Re: problems installing openmpi_ext

[c.]

On 8 Sep 2013, at 09:28, c. <address@hidden> wrote:

> 
> On 8 Sep 2013, at 00:55, Alexander Hansen <address@hidden> wrote:
> 
>>>> Fink's build process for the general package is pretty much identical to
>>>> just installing it manually, and it appears to produce the same failure.
>>>> :-(
>>>> 
>>>> My assumption is that some of the options that I'm using to build Octave
>>>> aren't the best.  Here's what Fink uses (at least for the variant that
>>>> uses untuned LAPACK):
>>>> 
>>>> octave:2> octave_config_info.config_opts
>>>> ans = '--prefix=/sw' '--with-lapack=-L/sw/lib/lapack/3.4.2 -lreflapack'
>>>> '--with-blas=-L/sw/lib/lapack/3.4.2 -lrefblas' '--with-framework-carbon'
>>>> '--with-x' '--x-libraries=/usr/X11/lib' '--x-includes=/usr/X11/include'
>>>> '--with-magick=GraphicsMagick'
>>>> '--with-qhull-includedir=/sw/include/libqhull'
>>>> '--host=x86_64-apple-darwin' '--build=x86_64-apple-darwin'
>>>> '--infodir=/sw/share/info' '--mandir=/sw/share/man'
>>>> '--libexecdir=/sw/lib' '--enable-shared' '--enable-dl'
>>>> '--disable-static' '--disable-docs' '--enable-dependency-tracking'
>>>> 'build_alias=x86_64-apple-darwin' 'host_alias=x86_64-apple-darwin'
>>>> 'CC=flag-sort -r gcc' 'CFLAGS=-O3 -MD' 'LDFLAGS=-Wl,-dead_strip_dylibs
>>>> -L/sw/lib' 'CPPFLAGS=-I/sw/include' 'CXX=flag-sort -r g++' 'CXXFLAGS=-O3
>>>> -MD -I/usr/X11/include' 'F77=/sw/bin/gfortran-fsf-4.8' 'FFLAGS=-O3'
>>> 
>>> I should have added that "flag-sort" is a script that sorts the flags
>>> fed to the compiler, and that "g++" and "gcc" are Apple's clang++ and
>>> clang in this case.
>>> 
>> 
>> Following up:
>> 
>> I tried rebuilding Octave (followed by rebuilding general) changing -O3 to 
>> -O0 in the CFLAGS and CXXFLAGS, as well as switching the C and C++ compilers 
>> to our FSF gcc-4.8, but I'm still getting the malloc() error that Francis 
>> reported
>> 
>> octave:1> parcellfun_example
>> Elapsed time is 9.4712 seconds.
>> parcellfun: child process 46393 created
>> parcellfun: child process 46394 created
>> parcellfun: 0/200 jobs doneoctave(46386,0x7fff7479d180) malloc: *** error 
>> for object 0x10c7f07e0: pointer being freed was not allocated
>> *** set a breakpoint in malloc_error_break to debug
>> panic: Abort trap: 6 -- stopping myself...
>> 
> 
> 
> Hi,
> 
> Sorry it took me long to respond but it's been a while since I 
> last built the 3.6 version and it took me a few attempts to get 
> it to work.
> 
> Here's my configure options:
> 
> ./configure                                                \
>    CC=/opt/local/bin/gcc-mp-4.7                           \
>    CFLAGS="-pipe -O2 -m64"                                \
>    CPPFLAGS="-D_THREAD_SAFE -I/opt/local/include"         \
>    LDFLAGS="-L/opt/local/lib -m64 /opt/local/lib/libsuitesparseconfig.a" \
>    CXX=/opt/local/bin/g++-mp-4.7                          \
>    CXXFLAGS="-pipe -O2 -m64"                              \
>    F77=/opt/local/bin/gfortran-mp-4.7                     \
>    FFLAGS="-pipe -O2 -m64"                                \
>    LLVM_CONFIG=/opt/local/bin/llvm-config-mp-3.1          \
>    --prefix=/opt/octave/3.6                               \
>    --without-framework-carbon                             \
>    --with-arpack                                          \
>    --with-lapack="/opt/local/lib/liblapack.a /opt/local/lib/libatlas.a 
> -lgfortran" \
>    --with-blas="/opt/local/lib/libcblas.a /opt/local/lib/libf77blas.a 
> /opt/local/lib/libatlas.a -lgfortran"   
> 
> Notice in particular I had to change blas/lapack as configure seemed to have 
> problems detecting atlas
> and that I had to add "libsuitesparseconfig.a" explicitely.
> 
> Finally I had to disable framework Carbon to avoid use of deprecated function 
> CGDisplayBitsPerPixel.
> 
> All my  dependencies are installed via MacPorts.
> 
> With these options I get:
> 
>>> parcellfun_example
> Elapsed time is 10.2874 seconds.
> parcellfun: child process 87208 created
> parcellfun: child process 87209 created
> parcellfun: 200/200 jobs done
> parcellfun: waiting for child process 87208 to close
> parcellfun: waiting for child process 87209 to close
> Elapsed time is 7.35937 seconds.
>>> 
> 
> I attach parcellfun_example.m jus to make sure we are using the same example.
> 
> c.

As our configure options are not much different I suspect the error might not 
be in Octave itself
but rather in some of the linked libraries.

If you run Octave in a debugger (run-octave -g) and look at the backtrace of 
the crash, that may give some hint ...
c.