From: vaeth@mathematik.uni-wuerzburg.de   
      
   Arthur Hagen  wrote:   
   > Martin Vaeth  wrote:   
   >>   
   >> But I doubt that you get all package compiled with gcc-4.3 which is   
   >> meanwhile almost the case with gentoo...   
   >   
   > But since gentoo is source based, just /why/ would one want to compile   
   > everything with gcc-4.3?  The erroneous belief that higher version numbers   
   > are always better?   
      
   Sooner or later, everybody will switch to gcc-4.3 anyway,   
   and it is better to not use mixed libraries on one system.   
   So better get rid of gcc-4.2.1 ASAP.   
   Personally, I wanted to test my projects with gcc-4.3 anyway and,   
   moreover, I wanted to recompile anyway, in order to push   
   -fstack-protector and -D_FORTIFY_SOURCE=2 in some apps where it was   
   yet missing and to add -Wl,-z,relro (about which I learnt only recently;   
   of course, I use throughout -Wl,-z,now anyway where possible).   
      
   > It also means that most hand-crafted optimisations are gone.   
      
   On most modern CPUs, it is really very hard to say what is a good   
   optimisation: What might be good for one CPU might be close to the   
   worst case for a slightly different model; one false branch prediction   
   can be worse than any cumbersome considerations.   
   E.g. 90% of the information which I heard about the effect of   
   various -f... flags were plain wrong with my CPU, although I have no   
   doubt they were correct for others.   
      
   > Speedwise, if you look at actual performance benchmarks instead of   
   > listening to anecdotal evidence ("my system seems faster")   
      
   For the above reason, I don't trust any benchmarks that I haven't faked.   
      
   > It was not made to produce faster or smaller binaries, and in some   
   > cases it sure doesn't.   
      
   In theory that's true, because most optimisations are a two-edged sword.   
   However, most users speak about a slight general speed increase of some   
   percent. I experienced nothing which is slower, and some things really   
   feel faster, but I made no benchmarks and did not use all new compiled   
   applications yet. So probably the about 5% general improvement which I   
   read at several places is correct (as a rule of thumb).   
   Of course, just for speed reasons this would not be worth all the   
   trouble.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)