core(link-elements): ignore non-HTMLLinkElements

[patrickhulce]

Summary
updates link elements gatherer to only consider real link elements and adds a smoketest for it

Related Issues/PRs
fixes #9764

[brendankenny]

I was only able to test it briefly, so I could have this wrong, but from the test site https://sundaysuppermovement.com/ in #9764, it appears that the only link rel="preload" elements are SVG elements, but they are triggering preloads, so dropping these might lead to unexpected results in things like uses-rel-preload and etc

[patrickhulce]

it appears that the only link rel="preload" elements are SVG elements, but they are triggering preloads, so dropping these might lead to unexpected results in things like uses-rel-preload and etc

No they shouldn't be triggering anything, they're just invalid elements. AFAICT, the preload requests come from the rel="preload" onload="this.rel = 'stylesheet'" pattern the page uses.

[brendankenny]

AFAICT, the preload requests come from the rel="preload" onload="this.rel = 'stylesheet'" pattern the page uses.

Nice catch!

Ah, sorry, I'm finally getting what you're saying. These end up as SVGUnknownElement (though that's not a public interface) since they're from a foreign namespace, so while they get parsed (and are apparently containers in svg, interesting) and have a link tagName, they are otherwise inert.

In general, the SVG user agent must include the unknown foreign-namespaced elements in the DOM but will ignore and exclude them for rendering purposes.

This sounds good then.

[brendankenny]

any reason to do this in dbw and not in perf or seo where LinkElements are already being used?

[patrickhulce]

just that dbw felt like the dumping ground of error catching appropriate for this sort of thing 😆

the others are so focused it felt mean to pollute them with SVG nonsense

[brendankenny]

LGTM!

[brendankenny]

this is a loss for documentation but I suppose there's always mdn :)