fix: markdown parser not detecting sections with horizontal rules

The markdownParser was incorrectly interpreting horizontal rules (---) as setext headers when they appeared after non-header text. This caused some sections to be missed in the output.

This fix:
- Makes setext header detection more strict by requiring at least 3 = or - characters
- Adds validation for the text line before a potential setext header
- Ensures horizontal rules are not confused with setext headers

Added a test case to verify the fix works correctly with horizontal rules.

Signed-off-by: Eric Wheeler <[email protected]>

Author: Eric Wheeler

src/services/tree-sitter/__tests__/markdownParser.test.ts

@@ -462,6 +462,42 @@ Line 5`
 			}
 		}
 	})
+
+	it("should correctly handle horizontal rules and not confuse them with setext headers", () => {
+		const content = `## Section Header
+
+Some content here.
+
+## License
+
+[Apache 2.0 © 2025 Roo Veterinary, Inc.](./LICENSE)
+
+---
+
+**Enjoy Roo Code!** Whether you keep it on a short leash or let it roam autonomously, we can't wait to see what you build.`
+
+		const captures = parseMarkdown(content)
+		expect(captures).toBeDefined()
+
+		// Format with default minSectionLines = 4
+		const formatted = formatMarkdownCaptures(captures)
+		expect(formatted).toBeDefined()
+		expect(formatted).toContain("## Section Header")
+		expect(formatted).toContain("## License")
+
+		// Verify that the horizontal rule is not treated as a setext header
+		const licenseCapture = captures.find((c) => c.node.text === "License")
+		expect(licenseCapture).toBeDefined()
+
+		// Check that the License section extends past the horizontal rule
+		const licenseCaptureIndex = captures.findIndex((c) => c.node.text === "License")
+		if (licenseCaptureIndex !== -1 && licenseCaptureIndex + 1 < captures.length) {
+			const licenseDefinitionCapture = captures[licenseCaptureIndex + 1]
+			expect(licenseDefinitionCapture.node.endPosition.row).toBeGreaterThan(
+				content.split("\n").findIndex((line) => line === "---"),
+			)
+		}
+	})
 })
 
 // Helper function to mimic the processCaptures function from index.ts

src/services/tree-sitter/markdownParser.ts

@@ -42,8 +42,11 @@ export function parseMarkdown(content: string): MockCapture[] {
 
 	// Regular expressions for different header types
 	const atxHeaderRegex = /^(#{1,6})\s+(.+)$/
-	const setextH1Regex = /^=+\s*$/
-	const setextH2Regex = /^-+\s*$/
+	// Setext headers must have at least 3 = or - characters
+	const setextH1Regex = /^={3,}\s*$/
+	const setextH2Regex = /^-{3,}\s*$/
+	// Valid setext header text line should be plain text (not empty, not indented, not a special element)
+	const validSetextTextRegex = /^\s*[^#<>!\[\]`\t]+[^\n]$/
 
 	// Find all headers in the document
 	for (let i = 0; i < lines.length; i++) {
@@ -80,7 +83,7 @@ export function parseMarkdown(content: string): MockCapture[] {
 		// Check for setext headers (underlined headers)
 		if (i > 0) {
 			// Check for H1 (======)
-			if (setextH1Regex.test(line)) {
+			if (setextH1Regex.test(line) && validSetextTextRegex.test(lines[i - 1])) {
 				const text = lines[i - 1].trim()
 
 				// Create a mock node for this header
@@ -106,7 +109,7 @@ export function parseMarkdown(content: string): MockCapture[] {
 			}
 
 			// Check for H2 (------)
-			if (setextH2Regex.test(line)) {
+			if (setextH2Regex.test(line) && validSetextTextRegex.test(lines[i - 1])) {
 				const text = lines[i - 1].trim()
 
 				// Create a mock node for this header