fix: derive a title from the URL when an external post has none

[alshedivat]

The defect

A degraded fetch left external posts with an empty title, which was then published.

fetch_content_from_url returns '' when the request fails or the page has no <title>, and an RSS <item> can carry a blank <title> of its own. create_document copied that straight into doc.data['title'].

Reproduced in a real build of the al-folio starter, where the Google Blog entry in external_sources produced:

<a class="post-title" href="https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/" target="_blank" rel="external nofollow noopener"></a>

a blank but clickable row in the blog index and on the home page, an untitled entry in the search index (id: "post-"), and 103 Warning: Empty `slug` generated for ''. lines from Jekyll (jekyll/utils.rb:222) slugifying that empty title downstream. With external_sources: [] the same build emits 0 such warnings.

How we got here

• fix: guard against nil post titles crashing external-post generation #10 made build_slug nil-safe, fixing a NoMethodError that previously aborted the whole build on a nil title.
• perf: optimize external-post slug generation (port of al-folio#3450) #11 optimized the slug pass.

Neither is wrong, but neither filled in the title. The failure mode went from "build crashes" to "build silently publishes a blank post". This PR fixes the remaining half. (al_ext_posts has its own private slugify and never calls Jekyll::Utils.slugify; the Jekyll warnings come from Jekyll slugifying the synthetic posts' empty titles downstream, so fixing the title fixes the warnings too.)

The fix

Derive rather than skip. These URLs are explicitly listed by the user in _config.yml, so dropping them would silently lose content they asked for. The entry is kept, made readable, and the degradation is reported.

resolve_title (lib/al_ext_posts.rb:113) returns the title when it carries any word character, and otherwise derives one from the URL's last meaningful path segment and logs a warning naming the URL:

ExternalPosts: No title found for https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/ - using "Google Gemini Update Flash Ai Assistant Io 2024" derived from the URL.

It is called from create_document, the single funnel both the RSS path (fetch_from_rss → process_entries) and the explicit-URL path (fetch_from_urls) pass through, so an RSS item with a missing or blank <title> gets exactly the same treatment.

Post slugs, and therefore post URLs, are unchanged: build_slug still receives the raw title and still uses its source-name fallback. The built sitemap is byte-identical across the before/after builds.

The usability check for a display title is /[[:word:]]/ (Unicode-aware), not the ASCII \w that build_slug uses for a filesystem slug — so a CJK title like 你好世界 is preserved as the title even though the slug still falls back.

Edge cases

URL	Title
.../google-gemini-update-flash-ai-assistant-io-2024/	Google Gemini Update Flash Ai Assistant Io 2024
https://example.com/posts/my-post/ (trailing slash)	My Post
.../my-post.html?utm_source=x#intro (extension, query, fragment)	My Post
.../hello%20world%21 (percent-encoding)	Hello World
.../my_post_title	My Post Title
https://example.com/archive/2024/05/deep-dive/	Deep Dive (numeric segments skipped)
https://example.com/posts/12345 (numeric last segment)	Posts
https://example.com/posts/---/ (no words in last segment)	Posts
https://example.com/post.v2 (unknown extension kept)	Post V2
https://blog.google (domain only)	blog.google
https://www.example.com/	example.com
https://example.com///// (empty segments only)	example.com
https://example.com/2024/05/12/ (no segment carries words)	example.com
nil, ''	External post

Malformed input never raises: URI::Error falls back to trimming the query/fragment by hand, bad percent-escapes keep the raw segment, and invalid UTF-8 bytes are scrubbed before any regexp match. The function cannot return an empty string.

Verification

Rebuilt the al-folio starter against this branch, in a sandbox with no general outbound network — a faithful simulation of the degraded-fetch case.

	before	after
Empty `slug` generated warnings	103	0
empty post-title anchors in blog/index.html	1	0

The blog index now renders:

<a class="post-title" href="https://blog.google/..." target="_blank" rel="external nofollow noopener">Google Gemini Update Flash Ai Assistant Io 2024</a>

The home page news-title anchor and the search-index entry (id: "post-" → id: "post-google-gemini-update-flash-ai-assistant-io-2024", with a real title) are fixed by the same change. The sitemap is unchanged, confirming no URL churn.

Tests

Added coverage for a missing title, a blank/whitespace title, a title of only non-word characters, a preserved non-Latin title, the warning naming the URL, every URL edge case above, a never-empty/never-raises sweep over malformed URLs, an end-to-end create_document assertion (title filled, slug unchanged), and the RSS path via process_entries.

CaptureGenerator#initialize now accepts and ignores an argument, because Jekyll::Site#setup instantiates every Generator subclass with the site config once a test builds a real site.

23 runs, 77 assertions, 0 failures, 0 errors, 0 skips

npm run lint:prettier passes.

Release note

Version bumped 1.0.2 → 1.0.3. 1.0.2 was never published to RubyGems — main is ahead of the published 1.0.1 — so the eventual release should cover both the 1.0.2 slug-generation optimization (#11) and this 1.0.3 fix.

---

Generated by Claude Code