fix(pandoc): recognize fence markers indented up to 3 spaces

markdownWithExtractedHeading scanned every notebook cell, not just
code cells. Code cells always emit fences at column 0, but markdown
cells carry arbitrary user-authored content, where a fenced block
nested in a list item can have its fence markers indented while a
line inside stays at column 0. That line was previously invisible to
fence tracking and could still be mistaken for a heading.

Author: cderv

src/core/pandoc/pandoc-partition.ts

@@ -72,10 +72,10 @@ export function partitionMarkdown(markdown: string): PartitionedMarkdown {
   };
 }
 
-// CommonMark also allows fences indented up to 3 spaces; intentionally
-// unsupported here since the fences this scanner sees (code cell source)
-// always start at column 0.
-const kFenceOpenRegex = /^(`{3,}|~{3,})/;
+// CommonMark allows fence markers indented up to 3 spaces (content lines
+// inside the fence can have any indentation, including none).
+const kFenceOpenRegex = /^ {0,3}(`{3,}|~{3,})/;
+const kFenceCloseRegex = /^ {0,3}(`{3,}|~{3,})\s*$/;
 
 export function markdownWithExtractedHeading(markdown: string) {
   const mdLines: string[] = [];
@@ -90,7 +90,7 @@ export function markdownWithExtractedHeading(markdown: string) {
     // never mistaken for the document heading.
     if (fence) {
       mdLines.push(line);
-      const closeMatch = line.match(/^(`{3,}|~{3,})\s*$/);
+      const closeMatch = line.match(kFenceCloseRegex);
       if (
         closeMatch && closeMatch[1][0] === fence.char &&
         closeMatch[1].length >= fence.length

tests/unit/pandoc-partition.test.ts

@@ -145,6 +145,22 @@ unitTest(
   },
 );
 
+// deno-lint-ignore require-await
+unitTest(
+  "markdownWithExtractedHeading - ignores an ATX-heading-like line inside a fenced code block whose fence markers are indented up to 3 spaces",
+  async () => {
+    const markdown = [
+      "- a list item with a nested code block",
+      "",
+      "  ```python",
+      "# not a heading",
+      "  ```",
+    ].join("\n");
+    const result = markdownWithExtractedHeading(markdown);
+    assertEquals(result.headingText, undefined);
+  },
+);
+
 // deno-lint-ignore require-await
 unitTest(
   "markdownWithExtractedHeading - extracts a setext-style heading",