[whatwg] Limiting the amount of downloaded but not watched video

[Glenn Maynard]

On Sun, Jan 23, 2011 at 6:32 AM, Philip Jägenstedt <philipj at opera.com>wrote:

> But presumably you want some kind of guarantee that the video will be able
> to keep playing without waiting for the network, right? So if you don't use
> preload=auto, you'll at least need preload=playthrough or similar. Maybe
> that's precisely what you mean by preload=buffer?


Yes, I think we all mean the same thing here: asking the browser to buffer
enough to avoid underruns and to begin playing immediately without blocking,
which is more than preload=metadata and less than preload=auto.



> I intend to make that impossible by only allowing scripts to increase the
> effective buffering strategy, because:
>
> 1. preload->none makes no sense.
> 2. I haven't seen any use cases for auto->preload other than working around
> browser buffering bugs.
> 3. It would be a silly API, because you can't reasonably mean "I need it
> all! No, I need only metadata! No, I need it all!", you're clearly trying to
> do something else.
>

I think it makes sense, with the right definitions.  preload=none means
"don't preload any more data than you already have".  preload=metadata means
"load metadata if you havn't already".  These make sense regardless of what
the state was previously.

By not catering to that, we can instead focus on fixing the
> API/implementations to support the actual use cases authors might have.


The most important unresolved use case is: how to allow limiting the amount
of prebuffered data, while also having a mechanism to disable that limit
when there isn't enough bandwidth.  The approaches I've seen proposed are 1:
have the browser try to guess statistically how big the buffer size needs to
be, and 2: disable the buffer limit when paused.

I don't trust #1.  It's impossible to predict bandwidth reliably.  If the
browser guesses too small, it'll be impossible to watch a video without
underruns.  Browsers should try to make a good guess, but there's got to be
an intuitive user override.

#2 is essentially that override, and it's what everyone is used to already:
if the video is underrunning, leave it paused until it buffers for a while.

However, there are cases where you'll want to pause the video without
triggering #2; for example, YouTube's pause captions, and after a
pause-and-rewind.  Simply having something like
preload=auto_only_when_paused wouldn't be enough.

That's why it seems like scripts should have some control over this.
Toggling between preload=metadata and preload=auto is one possible API to do
that.  Granted, browser-supplied controls should also be able to implement
this behavior, which suggests an attribute to allow it.  I'm not sure how to
get both cleanly.

-- 
Glenn Maynard