[MERGE] Handling frozen set inputs in mpdiff generation when a key is missing

[Andrew Bennetts]

Robert Collins wrote:
> On Thu, 2008-07-17 at 14:37 +1000, Andrew Bennetts wrote:
[...]
> 
> Yes, it was better :). I tested with a 334MB GraphIndex that would not
> parse at all with 2GB of ram, it made a modest improvement in memory
> use.
> 
[...]
> > That said, it's a fairly subtle tweak purely for performance reasons, and so it
> > probably deserves a brief comment explaining why it's there.
> 
> Sure.

bb:approve

-Andrew.