fix(parser): collapse a leading // in sanitizeUrl

packages/parser/src/utils/normalize-href.ts

@@ -232,6 +232,14 @@ export function normalizeMenubarPaths(items: any[]): void {
  */
 export function sanitizeUrl(url: string): string {
   if (!url) return url;
-  // Collapse double+ slashes, but preserve protocol://
-  return url.replace(/([^:])\/\/+/g, '$1/');
+  // Preserve a leading `scheme://` (e.g. `https://`) but collapse every other
+  // run of consecutive slashes. A leading `//` must collapse too: browsers
+  // treat `//host/path` as a protocol-relative URL pointing at a *different
+  // host*, so an accidental `//search/` (from base + `/search`) would resolve
+  // to `https://search/` instead of the intended same-site `/search/`.
+  const scheme = url.match(/^[a-z][a-z0-9+.-]*:\/\//i);
+  if (scheme) {
+    return scheme[0] + url.slice(scheme[0].length).replace(/\/{2,}/g, '/');
+  }
+  return url.replace(/\/{2,}/g, '/');
 }

packages/parser/test/sanitize-url.test.js

@@ -0,0 +1,48 @@
+/**
+ * --------------------------------------------------------------------
+ * docmd : the zero-config documentation engine.
+ *
+ * sanitizeUrl — slash-collapsing safety net.
+ *
+ * Regression coverage for a leading `//` collapsing to a single `/`.
+ * A leading `//host/path` is a protocol-relative URL: a browser resolves
+ * it against the current *scheme* but a *different host*, so an accidental
+ * `//search/` (produced by `base` + `/search`, base normalised to `/`)
+ * would navigate to `https://search/` instead of the same-site `/search/`.
+ * The scheme separator in a real absolute URL (`https://`) must survive.
+ *
+ * Run: `pnpm --filter @docmd/parser test`
+ * --------------------------------------------------------------------
+ */
+
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import { sanitizeUrl } from '../dist/index.js';
+
+test('sanitizeUrl collapses a leading // (protocol-relative) to /', () => {
+  assert.equal(sanitizeUrl('//search/'), '/search/');
+  assert.equal(sanitizeUrl('//docs//guide/'), '/docs/guide/');
+});
+
+test('sanitizeUrl collapses interior and trailing runs of slashes', () => {
+  assert.equal(sanitizeUrl('/a///b////c/'), '/a/b/c/');
+  assert.equal(sanitizeUrl('/guide/'), '/guide/');
+});
+
+test('sanitizeUrl preserves the scheme:// separator', () => {
+  assert.equal(sanitizeUrl('https://a.com//b'), 'https://a.com/b');
+  assert.equal(sanitizeUrl('https://x.com/a/b/'), 'https://x.com/a/b/');
+  assert.equal(sanitizeUrl('http://h//a//b'), 'http://h/a/b');
+  assert.equal(sanitizeUrl('ws://h//x'), 'ws://h/x');
+});
+
+test('sanitizeUrl leaves clean relative paths untouched', () => {
+  assert.equal(sanitizeUrl('../de/guide/'), '../de/guide/');
+  assert.equal(sanitizeUrl(''), '');
+});
+
+test('a browser resolves the sanitised path against the current host', () => {
+  // The whole point of the fix: `//search/` would resolve cross-host.
+  const href = sanitizeUrl('//search/');
+  assert.equal(new URL(href, 'https://example.com/docs/').href, 'https://example.com/search/');
+});

tests/cli-contracts/asset-base-url.test.js

@@ -297,19 +297,20 @@ export const test = runTestFile({
       assert(result.ok, 'URL-3b: workspace build for switcher test succeeds');
       const mainHtml = fs.readFileSync(path.join(proj, 'site/index.html'), 'utf8');
       // Find every project-switcher-item link and capture { href, title }.
-      // The href is protocol-relative (//search/) because buildAbsoluteUrl
-      // normalises the empty base to '/', which combines with /search to
-      // //search. Browsers treat // as the same-scheme prefix, so this is
-      // equivalent to /search/ in absolute terms.
+      // The href must be the same-site absolute path `/search/`. It must NOT
+      // be protocol-relative (`//search/`): a browser resolves `//search/`
+      // against a *different host* (`https://search/`), not the current site,
+      // so the switcher link would leave the docs entirely.
       const switcherHrefs = Array.from(mainHtml.matchAll(/<a\s+href="([^"]+)"\s+class="project-switcher-item[^"]*"[^>]*>([\s\S]*?)<\/a>/g))
         .map(m => ({ href: m[1], title: (m[2].match(/<span class="project-title">([^<]+)<\/span>/) || [])[1] }));
       const searchHref = switcherHrefs.find(h => h.title === 'search');
       assert(searchHref, 'URL-3b: project switcher has a link to "search" sub-site');
       // The previous bug emitted /search (no slash) which made the
       // browser treat the URL as a file when npx serve served the
-      // directory index. The fix keeps the trailing slash.
-      assert(searchHref && /\/search\/$/.test(searchHref.href),
-        `URL-3b: project switcher link to /search sub-site ends with /search/ (got: ${searchHref?.href})`);
+      // directory index. The fix keeps the trailing slash, and the link
+      // must stay same-site (exactly `/search/`, never `//search/`).
+      assert(searchHref && /^\/search\/$/.test(searchHref.href),
+        `URL-3b: project switcher link to /search sub-site is exactly "/search/" (got: ${searchHref?.href})`);
       // The root project link should be relative "./" or absolute "/" (no extra trailing slash
       // for the root).
       const mainHref = switcherHrefs.find(h => h.title === 'main');