Comment # 18 on bug 90887 from Ilia Mirkin

(In reply to Ilia Mirkin from comment #17)
> Created attachment 116761 [details] [review] [review]
> another attempt
> 
> OK, I like this a bit better since it doesn't use any weird maps, and incurs
> minimal costs in the common case. It unfortunately encodes the graph's
> insertion strategy, but... meh.

Ugh, actually this patch won't work. It needs to be

         phi->setSrc(j, tmp);

but then I lose the "old" phi source. So perhaps the map approach is ~required
:(