> More importantly, is there a penalty for reading and writing a one > file on the filesystem 2,470,018 times during one large parser run? > > Drop files bad, signals good.
Um, this *is* a signal. Or more precisely, this is one of the few portable IPC constructs. At heart, I'm a CORBA guy -- I'd much prefer to use that (see http://www.parc.com/ILU/). But files just work. And they provide the stable storage for the "estimate" part of things. Perhaps a canny user would use a file in a RAM-based filesystem. Bill
