lib/Epub/Epub/parsers/ChapterHtmlSlimParser.h |  1 +
 2 files changed, 27 insertions(+)

diff --git a/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.cpp b/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.cpp
index 3dc0c968..fefde980 100644
--- a/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.cpp
+++ b/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.cpp
@@ -408,6 +408,22 @@ void XMLCALL ChapterHtmlSlimParser::startElement(void* userData, const XML_Char*
     return;
   }
 
+  // Track SVG nesting depth. Must be checked before the svgDepth>0 guard below so that
+  // nested <svg> elements increment the counter rather than being swallowed as unknowns.
+  if (strcmp(name, "svg") == 0) {
+    self->svgDepth += 1;
+    self->depth += 1;
+    return;
+  }
+
+  // Inside SVG: only process <image> elements (raster images); skip everything else.
+  // SVG child elements like <path>, <rect>, <circle>, <text> must not reach the layout
+  // engine — they would accumulate path data and exhaust heap on large inline SVG.
+  if (self->svgDepth > 0 && !matches(name, IMAGE_TAGS, NUM_IMAGE_TAGS)) {
+    self->depth += 1;
+    return;
+  }
+
   // Extract class, style, and id attributes
   std::string classAttr;
   std::string styleAttr;
@@ -1319,6 +1335,12 @@ void XMLCALL ChapterHtmlSlimParser::characterData(void* userData, const XML_Char
     return;
   }
 
+  // Skip SVG text content (path data, coordinates, etc.) — it would be treated as words
+  // and exhaust heap on EPUBs with large inline SVG elements.
+  if (self->svgDepth > 0) {
+    return;
+  }
+
   // Collect footnote link display text (for the number label)
   // Remove leading/trailing whitespace and square brackets from the
   // footnote link text to normalize noterefs like "[1]" → "1"
@@ -1559,6 +1581,10 @@ void XMLCALL ChapterHtmlSlimParser::endElement(void* userData, const XML_Char* n
 
   self->depth -= 1;
 
+  if (strcmp(name, "svg") == 0 && self->svgDepth > 0) {
+    self->svgDepth -= 1;
+  }
+
   // Pop list entries whose ul/ol is now out of scope
   while (!self->listStack.empty() && self->listStack.back().depth >= self->depth) {
     self->listStack.pop_back();
diff --git a/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.h b/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.h
index 6c1f9eb9..c632b254 100644
--- a/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.h
+++ b/lib/Epub/Epub/parsers/ChapterHtmlSlimParser.h
@@ -36,6 +36,7 @@ class ChapterHtmlSlimParser final : public Print {
   int underlineUntilDepth = INT_MAX;
   int strikethroughUntilDepth = INT_MAX;
   int preUntilDepth = INT_MAX;  // set when inside a <pre> element; enables \n → line-break handling
+  int svgDepth = 0;             // nesting counter for <svg> elements; text inside SVG is skipped (path data etc.)
   // buffer for building up words from characters, will auto break if longer than this
   // leave one char at end for null pointer
   char partWordBuffer[MAX_WORD_SIZE + 1] = {};