How are paint worklets picked?

[dglazkov]

From https://groups.google.com/a/chromium.org/d/msg/blink-dev/Jex3idOld48/XWkgUzlxBQAJ

@bzbarsky:

What the spec defines is that a UA must have at least two paint worklets
and assign work to them somehow. The spec does not require randomness
of ordering of work assignment, randomness in which worklet is chosen,
or anything else.

This unfortunately makes it fairly easy to create worklets that depend
on specific assignment algorithms, and I expect pages to end up doing
just that if browsers do anything deterministic here. At which point
we'll end up having to specify those deterministic algorithms, and
whatever their limitations are (e.g. only two paint worklets, period).

[css-meeting-bot]

The Working Group just discussed How to spec non-determinism, and agreed to the following resolutions:

• RESOLVED: There must be at least two global scopes, can't reuse the same scope more than 1k times in a row, bounds will be improved over time.

The full IRC log of that discussion

<TabAtkins> Topic: How to spec non-determinism
<TabAtkins> github: https://github.com//issues/471
<TabAtkins> iank_: How do we spec how to do non-determinism?
<TabAtkins> iank_: Proposal is to have it in a note, becaus eit's hard to test for.
<TabAtkins> iank_: foolip suggested speccing a limit that we can't use a context more than 1k times, or something
<TabAtkins> iank_: I'd prefer just having it be a note; we can't reasonable test it
<TabAtkins> dbaron: I feel like if it's something we require, it should be normative even if we can't test it.
<TabAtkins> astearns: I'm happy having examles in a note, but the "have to" part would b enormative.
<TabAtkins> iank_: So add normative text that there should be non-determinism, and give a bunch of examples of what it should look like
<TabAtkins> iank_: will give our internal impl where we switch after N calls, which is non-deterministic
<TabAtkins> Rossen: Any upper bound?
<TabAtkins> iank_: Yeah, we switch after 30 or so
<TabAtkins> Rossen: So 1k seems plenty high enough
<TabAtkins> iank_: Current text says there must be at least 2 scopes, but no language about swapping
<TabAtkins> xidorn: So if the UA discards every time...
<TabAtkins> iank_: You're passing
<TabAtkins> dbaron: But it's deterministic!
<TabAtkins> Rossen: But you can't deterministically tell what new scope you'll get
<TabAtkins> RESOLVED: There must be at least two global scopes, can't reuse the same scope more than 1k times in a row, bounds will be improved over time.

[bfgeek]

Fixed by b57b101