freenode :: #whatwg

15 Apr 2017
01:49DomenicBasically the manifest spec screwed us by setting that precedent so now we follow it
01:53DomenicAlso, why is the stylesheet still not fixed for tables -__
01:53Domenic-__-, even
04:43annevkCan't we follow CSP report precedent instead? If that is different?
06:00Domenic... huh, so it's dash-delimited. That seems... worse.
06:01Domenicalso still map CSP uses "uri"
08:47annevkYeah, I complained about that bit already, might be "fixed" by a newer version
10:48smaugannevk: this agent thing. What defines the concept which BroadcastChannel uses?
10:50annevksmaug: they are origin-bound, no? Kind of cross-cutting with agents
10:50smaugok, perhaps I don't understand what an agent is
10:51smaugor not perhaps, I just don't :)
10:52annevksmaug: a dedicated worker is an agent, or a set of windows that are similar-origin and can access each other
10:52smaugok, so that window part is like, hmm, DocGroup in Gecko
10:53smaugis there anything for non-similar-origin?
10:53annevksmaug: no, we should prolly have terms for that too at some point
10:53smaug(that would be TabGroup in Gecko)
11:02smaugand the agent for BroadcastChannel isn't needed because something. ShareArrayBuffer just isn't supported by its postMessage
11:03smaugShould one be able to pass SharedArrayBuffer cross origin?
11:03smaugusing window.postMessage
11:04smaugI would expect so, if window.postMessage supports it in same-origin case
12:18annevksmaug: no, that will fail
12:19smaugsame-origin case?
12:49annevkNo cross fails
13:40Domenicyou can pass cross-origin if the origins could become same-origin-domain via document.domain
13:40Domenici.e. you can pass from to
13:41Domenicmaybe we should add that to the list of examples
14:01annevkWe should also add a note that same-origim can be different agents
14:02annevkEg two independently opened same-origin browsing contexts
14:33knunger004hi, i'm totally out of choice now. This is about XMLHttpRequest(...) ,
14:33knunger004the issue i'm facing is regarding 307 url redirection by the proxy server at localhost:8080, when some resource isn't found.
14:34knunger004i get the following error , "Uncaught NetworkError: Failed to execute 'send' on 'XMLHttpRequest': Failed to load 'http://localhost:7860/data.json'
14:35knunger004However, when the resource is found locally, it works perfectly fine.
14:36knunger004 this is the code that prepare a custom redirect response header.
14:51annevkknunger004: that is not really sufficient information to know where it goes wrong
14:52annevkknunger004: also, stackoverflow is prolly a better place
15:06knunger004annevk, alright
17:29Domenicannevk: oooh, interesting, i didn't realize URSOBC excluded such cases
17:30DomenicThat might be a reason it's based on browsing contexts instead of realms
17:46annevkBut the whole browsing context angle makes no sense given navigation
17:47annevkOr I guess you can see it as some kind of snapshot, but still, we can do better eventually
18:01DomenicI mean, it makes sense if you think of it as a definition, instead of a set that gets statically added to and removed from
18:01DomenicgetCurrentURSOBCs() instead of const ursobcs = []
18:01DomenicIt's a bit confusing conceptually, but I think it all *works*
18:04annevkWe add state to it though
18:06annevkWe should move that state to the corresponding agent over time and then figure out when it gets created; that will make it all go away
18:07Domenicah, yeah, we did
18:07Domenicnow i'm wondering if the custom elements reaction stack should be event-loop bound instead of URSOBC-bound
18:09annevkIt should be similar-origin window agent bound
18:10annevkAnd some stuff in DOM should be too, I should export these agents
18:11Domenicit seems more like a task queue-ish thing which makes me think event loop
18:11Domenici guess in implementations it's not going to be shared across processes?
18:12DomenicBut neither are event loops, in practice, I think
18:12annevkEvent loops cannot be shared across processes
18:12DomenicI thought per spec you could have one event loop for all browsing contexts...
18:13Domenicyeah, there is at most one event loop per URSOBC, but not at least one (like I thought there was until recently)
18:13annevkSure, in a single-process impl
18:13Domenicright ok so the mapping i have wrong is agent <-> process. it&#39;s agent <-> theoretically-possible process.
18:13annevkDomenic: we don&#39;t want to export subcluster right?
18:14annevkDomenic: agent&#39;s realm and agent&#39;s global object?
18:14Domenicannevk: I&#39;d lean toward not exporting them for now, but maybe that&#39;s over-conservative.
