Re: [whatwg] exposing dependencies + weights (aka, priorities) to web developers

On Fri, Aug 15, 2014 at 8:38 AM, Ben Maurer <ben.maurer@gmail.com> wrote:

> Hey,
>
> Thanks for writing this up!
>
> So to the extent we are simply exposing http/2 semantics, this spec seems
> pretty clear. Two questions that I have that aren't answered by the http/2
> spec:
>
> - What do we expect the browser to do with priorities set cross domain. Eg,
> if I express that a.com/foo.js depends on b.com/bar.js, does the browser
> do
> anything about this. What if b.com/bar.js depends on a.com/blah.js. Would
> we tell a.com that foo.js depends on blah.js?
>

Sure, the browser will take this into account in some ways, but not others.
I wouldn't think about it too much, because given the level of the
primitive being provided, this is an abstraction boundary that authors
shouldn't cross. But yes, the browsers will leverage the information
somewhat. More or less depending on the circumstances. Sorry for being
vague :) If you need more exact controls, we probably need to expose lower
level primitives.


> - Developers might have something they would want to speculatively fetch
> only if http/2 is being used (and therefore they know the resource won't
> compete for bandwidth with other resources). Maybe this could be solved by
> simply allowing the user to tell if the browser supports http/2. A more
> advanced solution could be to allow the user to say "don't send this fetch
> request unless it's over http/2".
>

Well, your server already knows about the http/2 support from the ALPN
negotation, so it can emit different client code based on that. But if you
want the logic to be in the client, then yes, the client must be able to
detect http/2 support. There have been proposals on public-web-perf@ for
that, and this reminds me that I promised Google web devs to push again for
this. So, yeah, we should do that.


> One other thing I didn't see any mention of in the write up -- it could be
> useful to change priority while the request is in flight. http/2 supports
> this, so it'd be great if this spec could expose it.


+1, although I don't know if Ilya was trying to subsume everything into
this doc.


>
>
> On Thu, Aug 14, 2014 at 6:16 PM, Ilya Grigorik <igrigorik@gmail.com>
> wrote:
>
> > (followup / continuation of [1])
> >
> > Trying to hash out some ideas for how to connect Fetch and the new
> > transport capabilities of HTTP/2. Would love to hear everyone's thoughts:
> >
> >
> >
> https://docs.google.com/document/d/1jSpWc6jkrUoYtGWcxev9Blkkv9RhoO1XtqinBvXqhgY/edit
> >
> > ig
> >
> > [1]
> >
> http://lists.whatwg.org/htdig.cgi/whatwg-whatwg.org/2014-July/297257.html
> >
>

Received on Friday, 15 August 2014 16:32:35 UTC