Add proposed mapping for in-band DASH emsg events to DataCue

[chrisn]

The explainer should describe how DASH emsg values are mapped to DataCue objects.

Consistent with the existing Webkit implementation, this could be something like:

DataCue field	emsg value
DOMString id	id
double startTime	Computed from timescale and presentation_time_delta
double endTime	Computed from timescale, presentation_time_delta, event_duration
boolean pauseOnExit	false
any value	Object containing data, schemeIdUri, and value
DOMString type	"urn:mpeg:dash:emsg" (or similar, TBD)

and

value field	emsg value	Description
ArrayBuffer messageData	message_data	Message body (may be empty)
DOMString schemeIdUri	scheme_id_uri	Identifies the message scheme. The semantics and syntax of the message data are defined by the owner of the scheme identified. The string may use URN or URL syntax
DOMString value	value	Specifies the value for the event. The value space and semantics are defined by the owners of the scheme identified by the scheme_id_uri

Does this proposal look OK?

Do we need to expose timescale, presentation_time_delta, and event_duration directly to web applications, rather than only using them to compute the cue's startTime and endTime?

For example:

• MPD validity expiration events (DASH spec section 5.10.4.2) gives particular meaning to the case where presentation_time_delta and event_duration are both zero
• SCTE 214-3 specifies an event_duration value of 0xFFFF as "unknown". How should this be interpreted?

What are the rules for placing the event on the media element's media timeline?

[technogeek00]

Through our discussions in the DASH-IF group we've aligned on the idea that the player utilizes the emsg data to compute the properties presentation_time and duration which are sourced differently depending on the emsg version used. This means we wouldn't have a burden of exposing the raw emsg values through the DataCue objects.

Technically the event specification asks to expose presentation_time and duration directly, but I'd argue startTime and endTime are sufficient if you can capture the unknown duration signalling case properly. (I think you are capturing that in #18)

The MPD expiration event signalling the presentation end with the dual zero values is an interesting case, I'm not sure we even discussed it in the DASH-IF Event TF, I'll take that back for further discussion: Dash-Industry-Forum/Events#66

On the proposal specifics, generally would it be correct to interpret that the current expectation is all emsg would be a single Track exposed for consumption? With this lens the top level attributes and value structure proposed look good.

However, if I were to map back our thoughts from the DASH-IF Event side, we generally treat the scheme_id_uri and value pairs as a standalone event track with the emsg events as a method that can carry multiple tracks. In the context of the UA parsing messages this could be tricky because you might be adding and remove Tracks constantly as emsg boxes with schemes are added and removed from the media buffer. Conceptually we've seen the model as the application forward registering for schemes that it understands, that way the player only needs to track and surface matching events.

Could I propose a similar pattern here to make emsg mapping simpler? Assuming a Type 3 playback context with the UA parsing the boxes and not the application level library. The expectation could be that the library creates the Track objects it expects to be filled from inband sources on the media element, identifying the Track with the scheme_id_uri and value pair it is looking for. As the UA discovers emsg boxes, it checks if a Track matching the scheme_id_uri and value pair has been created, if so it creates the DataCue as specified above except cue.value = message_data instead of the value structure. In this situation the application would understand the need to create a track based on the presence of an InbandEventStream element within an AdaptationSet in the MPD.

[chrisn]

would it be correct to interpret that the current expectation is all emsg would be a single Track exposed for consumption?

That's the current thinking, yes. But this does leave a gap, as compared to DASH-IF Events model, as DataCue has not defined a mechanism to let the web app tell the browser which event streams to surface (by providing the scheme_id_uri and value), and so the assumption would be that the browser surfaces all emsg boxes parsed from the media, and the web app is responsible for ignoring any that it doesn't want to handle. Is that reasonable, or do we need to provide such an API?