Please update the documentation at http://www.gentoo.org/doc/en/distcc.xml , currently it only says "Don't use -march=native", it doesn't mention that distcc actually gets DISABLED automatically, according to bug 223159. Reproducible: Always Steps to Reproduce:
I'll leave it as is, for a few reasons: 1. Because it does encourage smarter use of -march by stating it plainly. 2. That patch for distcc may or may not be around forever. This is playing it on the safe side. 3. There's a rather disturbing trend among users and even a few developers to use package-specific CFLAGS. While they're still a very bad hack, leading to an inconsistent world, implementing them has become rather easy with newer versions of Portage, thanks to all the new things you can put into /etc/portage/env and whatnot these days.
I disagree with all your points: 1. Actually it doesn't. It explains that it can lead to "code optimized for different processors", which implies that it is ok if all processors involved are the same. This is exactly my case. I wasted a LOT of time trying to find out why emerge wasn't using distcc, and just luckily stumbled upon that bug. If the doc mentioned that, I would know immediately. 2. The patch is in force now. Playing it safe would be to update the documentation to reflect the current state, not to wait for a potential change that may or may not come at some undefined time in the future. 3. I don't see how that's relevant. If some people want to mess up their systems on purpose, it's their responsibility to deal with the consequent problems. If you're suggesting that as a workaround, it's a terrible idea.
Is -march=native always equivalent to -march=(whatever family it is run on), or does it take advantage of all features it finds using cpuid? For example, I have core2 on both home and work systems, but the work system is a newer 'generation' and has some sse4 instructions which my home system does not have.
(In reply to comment #3) > Is -march=native always equivalent to -march=(whatever family it is run on), or > does it take advantage of all features it finds using cpuid? For example, I > have core2 on both home and work systems, but the work system is a newer > 'generation' and has some sse4 instructions which my home system does not have. Your question is already answered on the forums and on Google. Please check there, rather than on this already closed bug.
