The Minkowski's challenge

While mutating at the symbol and grammar level and allowing the potential for illegal symbols and syntax is a possible approach, mutations at the underlying semantic level would be much more productive as far as simple evolution is concerned. But if you're studying certain other aspects of evolution, such as the evolution of error tolerance, it might prove to be a very productive approach.Barbarian will have to confirm, but my understanding of his approach is that the instruction is the unit of mutation, not the individual bits coding for machine instructions. I don't think his abstract machine includes the concept of bit-encoded machine-level instructions, but I'm sure he'll let us know.--Percy

I can host, but the next couple weeks aren't good for me time-wise.--Percy

If you're spending a lot of time constantly allocating/deallocating same-sized blocks, create your own free-list of these blocks. Still, you say that's only 15% of the total, which means it's a meager performance opportunity.What are the other major performance consumers?For no local function to take more than 1% sounds extremely unusual. Especially in so repetitive an application you would expect a fairly expensive core. Are you sure of your profiling results?--Percy

Could Objective Caml itself be a performance bottleneck? I'm not myself familiar with the language, and in the five minutes I gave myself to study up I was unable to discover whether compilation produces actual object code or just preparsed scripts that still have to be interpreted. The frequent focus of Caml articles on performance, with frequent mention of such issues as avoidance of run-time type checking, leads me to believe that it is interpretive. If this is the case, then even the couple positive statements I saw claiming the performance is pretty good ("50% of a decent C compiler" was one comment) are not enough to persuade me away from the possibility that your biggest performance bottleneck could well be your choice of language.--PercyEdited by Percy, : Fix garbled last sentence.

Am I misinterpreting your table when I think it's telling me that 85% of your time is OCaml overhead? If so, are the services that the runtime OCaml system is providing of sufficiently high value to justify that?Is compiled OCaml known to be much, much faster than interpreted Perl? A couple years ago I tried to rig up something real quick because it was so easy in Perl, and that's when I discovered that Perl is less than a hundredth as fast as compiled C++. All that OCaml runtime stuff in your performance profile seems odd if it's really producing object code, and runtime reference tracking for GC is expensive. Taking a look at Objective CAML Tutorial, my guess is that, yes, compiling produces object code, but most of the object code is just calls to runtime system routines. The performance claims no doubt hold up extremely well for Caml code like "let max a b = if a > b then a else b;;", but I have my doubts about anything requiring calls to the runtime system.But it sounds like you're not really worried about performance. I guess something you said must have made me think you were, but if you're not worried about it then I'm certainly not.--Percy

To Jazzns and Barbarian,Okay, okay, I've been remiss. Yes, I can host. I promise to try to find time tonight to set up a subdomain for Jazzns, this will give him the ability to take things from there. Takes max of 24 hours for DNS name to propagate through the Internet, though.--Percy

The subdomain jazzns. has been created, but the DNS entry probably won't have propagated until tonight or tomorrow morning. There's a corresponding jazzns account, you can probably log in now, I'll send you a little info via email. The account has ftp, telnet and ssh access, your choice. It also has CGI capability, of course, and MySQL access.--Percy