FYI, my suggestion was to perform ordinary wrapping, then to keep reducing the width and re-wrapping the content until doing so would cause it to increase in height.  This is more expensive than normal wrapping, since it's performing normal wrapping repeatedly, but it's just an increase in the constant factor, not to the algorithmic complexity, and binary searching the width could reduce it to just a few passes.

An example in JS: https://zewt.org/~glenn/balanced-wrapping.html

I'm confused. What's the problem with the spec here?

http://krijnhoetmer.nl/irc-logs/whatwg/20120418#l-30

This isn't CSS, so... what's the problem here?

That it's not CSS is a problem.

> I'd like balancing
> to be a CSS feature available for arbitrary content

If you want CSS to define a balancing feature, that's great, but then this is in quite the wrong component. Even if we did have such a feature, we would still want to define it more strictly for WebVTT than CSS, since that would improve interop, IMHO. So this seems orthogonal.

How does it improve interop, when the results are going to depend on the UA's particular font rendering anyway?  This is just another word-wrapping mode, and if it's useful in general and not just for WebVTT, it doesn't make sense for WebVTT and CSS to have *different* balanced wrapping algorithms.

http://dev.w3.org/csswg/css-text-4/#valdef-text-wrap-balance

Thanks, bug 28432 is to use this in WebVTT.