18:15annevkseems fine to me, especially since they don&#39;t work across all agents
18:15DomenicYeah that&#39;s what&#39;s worrying
18:15DomenicMaybe they should be renamed &quot;worker agent&#39;s realm&quot; etc. Or... &quot;workerish&quot; so worklets are included? I dunno.
18:15Domenic&quot;single-global agent&#39;s global object&quot; is not short enough IMO
18:16annevkI think it&#39;s fine and we should worry about when either someone needs to use them elsewhere or someone gets confused
18:30annevkDomenic: with respect to event loop, I&#39;d like that to just become a queue or some such on each agent
18:31Domenicsure, 1:1 event loop <-> agent would work well for me
18:31annevkDomenic: and then we&#39;d allow UAs to merge queues of window agents or some such (no need to require those to be concurrent and not observable either)
18:32annevkDomenic: which brings us back to the status quo, except the formal model is more sensible (and better grounded as event loops are currently located nowhere...)
18:43annevkDomenic: other question, some of your JS PRs to remove willful violations gained consensus it seems
18:43annevkDomenic: did you lose track or just waiting for something?
18:45Domenicannevk: just hasn&#39;t gotten to the top of the priority queue unfortunately
18:45Domenicoh, PRs?
18:45DomenicI think none of my actual PRs have consensus
18:46DomenicUnless I lost track
18:46annevkDomenic:, but it needs modifications
18:46DomenicAh right yeah
18:47DomenicYeah doing that work is in a pile with a few others that got consensus
18:47Domenicbterlson was supposed to do HTMLAllCollection to show me how he envisions HTML integration working per the consensus
18:47DomenicBut he never did, so I should probably do the work myself
18:47annevkAh okay, I&#39;d just wait a bit on bterlson then
18:48DomenicNah it&#39;s been like 7 months, I gotta do it myself
18:48Domenic is updated
18:48Domenicannevk: should we specify worklet [[CanBlock]] in this PR?
18:49annevkDomenic: I decided yes, based on flagxor&#39;s feedback and nobody objecting
18:49DomenicCanBlock yes, or yes we should? It seems like it&#39;s not in the current PR
18:49annevkDomenic: but easily changed if we get some feedback
18:49Domenicannevk: oh nevermind
18:49Domenicwas looking in the wrong place
18:51annevkDomenic: seems we already had the same-origin domain iframe example too in your list of examples at the bottom of the PR btw, forgot to reply to that earlier
18:51DomenicHmm yeah I guess I did say &quot;could be&quot; and &quot;cannot be&quot;
18:52DomenicDid you want to add the &quot;two independently opened windows&quot; example?
18:52annevkDomenic: I think the note I added is good enough
18:52DomenicOh I see
18:53annevkIt seems better there, since it might directly conflict with what folks think when they see &quot;similar-origin&quot;
18:53annevkI couldn&#39;t think of a better term though
18:54annevkThe whole confusion with lth came from Gecko calling this a &quot;DocGroup&quot;
18:54annevkI think
18:56annevkDomenic: is Vat > Agent Cluster still true? I guess a Vat is the browser?
18:56DomenicI think vat is supposed to be agent
18:57annevkDomenic: hmm no, I guess Vat is Agent Cluster or Agent, okay
18:58annevkI was trying to figure out if there could be something between User agent and Agent Cluster, but that doesn&#39;t really seem possible due to BroadcastChannel and storage events
19:05DomenicHmm I don&#39;t follow the connection, but I do think there&#39;s probably not anything between them
19:23annevkI was imagining that something stronger than an Agent Cluster would not be able to use postMessage() between themselves
19:25annevkIf we call that kind of thing a Super Agent, then the problem is that as soon as you have two Super Agents, and then one of them creates a Window that has the same origin as a Window of the other, you&#39;re in trouble, as those need to be able to message each other
19:25annevkThrough BroadcastChannel and storage events
19:26annevkOh, and shared workers of course.
19:27annevkAnd service workers.
19:27annevkSo many holes.
19:43DomenicI think if we wanted to restrict sharing SABs to same-origin-domain, not just similar-origin, we would do that in the definition of agent clusters. I prefer being more permissive though.
20:10DomenicI guess an argument for more restrictive is that it helps us avoid propagating document.domain in the hope that maybe one day we could remove it or limit it to some mode
16 Apr 2017
No messages
Last message: 11 days and 19 hours ago