Fall back to English when translated frontmatter parses to a non-object (#62093)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: heiskr

src/article-api/tests/pageinfo.ts

@@ -58,6 +58,19 @@ describe('pageinfo api', () => {
     expect(res.headers['surrogate-key']).toBe(makeLanguageSurrogateKey('en'))
   })
 
+  test('corrupted translation frontmatter falls back to English title', async () => {
+    // The Japanese counterpart of this page has frontmatter whose ASCII `:`
+    // key/value separators were replaced with fullwidth colons (`：`), so the
+    // YAML parses to a scalar string rather than an object. When that happens
+    // none of the frontmatter keys exist. Without the non-object fallback in
+    // page-data, `meta.title` comes back empty and the search scraper rejects
+    // the record with "Record has empty title". It must fall back to English.
+    const res = await get(makeURL('/ja/get-started/foo/broken-frontmatter-translation'))
+    expect(res.statusCode).toBe(200)
+    const meta = JSON.parse(res.body) as PageMetadata
+    expect(meta.title).toBe('Broken frontmatter translation fallback')
+  })
+
   test('a pathname that does not exist', async () => {
     const res = await get(makeURL('/en/never/heard/of'))
     expect(res.statusCode).toBe(404)

src/fixtures/fixtures/content/get-started/foo/broken-frontmatter-translation.md

@@ -0,0 +1,14 @@
+---
+title: Broken frontmatter translation fallback
+intro: Demonstrates falling back to English when a translation's frontmatter is corrupted.
+versions:
+  fpt: "*"
+  ghes: "*"
+  ghec: "*"
+---
+
+## Intro
+
+This page exists to test that a translated file whose frontmatter is corrupted
+into a non-object (for example, fullwidth colons replacing the ASCII `:`
+separators) falls back to the English title instead of rendering an empty title.

src/fixtures/fixtures/content/get-started/foo/index.md

@@ -1,11 +1,11 @@
 ---
 title: Fooing Around
 shortTitle: Foo
-intro: 'The most basic of fixture data for {% data variables.product.product_name %}'
+intro: "The most basic of fixture data for {% data variables.product.product_name %}"
 versions:
-  fpt: '*'
-  ghes: '*'
-  ghec: '*'
+  fpt: "*"
+  ghes: "*"
+  ghec: "*"
 children:
   - /bar
   - /autotitling
@@ -19,4 +19,5 @@ children:
   - /table-with-ifversions
   - /code-snippet-with-hashbang
   - /journey-test-article
+  - /broken-frontmatter-translation
 ---

src/fixtures/fixtures/translations/ja-jp/content/get-started/foo/broken-frontmatter-translation.md

@@ -0,0 +1,9 @@
+---
+title： 壊れたフロントマターのフォールバック shortTitle： フォールバック intro： 翻訳のフロントマターが壊れているときに英語にフォールバックすることを示します。
+versions： fpt： '*' ghes： '*' ghec： '*'
+---
+
+## Intro
+
+このページは、翻訳ファイルのフロントマターが非オブジェクトに壊れている場合に、
+空のタイトルではなく英語のタイトルにフォールバックすることをテストするために存在します。

src/frame/lib/page-data.ts

@@ -164,7 +164,7 @@ async function translateTree(
     content = read.content
     data = read.data as Record<string, unknown>
 
-    if (!data) {
+    if (!data || typeof data !== 'object' || Array.isArray(data)) {
       // If the file's frontmatter Yaml is entirely broken,
       // the result of `readFileContents()` is that you just
       // get a `errors` key. E.g.
@@ -177,6 +177,15 @@ async function translateTree(
       //     }
       //   ]
       //
+      // A translated file can also be corrupted so that the frontmatter
+      // parses to a non-object. For example, if machine translation replaces
+      // the ASCII `:` key/value separators with fullwidth colons (`：`), YAML
+      // parses the whole block as a single scalar string. `data` is then a
+      // string rather than an object, so none of the frontmatter keys (like
+      // `title`) exist and the per-key English fallback below never fires,
+      // leaving the page with an empty title. Treat that the same as entirely
+      // broken frontmatter and fall back to English.
+      //
       // If this the case throw error so we can lump this error with
       // how we deal with the file not even being present on disk.
       throw new FrontmatterParsingError(JSON.stringify(read.errors), true)