Date: Sat, 22 Aug 1998 15:45:53 -0700 (PDT) From: Alex <garbanzo@hooked.net> To: Alexander Sanda <entropy@compufit.at> Cc: wwoods@cybcon.com, freebsd-current@FreeBSD.ORG Subject: Re: gcc 2.8 Message-ID: <Pine.BSF.4.00.9808221538220.254-100000@zippy.dyn.ml.org> In-Reply-To: <19980822135031.A358@compufit.at>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 22 Aug 1998, Alexander Sanda wrote: [...] > However, I have installed gcc-2.8.1 from the packages collection, and I > have the vague feeling, that this compiler has some problems. I > compiled one of my kde apps, using -O2 and -mpentiumpro and the app > started to segfault occasionally. Since I recompiled with gcc-2.7.2.1, > it never segfaulted again... I think this is an a.out problem. I'm using egcs from their anon cvs repository, and after building ELF libs and X (and all other X related goodies), egcs works wonderfully. I don't have any major cli C++ programs, so I didn't bother with building an ELF world. gdb seems to be the only stumbling block for me. [...] > Once a while ago, I did some experiments with compiling the Linux kernel > using different compilers (stock gcc-2.7, egcs, pgcc) and benchmarking > them with lmbench or byte. The results: They all ranged within > measurement tolerance, imho. Even if you run lmbench twice on the same > system, the results will slightly differ. I think the P5 (not clones) benefits more from compiler optimizations than do other chips (esp. the P6). > For kernel code, any compiler-induced problem can be lethal and will > probably result in an unstable system. Bad enough, those bugs would be > extremely hard to track. Fair enough, but I've built kernels with gcc 2.8 and earlier egcs snapshots, and pgcc and it's worked well for me. > As far as we all know, gcc-2.7.2.1 produces ok code. Known problems (if > any, I'am not an expert here) can be workarounded by the really skilled > kernel hackers. Moving to another compiler will perhaps introduce new > problems - as long as the performance gain is practically non-existant, > this is not very desireable. gcc 2.7 produces OK code, however, it's C++ support is really quite lacking. If you bend over backwards far enough to support it (which I have done) you really do miss out on some of the more fun features. Even Windows has the Unix world beat here (IBM and Sun's C++ compilers are far more broken than any post 2.6 gcc). I think most of the problems you'll encounter with newer compilers will be related to using a.out, and shared a.out libs, I think that since x86 Linux is such a popular platform, most code generation problems that would inflict pain on the FreeBSD will have been quashed long ago. - alex A person who has both feet planted firmly in the air can be safely called a liberal. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.00.9808221538220.254-100000>