Re: Unifying the rendering approach

YouTube-generated WebVTT always uses non-snap-to-lines cues. And we already
struggle to deal with the different levels of WebVTT support in the
browsers we need to work with. Removing non-snap-to-line cues outside of
regions will make this situation much more difficult for us.


On Fri, Mar 7, 2014 at 6:16 AM, Victor Carbune <victor.carbune@gmail.com>wrote:

> On Wed, Mar 5, 2014 at 12:03 PM, Silvia Pfeiffer
> <silviapfeiffer1@gmail.com> wrote:
> >
> > On 5 Mar 2014 18:29, "Victor Carbune" <victor.carbune@gmail.com> wrote:
> >>
> >> On Tue, Mar 4, 2014 at 10:46 PM, Silvia Pfeiffer
> >> <silviapfeiffer1@gmail.com> wrote:
> >> >
> >> > On 4 Mar 2014 20:58, "Victor Carbune" <victor.carbune@gmail.com>
> wrote:
> >> >>
> >> >> On Mon, Mar 3, 2014 at 11:55 PM, Silvia Pfeiffer
> >> >> <silviapfeiffer1@gmail.com> wrote:
> >> >> >
> >> >> > On Tue, Mar 4, 2014 at 12:34 AM, Victor Carbune
> >> >> > <victor.carbune@gmail.com> wrote:
> >> >> > > On Mon, Mar 3, 2014 at 12:53 PM, Silvia Pfeiffer
> >> >> > > <silviapfeiffer1@gmail.com> wrote:
> >> >> > >> On Mon, Mar 3, 2014 at 10:41 PM, Victor Carbune
> >> >> > >> <victor.carbune@gmail.com> wrote:
> >> >> > >>> On Mon, Mar 3, 2014 at 12:11 PM, Silvia Pfeiffer
> >> >> > >>> <silviapfeiffer1@gmail.com> wrote:
> >> >> > >>>>
> >> >> > >>>> Aha! I see. The first case is so as to keep the line counting
> >> >> > >>>> correct
> >> >> > >>>> for snap-to-lines cues, I assume? Couldn't we make these two
> >> >> > >>>> cases
> >> >> > >>>> into a single case if the line positioning both for
> >> >> > >>>> snap-to-lines
> >> >> > >>>> and
> >> >> > >>>> for non-snap-to-lines is done on the anonymous region that
> wraps
> >> >> > >>>> each
> >> >> > >>>> cue? What's the advantage of splitting these two cases?
> >> >> > >>>
> >> >> > >>> If we throw non-snap-to-lines cues within regions it means that
> >> >> > >>> we
> >> >> > >>> need to support a rendering case for these cues within regions,
> >> >> > >>> and
> >> >> > >>> also support named regions on them.
> >> >> > >>
> >> >> > >> I don't think so, since it will be the region that is placed,
> not
> >> >> > >> the
> >> >> > >> cue. So, the cue inside the region is still placed
> "snap-to-line"
> >> >> > >> even
> >> >> > >> if the line is basically just a single line (minus line wrapping
> >> >> > >> and
> >> >> > >> newlines).
> >> >> > >
> >> >> > > Well, it's one thing to deal with snap-to-lines, where you only
> >> >> > > move
> >> >> > > one line on top of the other until they don't overlap, and
> another
> >> >> > > one
> >> >> > > is to deal with overlap between a percentage-positioned cues
> >> >> > > together
> >> >> > > with line-positioned cues; moving lines is simple and
> >> >> > > straightforward.
> >> >> >
> >> >> > Correct. I don't see how this is relevant though. If we give all
> >> >> > non-region cues their own anonymous region box, then we never have
> to
> >> >> > worry about cue overlap inside regions. All we have to worry about
> is
> >> >> > region overlap.
> >> >> >
> >> >> > Was your intent to separate overlap avoidance for the percentage
> >> >> > positioned non-region cues from overlap avoidance of the regions?
> >> >> > That
> >> >> > would potentially cause overlap between non-region non-snap-to-line
> >> >> > cues and snap-to-line cues (in regions), right? Are you suggesting
> >> >> > not
> >> >> > to deal with that? Would we even do overlap avoidance for regions?
> >> >>
> >> >> I want to avoid solving overlap avoidance between non-snap-to-lines
> >> >> and snap-to-lines cues by:
> >> >> *) ensuring they never end up in the same region (thus, I don't see a
> >> >> need to support non-snap-to-lines cues with author-specified regions,
> >> >> there's no use-case for this situation)
> >> >> *) deferring the overlap avoidance mechanism to regions.
> >> >
> >> > Agree. That's why I wouldn't want all non-snap-to-lines cues end up
> in a
> >> > single full-viewport-sized region.
> >> >
> >> >> > >>> *) No need to think what happens if some percentage-positioned
> >> >> > >>> cue
> >> >> > >>> overlaps a line-positioned cue (see "underspecced overlapping
> >> >> > >>> positioning" bug)
> >> >> > >>
> >> >> > >> We still have to deal with overlapping cues, no matter whether
> >> >> > >> they
> >> >> > >> are in snap-to-lines regions or in non-snap-to-lines regions.
> >> >> > >
> >> >> > > This would move to dealing with overlapping regions - which we
> >> >> > > decided
> >> >> > > we don't want to support, right? Or at least differ it to a
> higher
> >> >> > > level mechanism that would deal with all the caption boxes from
