W3C

– DRAFT –
WebRTC June 2023 meeting

27 June 2023

Attendees

Present
AlfredHeggestad, Bernard, Carine, ColinRead, Dom, Elad, Fippo, Florent, FredericWang, Guido, Harald, Henrik, Jan-Ivar, JaredSiskin, PatrickRochill, PeterThatcher, Sameer, SunShin, TimP, Youenn
Regrets
-
Chair
Bernard, HTA, Jan-Ivar
Scribe
dom

Meeting minutes

Recording: https://youtu.be/RWGJX4MHIuo

Slideset: https://lists.w3.org/Archives/Public/www-archive/2023Jun/att-0003/WEBRTCWG-2023-06-27.pdf

Mediacapture-screen-share 🎞︎

Issue #268 Make CaptureController inherit from EventTarget 🎞︎

[Slide 12]

Elad: I suggest making capturecontroller an eventTarget help making that object more useful - with a specific use case inspired by the screen capture mouse events
… can easily think of more use cases

[+1 from Harald, Henrik, Youenn]

JIB: LGTM

RESOLUTION: merge PR for #268

Issue #263 Improve upon CaptureStartFocusBehavior.no-focus-change 🎞︎

[Slide 13]

Elad: this was discussed in the past and we decided to let the app express a preference that the UA is free to take into account or ignore
… no-focus-change is ambiguous given Safari's model with MacOS windows picker

[Slide 14]

Elad: proposed to add a new value "focus-capturing-application" and keep "no-focus-change" to be platform-dependent, possibly to be deprecated in the future

youenn: I like "focus-capturing-application"; maybe we could already add a warning about "no-focus-change" that it will be deprecated (and leave it to implementations to figure out a deprecation schedule)

Elad: not sure if we want to commit to deprecate right away; will want to look at web compat

TimP: would support deprecating ASAP

Elad: right, need to consult data usage before committing

JIB: for other implementors, knowing whether we implement 2 or 3 values is important
… otherwise, we would have to throw on unrecognized values

Harald: if we have deployed code that uses this, the usual magic is to not break running code which would require some deprecation timeline

Youenn: would be best to state the direction in the spec clearly, knowing that implementations will need to adjust over time

TimP: I wonder if this could be solved by feature detection and ensure that only 2 are ever implemented in a given browser

JIB: this would be concerning for web compat

Elad: +1
… there may be value for "no-focus-change" in itself, e.g. for accessibility to reduce change
… I propose we start by adding "focus-capturing-application" and have a separate conversation on deprecation

[JIB: +1]

Youenn: +1, as long as we converge reasonably quickly

Issue #261 Allow apps to avoid riskier display-surface types 🎞︎

[Slide 15]

[Slide 16]

[Slide 17]

Youenn: dynamic changes might be tricky to manage - auto-pause might be a solution
… combined with the current preference

Elad: when the user is offered the current screen, they may not understand they shouldn't which can create bad user experience
… there isn't any hint to disable that
… monitorExclusion would follow the footsteps of selfBrowserSurface

Youenn: I think a preference would be better than a hard-set requirement

Elad: the intent is for this to be a hint to the UA that it could ignore
… in Chrome's case, it would remove the "monitor" option

Youenn: in MacOS, the user can dynamically change to the screen - this wouldn't under the control of the browser UI
… which would create inconsistencies

Elad: this would reduce the number of clicks in the Safari workflow
… the OS itself could choose not to expose the monitor with that hint

TimP: I like this in principle, but think it would be hard to expose it in a non-confusing way to users
… this will create unexpected variations for users from one meeting to another

Elad: this wouldn't be more confusing than shutting down abruptly an ongoing capture

TimP: still, users will ask "why can't i share my screen when I could on my previous call?"

Elad: the app doesn't have to use that hint if it finds it too hard to communicate to end users

JIB: the use case makes a lot of sense; I support this; while I dislike limiting user choice, sharing full screen is risky which makes me supportive
… what would be the default?

Bernard: we're running out of time

Elad: let's follow up on github

Requesting keyframes via setParameters (WebRTC Extensions) 🎞︎

[Slide 20]

Fippo: this proposal would rely on WebCodecs-defined WebIDL - how do we feel about this?

