On Jun 22, 2010, at 2:45 PM, Eelco Dolstra wrote: > Hi, > > On 06/22/2010 08:18 PM, Jeff Johnson wrote: > >> The "portability" issue is that there isn't any clear API to >> extract the number of CPUS programatically. > > I don't see the portability issue, since this patch doesn't actually try to > determine the number of CPUs. Instead it's up to the user to set > `build-cores'. > But even if we would interpret `build-cores = 0' as "guess the number of > CPUs", > it doesn't have to be perfect. If it doesn't work on some platform, you can > always set the value manually. >
I don't disagree. My suggestion was for "clearly documented source", not for "portability". If the source is "user sets build-cores" that's find by me. But if "you can always set the value manually", why not just design that in directly without the fiction(s) of #cpus as a basis for the heuristic. There's nothing at all wrong with leaving it to the user to configure the build-cores. There _IS_ however something wrong with benchmarking heuristics based on a fictional configurable and arguing for/against some change because of the benchmark results imho. 73 de Jeff _______________________________________________ nix-dev mailing list [email protected] https://mail.cs.uu.nl/mailman/listinfo/nix-dev
