fix(context): preserve language fences in formatted pre/code blocks

.changeset/preserve-html-code-languages.md

@@ -0,0 +1,5 @@
+---
+"@neuledge/context": patch
+---
+
+Preserve language fences and code indentation when HTML pre/code blocks contain surrounding formatting whitespace or comments.

packages/context/src/html.test.ts

@@ -34,4 +34,54 @@ describe("DocBook HTML examples", () => {
     );
     expect(parsed.sections[0]?.content).toContain("```sh\necho hello\n```");
   });
+
+  it.each([
+    ["spaces", "  ", "  "],
+    ["newlines", "\n  ", "\n"],
+    ["tabs", "\t", "\t\n"],
+    ["comments", "\n<!-- example -->\n  ", "\n<!-- end -->\n"],
+  ])("preserves language and indentation with %s around code", (_, before, after) => {
+    const parsed = parseHtml(
+      `<h1>Guide</h1><h2>Example</h2><pre>${before}<code class="highlight language-sh">if true; then\n  echo hello\nfi\n</code>${after}</pre>`,
+      "guide.html",
+    );
+    expect(parsed.sections).toHaveLength(1);
+    expect(parsed.sections[0]?.hasCode).toBe(true);
+    expect(parsed.sections[0]?.content).toBe(
+      "```sh\nif true; then\n  echo hello\nfi\n```",
+    );
+  });
+
+  it("keeps embedded backticks inside a language fence with wrapper whitespace", () => {
+    const parsed = parseHtml(
+      '<h1>Guide</h1><h2>Example</h2><pre>\n  <code class="language-markdown">first\n```\nlast</code>\n</pre>',
+      "guide.html",
+    );
+    expect(parsed.sections).toHaveLength(1);
+    expect(parsed.sections[0]?.hasCode).toBe(true);
+    expect(parsed.sections[0]?.content).toBe(
+      "````markdown\nfirst\n```\nlast\n````",
+    );
+  });
+
+  it.each([
+    [
+      'prefix <code class="language-sh">echo hello</code> suffix',
+      "prefix echo hello suffix",
+    ],
+    [
+      '<span>prefix </span><code class="language-sh">echo hello</code>',
+      "prefix echo hello",
+    ],
+    [
+      '  <code class="language-sh">echo hello</code><code> suffix</code>',
+      "  echo hello suffix",
+    ],
+  ])("preserves meaningful siblings in mixed preformatted content: %s", (content, expected) => {
+    const parsed = parseHtml(
+      `<h1>Guide</h1><h2>Example</h2><pre>${content}</pre>`,
+      "guide.html",
+    );
+    expect(parsed.sections[0]?.content).toBe(`\`\`\`\n${expected}\n\`\`\``);
+  });
 });

packages/context/src/html.ts

@@ -40,17 +40,31 @@ for (const tag of REMOVED_TAGS) {
 
 // DocBook emits bare <pre> elements; Turndown's code rule requires <pre><code>.
 // Preserve their whitespace and prevent Markdown escaping of unit-file examples.
+// Also recognize a sole <code> child preceded by whitespace or comments.
 turndown.addRule("barePre", {
   filter: (node) =>
     node.nodeName === "PRE" && node.firstChild?.nodeName !== "CODE",
   replacement: (_content, node) => {
-    const code: string = node.textContent ?? "";
+    const codeElement = node.firstElementChild;
+    const isWrappedCode =
+      codeElement?.nodeName === "CODE" &&
+      [...node.childNodes].every(
+        (child) =>
+          child === codeElement ||
+          (child.nodeType === 3 && !/\S/.test(child.textContent ?? "")) ||
+          child.nodeType === 8,
+      );
+    const className = isWrappedCode
+      ? (codeElement.getAttribute("class") ?? "")
+      : "";
+    const language = className.match(/\blanguage-(\S+)/)?.[1] ?? "";
+    const code: string = (isWrappedCode ? codeElement : node).textContent ?? "";
     const longestRun = (code.match(/`+/g) ?? []).reduce(
       (longest, run) => Math.max(longest, run.length),
       0,
     );
     const fence = "`".repeat(Math.max(3, longestRun + 1));
-    return `\n\n${fence}\n${code.replace(/\n$/, "")}\n${fence}\n\n`;
+    return `\n\n${fence}${language}\n${code.replace(/\n$/, "")}\n${fence}\n\n`;
   },
 });