Stefan. Don't get me wrong. I absolutely agree with Mike, but not fully when it comes to that you get slow if you step out the standard library eventually. This is not always the case, but I do agree that it is utterly painful.
My divsigma is faster as the internal DivisorSigma, but cheated a bit as well, since it is compiled to C and not as flexible as the latter (Julia is jit'ing...hmmm...wonder if this is cheating really...). The next thing that shows up, is, although it is faster you loose in another place (in this case the Block definition...which is annoying and you're asking yourself why is that?) This is what I think is lovely about Julia. You can have absolute control of what code is generated etc. Currently I am lightyears away from being able to predict the cost of what I am doing, since I've to get more familiar with the semantics, but so far it is amazing. I have some mediocre decent knowledge about llvm, but should be enough to recognise where something is fishy. Btw. I think if Laszlo continues on that matter, there will be soon another factor algorithm as well :) Having said that. It is fascinating what this innocent question created. What a wonderful community. Fairly constructive indeed. Plus. I'm not getting tired on that matter... I do think a decent number theoretical package would be rather decent and it would be a nice showcase how Julia deals with 100 digits SIAM challenge. Stefan
