changeset in /hg/icedtea6: 2008-10-27 Matthias Klose <doko at ubu...

[Matthias Klose]

Christian Thalinger schrieb:
> On Mon, 2008-10-27 at 11:05 +0000, doko at ubuntu.com wrote:
>> -+  ifneq (,$(filter $(mach),mips s390 s390x))
>> ++  ifneq (,$(wildcard /usr/bin/dpkg-architecture))
>>  +    mach := $(shell dpkg-architecture -qDEB_BUILD_ARCH_CPU 2>/dev/null || echo $(mach))
>>  +  endif
>>     archExpr = case "$(mach)" in \
>> @@ -20,13 +20,13 @@
>>                   *) \
>>                       echo $(mach) \
>>                       ;; \
> 
> This patch breaks the build on powerpc, as the output of
> dpkg-architecture is:
> 
> $ dpkg-architecture -qDEB_BUILD_ARCH_CPU
> powerpc
> 
> This results in setting ARCH to powerpc and later in the build in:
> 
> -L/localtmp/cacao/jvmtester/work/b6-icedtea-zero/icedtea/build/bootstrap/jdk1.6.0/jre/lib/powerpc/ -ljvm 
> 
> /usr/bin/ld: cannot find -ljvm
> 
> Because the directory where libjvm.so lays is called jre/lib/ppc/.  Why
> not simply adding cases for all architectures we support and not
> depending on such utilities?

Why not simply removing the use of uname -m? Yes, I know that the current
approach is a hack, but some debian buildds are configured in such a way that
they don't set the linux32 personality (no, I don't want to argue why buildd
admins do this). I checked in a fix for the ppc/powerpc problem, but in the long
term this should be better fixed upstream. The upstream build already uses
-m32/-m64 explicitely to allow building in configurations where the default gcc
(without -mxx) is not the compiler you want to use.

  Matthias