BUG: python/py25-scipy fails to build when py25-numpy built using gcc42/gcc43 variants

[skymoo (Adam Mercer)]

If you try to build py25-scipy using a py25-numpy built with either the gcc42 or gcc43 variant the build fails - build log attached as install_1.log.

The build fails because it can't find the binary of the fortran compiler, if you add the fortran compiler variants like in py25-numpy, patch attached, the build this time it manages to find the fortran compilers during the build phase, but fails in the install phase as it is unable to find the fortran compiler - build log attached as install_2.log.

It seems that the install phase is ignoring the compiler choice that was set using config_fc.

[skymoo (Adam Mercer)]

build log

[skymoo (Adam Mercer)]

patch to add fortran variants

[skymoo (Adam Mercer)]

build log of patched Portfile

[skymoo (Adam Mercer)]

Just tried building/installing manually with

$ python setup.py config_fc --fcompiler gnu95 \
                  --f77exec /opt/local/bin/gfortran-mp-4.2 \
                  --f90exec /opt/local/bin/gfortran-mp-4.2 build
$ python setup.py install --prefix=$HOME/scipy

and it fails with the same error. But if I install with

$ python setup.py config_fc --fcompiler gnu95 \
                  --f77exec /opt/local/bin/gfortran-mp-4.2 \
                  --f90exec /opt/local/bin/gfortran-mp-4.2 install --prefix=$HOME/scipy

then the install succeeds, therefore I added the appropriate config_fc... to destroot.cmd in each of the different variants and installing from the port succeeds.

I've attached a new patch that does this for the py25-scipy Portfile, a similar change will therefore be needed for the py-scipy port.

[skymoo (Adam Mercer)]

specify fortran compilers for build and destroot commands

[skymoo (Adam Mercer)]

specify fortran compilers for build and destroot commands

[erickt@…]

Commited. Let me know if it doesn't work.

[skymoo (Adam Mercer)]

Builds fine for me with the gcc42 variant.

It seems to me that the install phase should inherit the options specified during the build phase, as such I've emailed the scipy users mailing list to see if this is the expected behaviour. I'll post what I find out...

[skymoo (Adam Mercer)]

Just head back from the scipy developers, this is the expected behaviour and the approach used to fix the install is correct.

[(none)]

Milestone Port Bugs deleted