> I'm surely not the only one that has found scons builds to be slow, have > they always been slow. M5 is a reasonably sized body of code, but it is by > no means a behemoth, and from experience I have seen faster builds on much > bigger projects. A colleague also suffering the same problem of waiting for > builds to complete pointed me at this link. > > http://blog.electric-cloud.com/2010/07/21/a-second-look-at-scons-performance/ > > Apart from legacy, and the fact that someone would need to spend time on the > build system, is there a particular reason why scons is being used? what was > the original decision making process?
The primary reason that we use SCons is because it is python and we do a ton of code generation in python. Part of the slowness is all of that code. Comparing SCons to make is also not very interesting since gmake has serious problems with parallel builds. The one option that I see is trying to port to Waf, but that would likely be a monumental effort. We could probably optimize a lot of our SCons usage if we tried. I know of several things that we could do to improve performance, but it would take a fair amount of work. Nate _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
