Hi Aaron
Thanks for your test code - it was very helpful in fixing a few more bugs. The fibonacci heap code is based very closely on Knuth's implementation from "The Stanford GraphBase" - but he uses some bit-level tricks, a few of which got lost in the translation.
I just committed a fix to HEAD that I hope clears up the remaining bugs. Will you please pull it down and let us know? It would be great to have the fibonacci heap working to compare against mutable_queue and relaxed_heap.
After doing some tests with the new code you committed I integrated it into my program. As far as I can see it seems to work perfectly well! Having an assessment routine computing dijkstra's shortest path nearly 4000 time on different graphs I get *exactly* the same results I got before with my old implementation -- just faster :) Thank you very much for your fast response this takes me a big step further. Best wishes Florian