Add the Priority Hints changes to the fetch spec

[pmeenan]

This applies the specified changes from the Priority Hints spec to fetch for #1319.

The main changes are:

• Rename the existing implementor-internal priority to internal priority.
• Add a new priority enum to the Request and RequestInit objects.

• At least two implementers are interested (and none opposed):
  • Chrome (Shipped in 103)
  • Mozilla
• Tests for all of the interfaces are already being run as part of the Priority Hints spec:
  • https://github.com/web-platform-tests/wpt/blob/master/interfaces/priority-hints.idl
  • https://github.com/web-platform-tests/wpt/blob/master/priority-hints/fetch-api-request.tentative.any.js
  • https://github.com/web-platform-tests/wpt/blob/master/priority-hints/service-worker-fetch-request-priority.tentative.https.html
• Implementation bugs are filed:
  • Chromium: https://bugs.chromium.org/p/chromium/issues/detail?id=821464
  • Gecko: https://bugzilla.mozilla.org/show_bug.cgi?id=1797715
  • WebKit: https://bugs.webkit.org/show_bug.cgi?id=247291
  • Deno (not for CORS changes): Add support for Priority Hints denoland/deno#16502
  • Node.js: Add support for Priority Hints nodejs/undici#1748
• MDN:
  • Already documented as experimental from the existing spec, can be moved to non-experimental after merged.

---

Preview | Diff

[pmeenan]

From what I understand, Mozilla will be implementing it so I'll ping them to commit to the implementation on the issue.

[valenting]

From what I understand, Mozilla will be implementing it so I'll ping them to commit to the implementation on the issue.

That is correct. We are tracking this in https://bugzilla.mozilla.org/show_bug.cgi?id=1797715

[annevk]

Thank you for working on this!

I'm not seeing the outcome of WICG/priority-hints#69 incorporated here.

We might also want to say a bit more about "internal priority".

Please also draft a commit message (as a comment) that links the appropriate HTML changes and makes it clear that all of this essentially obsoletes the Priority Hints specification. (I'm assuming that's the intent given there's no normative references.)

[pmeenan]

@annevk I think I got most of the changes done and the last commit message includes the obsoleting language. I added a comment to the render-blocking issue to discuss it a bit more before we figure out what the changes mean for the spec (needless to say, this merge will wait on the outcome of that thread).

[pmeenan]

I removed priority from the exposed Request attribute interface and have it as internal-only as discussed here.

I also have render-blocking and priority both being copied along with internal-priority when constructed from an existing Request in step 12 of https://fetch.spec.whatwg.org/#dom-request (let me know if you'd rather not explicitly require that they be copied).

[annevk]

What will happen to the Priority Hints specification? Does it essentially disappear?

[pmeenan]

As far as the existing Priority Hints spec goes, once both HTML and fetch are updated I'm planning on removing most of the content but leaving a placeholder with some high-level details and links to the relevant parts in the HTML and fetch specs. That way any external links in blog posts and articles will still find their way to the specs with the relevant details.

[annevk]

The Request constructor is not correct:

• It copies priority rather than internal priority.
• When priority is given this is not taken into account. I suspect that when it's given we want to set request's priority to it and we want to set internal priority to null (so request's priority will be used upon fetching to initialize it). Or would we want to adjust internal priority instead in some manner so the existing prioritization information is taken into account?

[pmeenan]

Thanks, looks like I accidentally deleted internal priority instead of priority from the copy step. Should be fixed now.

I also added a step 27 to the constructor to take init's priority into account in setting internal priority (I don't think we can be more specific than that for how each browser will choose to modify or override an existing internal priority).

[annevk]

Thanks! This looks good to me. I have a couple final nits I can take care of when merging.

Where is the corresponding HTML PR again?

Could you post a suggested commit message in a new comment that includes all the relevant details and pointers to issues/documents/PRs? You can use https://github.com/whatwg/meta/blob/main/COMMITTING.md for guidance.

I think this can be merged Wed or Thu next week assuming it all checks out.

[pmeenan]

The HTML PR is here (#8470) (though I still have a bit of updating from Domenic's feedback and I assume a few rounds of tweaks after that before it can land).

The latest commit (afce7fc) has a commit message with the summary details of the change and links to the Issue and previous spec.

[annevk]

Tests:

• Let's have a non-throwing test as well for Request and fetch(). (You can use about:blank as URL to avoid it taking up more resources than needed as I guess we cannot actually test priority that well?)
• Maybe: test that Request doesn't have priority.
• Fetch tests need to be inside fetch/.
• /interfaces/priority-hints.idl needs to be removed as it'll be obsoleted by Fetch and HTML. @foolip how can we stop Priority Hints from being imported?

[annevk]

The test situation still needs a bit of work, but once that is done I think this can be merged. It seems okay that HTML will be merged later as it builds on top of this.

[pmeenan]

I have a CL up with the changes to the tests (will update here after it lands). Since the effects of priority are all implementation-specific and not exposed, other than not erroring when they are set there's not much we can test.

As far as the IDL import, I pinged @foolip to see if there's a workflow to stop importing it, but I can also remove the fetch integration section from the spec which has the IDL bits (and remove the HTML part when HTML is ready to land). That should also avoid any conflicts. Main question is if I remove them before the fetch changes land or wait until they have landed and there is content to point to from the Priority Hints spec.

[foolip]

IDL files in WPT come from https://github.com/w3c/webref, and there's a consistency check that runs before releasing @webref/idl. In other words, you don't need to worry about a two-sided spec change making the sum of all IDL inconsistent for a period. Ideally that only lasts for a few days, but a broken set of IDL won't be imported into WPT.

Hope that helps?

[annevk]

I decided to merge this because of:

• @foolip's comment
• triaging https://github.com/WICG/priority-hints/issues didn't yield anything new
• tests are written up and reviewed by Domenic

I hope @pmeenan can sort out the WPT IDL situation and make progress on the HTML PR next.

Thanks @pmeenan for your work on this feature!