On Sun, 2009-05-03 at 09:39 +0200, Ralf Corsepius wrote: > Lubomir Rintel wrote: > > On Sat, 2009-05-02 at 13:52 +0300, Ville Skyttä wrote: > > > One subtle advantage of dynamic linking would be saving space by making > > the executable image as well as text of the running process smaller, but > > I guess in your case (trading space for performance) is a quite good > > deal. > > The key word in your sentence is "subtile" ... > > How much performance gains does static linkage actually provide in this > particular case? > > In the overwhelming majority of general cases, such kind of > performance gains are close to immeasurable, distorted/overlayed by > other effects (ld.so caches, ld.preload, disk caches, etc.) and hardly > ever user noticeable. I tend to agree with you, hardly being able to imagine the PLT jumps add any real overhead unless the every third instruction is a jump to library function. I'd be interested to see the results of benchmarks upstream has done, unfortunately they don't provide those (which don't contribute to the credibility of the claim anyhow). Point is that there's nothing really seriously wrong against using the static linkage internally, especially when upstream decided so for a reason. Asking for the explanation for the reason once unclear might be a good idea though. -- "Excuse all the blood" -- Dead -- Fedora-packaging mailing list Fedora-packaging@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-packaging
