On Tue, Mar 04, 2008 at 10:54:10AM +0100, Raphael Hertzog wrote:
> On Sun, 27 Jan 2008, Aurelien Jarno wrote:
> > That's a good start, but there is still some work to be done:
> > - The minimum dependencies on sparc should be 2.6-1 due to the SPARCV8
> > to SPARCV8PLUS ABI transition
> 
> I suggest to modify the headers in libc6.symbols.sparc to include
> the dependency written as:
> <soname> libc6 (>= 2.6), libc6 #MINVER#
> 
> This will be correctly simplified by dpkg-gencontrol at the end.

Looks correct.

> > - The patch uses symbols versioning to determine the minimal Debian
> > version, but some symbols have been introduced later in Debian, due to
> > some build options that weren't enabled or availabled on some
> > architectures. TLS support is such an example.
> 
> Okay, I have no idea when those got introduced, hopefully we can find
> that in the Debian changelog.

alpha           2.3.5
amd64           always
armel           always
arm             2.4
hppa            2.5
i386            2.3.6-6
ia64            2.3.5
m68k            never
mips            2.4
mipsel          2.4
powerpc         2.3.5
s390            2.3.5
sparc           2.3.5 (libc6) 2.4 (libc6-sparc64)
hurd-i386       2.6-3

> > - As the symbols versioning in dpkg-shlibdeps only takes care of the
> > symbols and not at the include headers, one has to search for #define or
> > enum values that have been added, and bump the minimum version of the
> > corresponding symbols if this value is uses internally in the glibc.
> 
> No idea for this one. Is that documented in the upstream ChangeLog?

Maybe, but it is really huge, so I doubt it can be used for that. The
easiest way is to make a diff between the .h files from etch and sid and
see if the new enum/defines are passed directly to the kernel (which is
fine in that case) or used internally by the libc.


Also note that the dynamic linker name is wrong on some architectures,
like on amd64. This cause a build failure. It would be nice if your
patch can be tried on more than one architecture.

I am also wondering if it worth adding symbol files for hurd, it adds a
a magnitude of complexity for no real gain.

-- 
  .''`.  Aurelien Jarno             | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   [EMAIL PROTECTED]         | [EMAIL PROTECTED]
   `-    people.debian.org/~aurel32 | www.aurel32.net



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to