Multiple <link rel=preload>s with the same preload key appear to cause multiple requests

[domenic]

What is the issue with the HTML Standard?

Reading the spec in https://html.spec.whatwg.org/#link-type-preload , I am pretty sure that

<link rel=preload as=image href=foo.png>
<link rel=preload as=image href=foo.png>

requires doing two fetches. If I am reading the Fetch spec right, the second one might even consume the preload response from the first one?

This seems like a bug. @noamr, do you agree?

I know there are more general issues around to what extent the "memory cache" is specified. But my understanding was that @noamr's work on <link rel=preload> was intended to start putting us on solid footing there. So maybe tackling just this particular case first is worthwhile.

/cc @nidhijaju

[noamr]

What is the issue with the HTML Standard?

Reading the spec in https://html.spec.whatwg.org/#link-type-preload , I am pretty sure that

<link rel=preload as=image href=foo.png>

<link rel=preload as=image href=foo.png>

requires doing two fetches. If I am reading the Fetch spec right, the second one might even consume the preload response from the first one?

This seems like a bug. @noamr, do you agree?

I know there are more general issues around to what extent the "memory cache" is specified. But my understanding was that @noamr's work on <link rel=preload> was intended to start putting us on solid footing there. So maybe tackling just this particular case first is worthwhile.

/cc @nidhijaju

The second preload consumes the first preload, and the following fetch will consume the second preload.
So this works as intended I think?

[domenic]

Yeah, I think you're right. And that's not really observable, so an implementation could optimize it by just not doing the second preload, I think?

[noamr]

Yeah, I think you're right. And that's not really observable, so an implementation could optimize it by just not doing the second preload, I think?

It's observable in resource timing, we have a WPT for that IIRC but I am not sure about the interop of this in particular, it's been a while. Happy to check.

[domenic]

Hmm OK. @nidhijaju is looking into optimizations here where we want to skip doing redundant things, but I don't think this particular case of multiple preloads showed up in the real world. (It just came to my mind while we were discussing a more realistic scenario.) So I think we'll be fine.