Re: gcc 3.4: old xine-ui packages are lost?
On 05-Jan-08 17:06, Harald Dunkel wrote:
> AFAIR the plan was to keep gcc 3.4.x at least till Sarge is
> released. After release date the changes done for gcc-3.4
> were supposed to be integrated into the Unstable tree.
The amd64 porting team decided to stay with gcc-3.3 and to
integrate the pure64 branch which uses gcc-3.3 into unstable
after sarge is released. However, Debian unstable will likely switch
to gcc-3.4 or gcc-4.0 some time after sarge is released. So the end
result will be an unstable archive with gcc-3.4 or gcc-4.0 after a while.
> | Since a few weeks, every new package version has been built with gcc-4.0.
> | There were a few problems at the beginning and for some packages I had
> | to switch back to gcc-3.4, but generally it worked quite well. In my
> | own archive I have successfully built 95% of all source packages with
> | gcc-4.0.
> |
>
> If you have obvious compilation failures for 5% of all packages
> when using gcc 4.0, how many packages do you expect to have
> additional internal problems that haven't been detected yet?
Many of the remaining 5% also fail to compile with gcc-3.4 and/or
gcc-3.3 and even on i386 or other official architectures.
At the beginning I found about 600 packages which failed to
build with gcc-4.0 but compiled fine with earlier versions.
Eventually I found patches for most of those packages which made them
compile again. Many of these patches have already been applied by the
package maintainers and uploaded to the unstable archive.
Most of the gcc-4.0 compilation failures were due to stricter warnings
and checks for standard conformity of the code. Gcc-4.0 also found quite
a lot of dubious pointer to integer conversions which may cause severe
problems on amd64 regardless which compiler version is used to build
the package.
Of course there will be some problems with gcc-4.0. Up to now I found
five packages which were somehow miscompiled with gcc-4.0. The most
notable of these is 'openssl' for which I still do not have a fix.
This is why the current 'openssl' version in the archive has
still been compiled with gcc-3.4.
Regards
Andreas Jochens
Reply to: