Re: TextTrackCue discussions

On Thu, Sep 19, 2013 at 12:11 AM, Philip Jägenstedt <> wrote:
> On Wed, Sep 18, 2013 at 4:05 PM, Glenn Adams <> wrote:
>> On Wed, Sep 18, 2013 at 2:42 AM, Philip Jägenstedt <>
>> wrote:
>>> On Wed, Sep 18, 2013 at 9:59 AM, Silvia Pfeiffer
>>> <> wrote:
>>> > On Tue, Sep 17, 2013 at 10:15 PM, Philip Jägenstedt <>
>>> > wrote:
>>> >> On Tue, Sep 17, 2013 at 2:11 PM, Silvia Pfeiffer
>>> >> <> wrote:
>>> >>> On Tue, Sep 17, 2013 at 6:01 PM, Philip Jägenstedt <>
>>> >>> wrote:
>>> >>>> On Tue, Sep 17, 2013 at 12:52 AM, Silvia Pfeiffer
>>> >>>> <> wrote:
>>> >>>>
>>> >>>>> UnparsedCue can be created by JS and the JS could try to add it to a
>>> >>>>> track that has a @kind=captions.
>>> >>>>
>>> >>>> Why do we need to allow scripts to create UnparsedCue at all? That
>>> >>>> will require some new API surface that isn't needed to solve the
>>> >>>> original use case -- in-band metadata tracks which aren't just a
>>> >>>> special-case of a captioning format.
>>> >>>
>>> >>> It also allows JS devs to create @kind=metadata cues without having to
>>> >>> decide to use a more specific format such as WebVTT.
>>> >>
>>> >> Given that UnparsedCue would be a strict subset of VTTCue, that
>>> >> doesn't sound at all worth adding APIs for.
>>> >
>>> > It avoids developer confusion, which is sufficient reason for me.
>>> >
>>> > Here's another use case: when a browser exposes in-band text tracks
>>> > through a @kind=metadata TextTrack, this allows developers to make
>>> > corrections to the list of cues - add cues if necessary to e.g. fill
>>> > gaps.
>>> It turns out I'm not up to date with what the spec says. addCue and
>>> removeCue are now on the TextTrack interface and take a TextTrackCue,
>>> whereas there used to be a function which created *and* added a cue. I
>>> thought we'd need a new such function, but it's only a question of
>>> whether or not UnparsedCue has a constructor.
>>> I would prefer if UnparsedCue was only ever used for in-band metadata
>>> tracks of an unknown kind, but if browsers are going to expose the
>>> UnparsedCue interface anyway, then exposing a constructor by that same
>>> makes no difference, and would indeed be more consistent.
>> Would it be better to call this interface RawCue or RawTextCue, so that it
>> doesn't necessarily imply that it is not parsed? This would be useful to
>> permit a rendered cue interface type (that exposes getCueAsHTML) to inherit
>> from Raw{Text}Cue in order to provide the text attribute.
>> I have specified a RawTTMLCue to serve this function (as a base interface
>> for TTMLCue) in [1].
> Hmm, this makes we wonder what the property for accessing the data is
> supposed to be. If it's a string, then it requires that the UA at
> least know what the encoding of the text is, which seems like it might
> not be true. RawCue and exposing the data as a typed array .data
> property seems OK to me. The other option is to base64-encode cues
> which are of an unknown encoding, I guess?

ArrayBuffer .data works for me.
Should we keep .text as well, because converting from ArrayBuffer to
String can be inefficient, see

Also, I am not fussed about the naming - RawCue works for me just like

However, I would suggest not to inherit TTMLCue from RawCue, since as
Simon pointed out earlier, you'd like JS devs to be able to do

   var raw_cue = cue instanceof RawCue;

to detect that a cue is not taken care of by the UA. If TTMLCue
inherits from it, that doesn't work.


Received on Friday, 20 September 2013 03:22:43 UTC