Move instead of clone shadow tree children

Move instead of clone shadow tree children

Move instead of clone shadow tree children when expanding <symbol> and <use>. That way we avoid
deleting perfectly fine shadow tree nodes.

I see a 38% speedup of Sierpinski perf test on my linux desktop, others are unchanged.

BUG=313438

[rwlbuis, 2014-05-09 18:24:17]

PTAL. Not needed for overall bug but this is more efficient.

[pdr., 2014-05-09 21:10:11]

On 2014/05/09 18:24:17, rwlbuis wrote:
> PTAL. Not needed for overall bug but this is more efficient.

I see how this would be a benefit but I think it adds some tricky logic. I'd
prefer to wait until we can prove the perf benefit for this one :/

[Erik Dahlström (inactive), 2014-05-12 08:08:08]

On 2014/05/09 21:10:11, pdr wrote:
> On 2014/05/09 18:24:17, rwlbuis wrote:
> > PTAL. Not needed for overall bug but this is more efficient.
> 
> I see how this would be a benefit but I think it adds some tricky logic. I'd
> prefer to wait until we can prove the perf benefit for this one :/

I'd be curious to see how this would work out with animations, event handlers
listening for tree mutations and such. It might be possible to do this only in
some scenarios, if it shows major perf benefits.

Ideally there wouldn't even be a need for cloning elements at all, but that
implies some other way of dealing with the instance tree state.

[rwlbuis, 2014-05-20 18:23:13]

On 2014/05/09 21:10:11, pdr wrote:
> On 2014/05/09 18:24:17, rwlbuis wrote:
> > PTAL. Not needed for overall bug but this is more efficient.
> 
> I see how this would be a benefit but I think it adds some tricky logic. I'd
> prefer to wait until we can prove the perf benefit for this one :/

Understood. Now that I look at it again, it does gain a lot on Sierpinski test
on my desktop, 38%. I'd like to see more measurements first though.