Re: [PATCH 1/3] distutils-common-base.bbclass: Prepend STAGING_LIBDIR to linker library search path.

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Mon, 2011-04-18 at 07:54 +0200, Martin Jansa wrote:
> On Thu, Mar 31, 2011 at 03:26:39PM +0100, Richard Purdie wrote:
> > On Thu, 2011-03-31 at 12:40 +0200, Martin Jansa wrote:
> > > * This was done generically in bitbake.conf this should be done on
> > >   demand and distutils based recipes seem to need it. So we define it
> > >   here.
> > > * imported from OE 629ae87eadeec1068c3778e14488b549029adffb
> > > 
> > > Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
> > > ---
> > >  meta/classes/distutils-common-base.bbclass |    2 ++
> > >  1 files changed, 2 insertions(+), 0 deletions(-)
> > > 
> > > diff --git a/meta/classes/distutils-common-base.bbclass b/meta/classes/distutils-common-base.bbclass
> > > index f66a5cd..b6caf24 100644
> > > --- a/meta/classes/distutils-common-base.bbclass
> > > +++ b/meta/classes/distutils-common-base.bbclass
> > > @@ -4,6 +4,8 @@ EXTRA_OEMAKE = ""
> > >  
> > >  export STAGING_INCDIR
> > >  export STAGING_LIBDIR
> > > +# libtool 2.4 does not export with LDFLAGS but distutils need it
> > > +LDFLAGS_prepend = "-L${STAGING_LIBDIR} "
> > 
> > This should not be needed, the linker itself should automatically look
> > in that path at link time. What problem are you seeing?
> 
> sorry for delay, my previous reply was refused by mailman at Mon, 04 Apr 2011
> with "The message's content type was not explicitly allowed", maybe PGP
> sign is not allowed? ;/.
> 
> 
> when building any python extra module ie python-numeric I get error like
> this
> 
> | arm-oe-linux-gnueabi-gcc -march=armv7-a -mtune=cortex-a8 -mfpu=neon -mfloat-abi=softfp -mthumb-interwork -mno-thumb --sysroot=/OE/shr-core/tmp/sysroots/nokia900 -DNDEBUG -g -O3 -Wall -Wstrict-prototypes -fexpensive-optimizations -fomit-frame-pointer -frename-registers -Os -g -feliminate-unused-debug-types -fPIC -IInclude -IPackages/FFT/Include -IPackages/RNG/Include -I/OE/shr-core/tmp/sysroots/arm-oe-linux-gnueabi/usr/include/python2.6 -I/OE/shr-core/tmp/sysroots/nokia900/usr/include/python2.6 -c Src/ufuncobject.c -o build/temp.linux-x86_64-2.6/Src/ufuncobject.o
> | arm-oe-linux-gnueabi-gcc -march=armv7-a -mtune=cortex-a8 -mfpu=neon -mfloat-abi=softfp -mthumb-interwork -mno-thumb -D__SOFTFP__ --sysroot=/OE/shr-core/tmp/sysroots/nokia900 -shared -Wl,-O1 -Wl,--hash-style=gnu -fexpensive-optimizations -fomit-frame-pointer -frename-registers -Os -g -feliminate-unused-debug-types build/temp.linux-x86_64-2.6/Src/_numpymodule.o build/temp.linux-x86_64-2.6/Src/arrayobject.o build/temp.linux-x86_64-2.6/Src/ufuncobject.o -L/usr/lib -lpython2.6 -o build/lib.linux-x86_64-2.6/_numpy.so
> | /OE/shr-core/tmp/sysroots/x86_64-linux/usr/libexec/armv7a-oe-linux-gnueabi/gcc/arm-oe-linux-gnueabi/4.5.3/ldERROR: Function 'do_compile' failed (see /OE/shr-core/tmp/work/armv7a-oe-linux-gnueabi/python-numeric-24.2-ml1/temp/log.do_compile.16227 for further information)
> | : warning: library search path "/usr/lib" is unsafe for cross-compilation
> | /OE/shr-core/tmp/sysroots/x86_64-linux/usr/libexec/armv7a-oe-linux-gnueabi/gcc/arm-oe-linux-gnueabi/4.5.3/ld: skipping incompatible /usr/lib/libc.so when searching for -lc
> | /usr/lib/libc.a: could not read symbols: File format not recognized
> | collect2: ld returned 1 exit status
> | error: command 'arm-oe-linux-gnueabi-gcc' failed with exit status 1
> | FATAL: python setup.py build_ext execution failed.
> NOTE: package python-numeric-24.2-ml1: task do_compile: Failed
> ERROR: Task 4075 (/OE/shr-core/meta-shr/recipes-devtools/python/python-numeric_24.2.bb, do_compile) failed with exit code '1'
> 
> in oe we had the same problem after switching to libtool 
> with sysroot support IIRC and this was fix for it 
> (and it helps with oe-core too)

This isn't a libtool problem, we've patched that problem out in oe-core
(and I think in OE-dev too now). In this case something is adding
"-L/usr/lib" to the compiler arguments and that is the reason things are
failing. I'd suggesting finding out where that is coming from and fixing
it at source. Yes, it so happens adding other search paths first masks
the problem but it is just masking it, its not a proper fix.

Cheers,

Richard