> any
> >> >> > > format ending up on the screen.
> >> >> >
> >> >> > Hmm... I thought we didn't want to deal with overlap for
> region-cues.
> >> >> > But you're now also saying we don't want to deal with overlap for
> >> >> > non-region snap-to-line cues. I don't think that was the intention.
> >> >>
> >> >> We need unification: imagine, exaggerating here, having
> >> >> {snap-to-lines, non-snap-to-lines} x {region, non-region} type of
> >> >> cues.
> >> >>
> >> >> One solution is for all cues to end up in regions, anonymous or
> >> >> author-specified, for rendering purposes.
> >> >
> >> > Yes, that's the best approach IMO.
> >> >
> >> >> > I can imagine a single overlap avoidance algorithm that works on
> >> >> > lines
> >> >> > only where for percentage-positioned cues a line is deemed occupied
> >> >> > if
> >> >> > a part of a cue is in it.
> >> >> >
> >> >> >
> >> >> > >>> *) Better abstraction: author can already obtain exactly the
> same
> >> >> > >>> positioning using regions that they can with
> >> >> > >>> percentage-positioned
> >> >> > >>> cues. Why integrate two different elements solving the same
> >> >> > >>> problem
> >> >> > >>> together, if we can keep only one?
> >> >> > >>
> >> >> > >> Because it avoids another big case statement in the rendering
> >> >> > >> algorithm. This way we have all three cases in one branch rather
> >> >> > >> than
> >> >> > >> 2 different branches. Also, this is just about the rendering,
> >> >> > >> since
> >> >> > >> we're still keeping the two different ways of specifying
> >> >> > >> positioning
> >> >> > >> (cues with line cue setting and cues inside regions).
> >> >> > >
> >> >> > > Wouldn't this simply be something like: if non-snap-to-lines=true
> >> >> > > create on the fly an anonymous region, render the text in it
> >> >> > > according
> >> >> > > to the rules in "paragraph where layout in a region is done" and
> >> >> > > then
> >> >> > > resize the anonymous to perfectly match the cue and set the
> region
> >> >> > > positioning parameters accordingly?
> >> >> >
> >> >> > Hold on. Earlier you said that all non-snap-to-lines cue will be
> >> >> > rendered in a single anonymous region that covers the full
> viewport.
> >> >> > What you are instead describing here is the rendering approach for
> >> >> > snap-to-lines-cues.
> >> >>
> >> >> This was for snap-to-lines cues with no author-specified region.
> >> >
> >> > Oh! But then you can't do overlap avoidance with these cues either.
> >>
> >> Well if all the snap-to-lines cues without an author-specified region
> >> go into the same anonymous region of the size of the video, then you
> >> are just using the cue snap-to-lines positioning algorithm to do
> >> overlapping.
> >
> > I didn't mean for cues to share a region unless they were authored with a
> > region.
> >
> >> > I'd rather they go into  individual regions, too, and are all dealt
> with
> >> > by
> >> > a single overlap avoidance approach that works on regions.
> >>
> >> Then how do you honor line positioning for a cue that has no region,
> >> and has line:3 attribute? You will have to make position the region in
> >> line 3 of the video viewport, rather than the text lines of the cue in
> >> a region.
> >
> > Correct. That's what I thought we are doing with all cues now.
>
> The cleanest way to me looks like having regions always absolutely
> positioned within the video viewport and cues always snapped to line
> within a region.
>
> On Fri, Mar 7, 2014 at 11:27 AM, Philip Jägenstedt <philipj@opera.com>
> wrote:
> > On Mon, Mar 3, 2014 at 6:41 PM, Victor Carbune <victor.carbune@gmail.com>
> wrote:
> >
> >> A more personal comment: I feel that non-snap-to-lines cues are hard
> >> to use for authors that want to actually position things precisely on
> >> top of the video, and I'm not aware of other use-cases for it, so I
> >> would even go as far as removing them as soon as we support regions.
> >> But since they are already here, we can easily keep them for
> >> backwards-compatible purposes with wrapped anonymous regions
> >> fulfilling the same positioning behavior.
> >
> > If we go down this route I think we should try just removing the old
> > way. I could add use counters to Blink to see if it's still possible.
> > Would counting VTTCues where snapToLines is false be enough, or would
> > something about position/size/align also change?
>
> I'm certainly in favor of this, but I'm sure on this list there might
> be other vtt-users that are able to tell how important
> percentage-positioned cues in their current form are.
>
> Victor
>
>