On Mon, Mar 19, 2007 at 01:47:13PM +0100, Enrico Forestieri wrote: > The swap is already used when needed, and indeed buffer.C takes forever > to be compiled.
The size of the compiler process does not (primarily) depend on the input size (i.e. the size of the compilation unit) and even less on the line count of the source file. It depends mainly on the complexity of the code, most notably when optimizing. > So, using the swap is defeating the original purpose > of faster compilations for me. Indeed, these proposal could even increase > the required time. Right now, compiling from scratch takes almost 2 hours > on cygwin and I remember that I was able to compile 1.3.x in about 45 > minutes. Compiling with scons is a bit faster but not that much as it > takes about 1 hour and 40 minutes. I think that the difference is to be > attributed to autotools and the fact that a fork() is quite expensive > as it has to be emulated with native Windows calls. Anyway, this is not > a cygwin problem, as the required times for compilation are not much > different on Solaris 10 with a Sun-Blade-1000 with 512 Mb of memory. > Indeed, they are quite the same as with cygwin/scons (the Sun machine > is now 5 years old and features an ultrasparc III at 750 MHz). > As a side note, compilation on Linux with a similar powerful processor > as the Sun machine is much faster. This means that compilation speed > is depending on the architecture and a solution which is good for a given > system may not be good for another one. Sure. And especially Windows is notoriously slow on small files. Go check the SVN devel list for discussion why 'svn co' is slow on Windows. Andre'
