Tess: there are uncontroversial changes ... David & I to schedule 121?

[decide to do it here]

HTML issue 4617

Tess: this change to HTML - ties together browsing contexts and associated agent cluster. [gives 50k foot overview of agent clusters - agents are thread - clusters are process boundary]. In HTML there is a browsing context - a browsing context group are contexts able to reach each other - e.g. access to each others globals. Therefore they have to be in the same agent cluster. This change tries to tie together the ecmascript notion of agent clusters with html notion of browsing contexts. You can also have dedicated workers. A browsing context group and an agent cluster are roughly 1:1 but an agent and a browsing context are not 1:1.

... If you look at the actual diff.. what's interesting ... this adds the text that says every browsing context group has an associated agent cluster map. If you consider a browser tab with some iframes and popup windows.. relationship between these things becomes fuzzy. the process boundary tends to be at a site rather than an origin. A big chunk got removed - a hand-wavey description of how these things related and they've added in something that's much more detailed. This is a better description of how browsers work today that ties html concepts more closely to ecmascript concepts.

Dan: are there still areas where things fall through the cracks?

Tess: the followup work has covered all these cases - as of today. One of the things that this changes - it adds several algorithms that tie things together. Every agent cluster has one agent in it that corresponds to a ... window. A window has an agent but a window can also spin off things... but each agent cluster has only one agent that is a window. this change has the algo that specifies that...

... good example of a change to html that is difficult to do but not controversial.

David: how much is this a thing that .. given that there is variance in implementation .. how much room does it leave?

Tess: leaves as much as ecmascript leaves... there is leeway to have different process of models. It just clarifies the wiggle room. The idea is we