Give each engine archive its own feed, and make the feeds validate

[yusuf-gundogdu]

The blog groups its posts by engine, but it publishes one feed for all of them, so anyone who wants a single engine has to filter us from the outside. Planet for the MySQL Community does exactly that today, and it asked us for a better input.

Two commits, in this order, so the change to the feed people already subscribe to can be read and reverted on its own.

What was wrong

The only thing a third party can filter on is the title. We are aggregated at planet.oursqlcommunity.org through siftrss with an include/title//mysql/i rule, because that is all an outside service can see. Eight published posts are about MySQL and the rule reaches seven. The one it drops is "Nine servers, one connection type, different answers", whose title names the behaviour rather than the engine, which is how we title most posts. The maintainer predicted this when we submitted and asked whether we could expose a category feed instead.

We are paying for that workaround in public. Our row in their blogroll points its RSS icon at siftrss.com rather than at libredb.org, and carries their filtering icon beside our name. The maintainer also had to create the siftrss link himself and now maintains it.

Our feed does not validate. It writes the author's name into <author>, which RSS 2.0 defines as an email address. The W3C validator answers Invalid email address once per item and refuses the feed, 104 times on /rss.xml as it stands.

One deleted line could have broken every feed silently. The namespaces are declared on <rss>; without them the documents are not well-formed XML and every reader rejects them. I checked: deleting that line left the whole suite green.

What changed

Each engine archive serves the feed for its own posts at /blog/engine/<engine>/rss.xml, built from the post set the archive page lists, with the engine as the first category on every item so a category filter works without touching the title. The archive page links it in the body and declares it in the head ahead of the site feed, which stays advertised so the whole blog is still reachable from there.

The author's name moved to <dc:creator> on every feed and each channel now carries atom:link rel="self".

The grouping and its threshold already existed three times: the archive route, the sibling links on an archive, and the engine index on /blog. The feed would have been a fourth, so all four now read engineArchives() in lib/posts and the threshold is one constant.

	before	after
MySQL posts an aggregator can reach	7	8
feeds published	1	18
feeds the W3C validator accepts	0	18
copies of the engine grouping and its threshold	3	1

Tests

Six new assertions, each checked to fail without the thing it guards rather than assumed to:

removed from the source	what went red
the xmlns declaration	feeds parse as XML
dc:creator, back to <author>	that, plus the escaping test
atom:link rel="self"	feeds parse as XML
pairing between page route and feed route	pairs one feed with one archive
dist/blog/engine/mysql/rss.xml	keeps serving the feed an aggregator subscribed to

The fixture behind the archive tests counted every post file on disk while the routes count published ones, so saving a draft turned the suite red with nothing wrong in the site. I reproduced that, then filtered the fixture on status, which also settles nine older tests that shared it.

Follow-up outside this repo

Once this is live their planet.ini entry can point at https://libredb.org/blog/engine/mysql/rss.xml and drop the siftrss link, the filtering comments and the _FI_ title prefix that draws the icon. The guids are identical between the two feeds, so switching will not re-flood their front page with posts they have already carried. I am answering the submission thread with the address; their own contributing history says the maintainer makes that edit himself, so I am not sending a pull request there unless he asks.