[arch-dev-public] New toolchain stuff in testing
Hi all, Andy and I uploaded a new toolchain to testing. This includes a new binutils, a new gcc and a new glibc. On AMD64, the old stable 2.17 binutils doesn't function correctly in all cases with gcc 4.2 and glibc 2.6 (I was waiting for a binutils 2.18 release, but it's not coming within the next month I'm afraid). One big new feature in the new binutils is the support for GNU hashing, which is, according to Redhat, 50% faster in symbol lookups when doing dynamic linking. The GNU hashing is incompatible with the older hashing. There's a way inbetween: supporting both hash styles. The gcc in testing enforces both hash styles in the resulting linked binary. This means the size increases (I've seen libstdc++.so.6 and libc.so.6 getting increased by 10-20KB on a 1-2MB sized binary), but we maintain compatibility with other distributions and with our current "old style" glibc whenever something gets built with this toolchain. this is also the road taken by Debian. As long as binutils 2.17.50.xx isn't a so called stable release, we will keep this duplicate hashing style around, we have to decide what to do after the 2.18 release of binutils. (gcc-fortran will follow later tomorrow btw). Have fun testing this new toolchain, please report any bugs you find in flyspray.
Am Mon, 25 Jun 2007 00:45:11 +0200 schrieb Jan de Groot <jan@jgc.homeip.net>:
Have fun testing this new toolchain, please report any bugs you find in flyspray.
i haven't uploaded new gcc-gcj for a good reason. first i wanted to make sure we can still compile and run OpenOffice. and so i have already found out that the new binutils/glibc from testing have horrible issues handling gnujava. calling OOo wizards eats up tons of memory and almost freezes the system. same when trying to compile OOo. i will look for a fix. so be warned not to use new bintuils/glibc together with any java stuff right now. Andy
On Mon, 2007-06-25 at 12:28 +0200, Andreas Radke wrote:
Am Mon, 25 Jun 2007 00:45:11 +0200 schrieb Jan de Groot <jan@jgc.homeip.net>:
Have fun testing this new toolchain, please report any bugs you find in flyspray.
i haven't uploaded new gcc-gcj for a good reason. first i wanted to make sure we can still compile and run OpenOffice.
and so i have already found out that the new binutils/glibc from testing have horrible issues handling gnujava. calling OOo wizards eats up tons of memory and almost freezes the system. same when trying to compile OOo.
i will look for a fix. so be warned not to use new bintuils/glibc together with any java stuff right now.
Andy
Couldn't reproduce this on i686, Openoffice and its java wizards load a bit faster and I see no big memory issues on my laptop.
filed a bugreport http://sourceware.org/bugzilla/show_bug.cgi?id=4781. ml thread http://sourceware.org/ml/binutils/2007-07/msg00144.html wasn't very helpful so far. this bug has to be solved before we can move new binutils+gnuhashing to current. Andy
On 6/24/07, Jan de Groot <jan@jgc.homeip.net> wrote:
Hi all,
Andy and I uploaded a new toolchain to testing. This includes a new binutils, a new gcc and a new glibc. On AMD64, the old stable 2.17 binutils doesn't function correctly in all cases with gcc 4.2 and glibc 2.6 (I was waiting for a binutils 2.18 release, but it's not coming within the next month I'm afraid).
One big new feature in the new binutils is the support for GNU hashing, which is, according to Redhat, 50% faster in symbol lookups when doing dynamic linking. The GNU hashing is incompatible with the older hashing. There's a way inbetween: supporting both hash styles. The gcc in testing enforces both hash styles in the resulting linked binary. This means the size increases (I've seen libstdc++.so.6 and libc.so.6 getting increased by 10-20KB on a 1-2MB sized binary), but we maintain compatibility with other distributions and with our current "old style" glibc whenever something gets built with this toolchain. this is also the road taken by Debian. As long as binutils 2.17.50.xx isn't a so called stable release, we will keep this duplicate hashing style around, we have to decide what to do after the 2.18 release of binutils.
(gcc-fortran will follow later tomorrow btw).
Have fun testing this new toolchain, please report any bugs you find in flyspray.
No problems here, looks pretty good to me. -Dan
participants (3)
-
Andreas Radke
-
Dan McGee
-
Jan de Groot