JIB: I like that proposal direction; not sure about the value of reusing WebCodecs IDL which may evolve in ways that wouldn't work for us

Fippo: true - that's already the case since there are codecs-specific fields already which we wouldn't want to import

Florent: we would still want to keep encoding options in sync with WebCodecs when they make sense

Youenn: WebCodecs is per frame when setParameters isn't - I prefer a separate dictionary, but keep the definition aligned with WebCodecs

JIB: if I specify false - what does that mean?

Fippo: you're not requesting keyframes

JIB: and setParameters() with no change but keyframe true, I get a keyframe?

Fippo: yes

JIB: a bit odd of an API, but it makes sense for synchronicity

Florent: setParameter is supposed to resolve the promise when all the params have been applied
… how would this sequenced with the keyframe?

Fippo: we can't know when a keyframe would be generated; and it gets more complex with several layers
… so we wouldn't wait for the keyframe to resolve the promise

Florent: SGTM

Fippo: I'll update the PR in that direction, with a similar but different object than WebCodecs

RESOLUTION: use second parameter with a similar but different dictionary than WebCodecs, clarify promise doesn't wait for the keyframe to resolve

WebRTC Extended Use Cases 🎞︎

[Slide 23]

Remove Use Cases That Don’t Add New Req’ts Pull Request #112 Pull Request #113 🎞︎

[Slide 25]

TimP: when this was raised, it was brought up this could be achieved with a JS library, which WHEP has kind of demonstrated
… unless anyone can think of a use case where UA assistance would be needed
… (although WHEP hasn't been implemented yet)

Bernard: in future meetings we would discuss streaming which relates to WHIP and WHEP as well
… in the meantime, this use case doesn't bring any requirement - any pushback on removing it?

TimP: the only reason would be to validate this is a valid usage of WebRTC

Bernard: the fact that WISH took this up kind of validates this (and they don't need our validation)

JIB: I support our use cases should only drive decisions in our WG

Dom: I think keeping track of edge usages is interesting, but I'm not sure the WG is equipped for that
… and this document should really focus on use cases that generate new requirements

Harald: not sure; but not strong objection either

RESOLUTION: remove section 3.9

[Slide 26]

Youenn: related to N22 - we should look at what we're defining in the WG
… the only thing we need to define is efficient access to a VideoFrame

Dom: remaining question for me is whether there are memory-copy reduction requirements the WebRTC WG would need to cover

Bernard: this is being worked on, but in the Media WG

TimP: yes, I think this has been overtaken by events (i.e. it is now available)
… I don't think there is anything left for us - even though clearly this is something that WebRTC supports
… ensuring proper integration with ML is definitely important, but maybe no longer in our scope

JIB: N22 is a superset requirement that satisfies funny hats & ML

Bernard: it doesn't really satisfy fully ML

JIB: I was going to suggest a requirement about "processing" rather than "manipulation"

Harald: I think the requirement was misguided in tying it to GPU
… not all media manipulation needs or benefits GPU

Bernard: typically audio doesn't go through GPU
… so N22 should be revised

RESOLUTION: remove ML use case section 3.7 #PR 113

Bernard: we'll also update N22 for funny hats

Process changes 🎞︎

[Slide 31]

TimP: how do we shape the doc moving forward? seeking guidance on the relationship between explainers and the use case docs? conversely, should we ensure API changes tie back to use cases?

Dom: a possibility would be to remove use cases & requirements that already have a home in a spec/explainer?

JIB: an explainer fulfills a different role - having early use cases remain useful

Dom: right, but they could be still be sequenced

JIB: maybe

Bernard: conversely, would we want to require use cases for new API proposals?

Dom: we already have a goal of having API proposals be accompanied by explainers that have fairly detailed use cases

TimP: then this document would be a queue of future use cases without backing proposals? This sounds workable

Harald: it often feels easier to focus on an explainer with a specific proposal than to get consensus on an addition to the use case doc
… or we need a lower bar of entry to the document

TimP: there are a bunch of developer requirements that are impossible to implement without getting implementers on board

IceController 🎞︎

Repository: w3c/webrtc-extensions

Issue #166 PR #168 - prevent candidate pair removal 🎞︎

[Slide 34]

Sameer: please keep feedback coming on #168

Issue #170 #171 - candidate pairs management 🎞︎

[Slide 37]

[Slide 38]

[Slide 39]

JIB: if you use a promise, would there also be an event?

Sameer: the event already exists, so it would have to be fired as well

JIB: with pruning and preventDefault, this could get awkward and footgunny

Sameer: prune() would not fire an event - it prunes it immediately without an event

JIB: we should be consistent; re promise, can it fail?

Sameer: in other additions we're considering, there is an event for deletion
… the only possible failure I can think of is pruning a candidate pair that doesn't exist

JIB: I meant fail async - input validation can be done sync

Sameer: can't think of anything of failure for prune()
… for setSelected(), there may be async failure cases depending on the implementation approach

Youenn: is it fine to prune the currently selected candidate pair? if we don't allow it, it would have to be async

Sameer: pruning the current selected pair is the equivalent of a pair going away for any other reason (e.g. network interface going down)

Youenn: but this could lead to situation of pruning the selected candidate pair without realizing it due to race conditions
… why does prune() take a sequence? is it for an optimization? (could use variadic arguments)

Sameer: indeed, that's to optimize it

TimP: +1 to allow for several pairs
… I'm slightly worried that the API isn't taking account the asymetry of control

Sameer: setSelected fails immediately if called on the controlled side

TimP: maybe other asymetricalities in the timing of things; maybe to discuss on a specific PR

Peter: I think we should allow the controlled side to override to set a selected pair, but it should be an explicit call

Sameer: that might make sense, indeed
… any thoughts selection by directly sending media vs doing an exchange?

JIB: I'm nervous that the ICE Transport is running async and with PC being a huge state machine - there could be a lot of races e.g. when considering ICE restart
… would these decisions be reversed by an ICE restart?

Sameer: I expect ICE restart would be a clean slate

JIB: I'll need to think more about potential races

TimP: re ICE round trip, it should happen, to the risk of messing up bandwidth estimation

Peter: wrt ICE restart - ICE restart is make-before-break, typically adding candidate pairs
… would newly added candidate pairs be able to override already selected pairs?
… I'm inclined the app should be in control
… There is no reason both sides would need to use the same pairs
… I would support sending media right away when selecting a pair; I don't see a good reason to wait for another roundtrip

Encoded Transform Codec Negotiation 🎞︎

Repository: w3c/webrtc-encoded-transform

[Slide 42]

[Slide 43]

[Slide 44]

[Slide 45]

Bernard: what does it mean "not to know about it", or "to know about it"?

Harald: the UA has to know about the packetization mode

Bernard: so having a WebCodec decoder counts as "known"?

Harald: no, this is in the context of the WebRTC chain

[Slide 46]

Harald: I re-use mime types for packetization mode - in most cases, you would want to use a simple packetization mode (not H264)

[Slide 47]

Harald: note the payload type uses out-of-range SDP payload types

[Slide 48]

[Slide 49]

Harald: have filed PR #186

[youenn on the chat: My main question is on which IETF work would be needed here. payloadType setter on a frame seems fine, I am less sure about the other API.]

Harald: I don't think any IETF work is needed, as long as we recognize we would be using non-standardized mime type
… for packetization

JIB: this makes sense; I see there are pre-negotiation methods, then a negotiation would happen where the other side may reject a payload type
… would we need new APIs to set codecs?

Harald: I think we should align with what Florent is proposing

JIB: +1

TimP: much to my chagrin, I see the need to involve SDP in this
… how will we deal with scope rules of SDP, RTX, etc?

Harald: I haven't figured that out yet
… at the moment, we have a very imprecise surface for deciding what kind of protection features we want to turn on
… that's done through SDP munging, which is sad
… we should figure that out, and apply it to this case as well
… (deferring to a previously unsolved problem in other words)

JIB: there are only methods to add things, not remove; maybe this could be part of setConfiguration?

Harald: I thought about adding/removing - we're able to turn off codecs by removing them with setCodecs
… I prefer to focus on well-known needs, and small-scoped APIs rather than extending setConfiguration

Harald: any sentiment on whether we should adopt this?

[JIB: +1]

[Jared, Bernard, Guido: +1]

RESOLUTION: Adopt PR #186 with details to be discussed in the PR

JIB: Youenn's comment on IETF?

Bernard: there is a proposed work item in front of the IESG in this space (SKIP?)

Summary of resolutions

  1. merge PR for #268
  2. use second parameter with a similar but different dictionary than WebCodecs, clarify promise doesn't wait for the keyframe to resolve
  3. remove section 3.9
  4. remove ML use case section 3.7 #PR 113
  5. Adopt PR #186 with details to be discussed in the PR
Minutes manually created (not a transcript), formatted by scribe.perl version 208 (Wed Dec 21 15:03:26 2022 UTC).