---------- Forwarded message ----------
From: Robert DeLisle <[email protected]>
Date: Tue, Nov 16, 2010 at 3:03 PM
Subject: Re: Re: [Rdkit-discuss] Building RDKit on CentOS 5
To: Greg Landrum <[email protected]>, Robert DeLisle <
[email protected]>
'm sorry for the slow response. Busy day.
For this install, I started on a CentOS 5.5 system that was up to date with
all package upgrades. Following is what I've done so far:
installed blas, blas-devel, lapack, lapack-devel through yum
I had problems in the past with the standard GCC package on CentOS which is
version 4.1.2, so I rebuilt the GCC 4.4.5 package and included mpfr 2.4.1
Installed cmake 2.8.2
Installed flex 2.5.35
CentOS's Python installation is v2.4.1, so I built and installed 2.7. Due
to errors found later in the process, I built this with the -fPIC switch and
also enabled Unicode UCS4 support
./configure CFLAGS=-fPIC --enable-unicode=ucs4
Built and installed NumPy 1.5.0
Boost on CentOS 5.5 is v1.33, so I built and installed boost 1.44 with the
following commands:
./bootstrap.sh --with-libraries=python,regex
./bjam address-model=64 install
Finally, with RDKit I have $LD_LIBRARY_PATH set with /usr/local/lib first to
avoid conflicts with the system packages. GCC and Python are both in
/usr/local and these are the instances referred to by my user and root. For
RDKit, the following commands were done:
cmake -DBoost_USE_STATIC_LIBS=OFF -DBOOST_ROOT=/usr/local ..
make
make install
I have also installed FreeType2 and PIL - both seem fine with Python 2.7. I
attempted aggdraw, but the self-test seem to always give me a Segmentation
Fault. I found that I can build aggdraw using the code as-is as long as I
include CFLAGS="-fpermissive", or there is a one line code change that makes
the compiler happy on 64-bit. Either way I still get the seg fault upon
testing.
Regarding RDKit, the first group of errors I received consisted of that
requiring Python be built with -fPIC and what seems to be the typical
USE_STATIC_LIBS error.
Initially, an -fPIC error would occur around 87% which was not cured by the
Python rebuild or any other modification. I found that by switching to the
SVN code, the problem was solved. Upon inspecting the errors logs, it
appeared that the build process was always referring to the system Boost
install and not my new install despite having set -DBOOST_ROOT correctly.
Currently, the build goes to completion but upon issuing 'from rdkit import
Chem' wihtin Python 2.7, I get an error related to Py_InitModule4 not being
defined. From a little Google searching for Py_InitModule4 the only thing
I've seen thus far is a conflict in various packages on code built on 32-bit
or 64-bit systems. It seems that this name has been renamed to
Py_InitModule4_64 on 64-bit systems but that change may not be reflected in
all code necessary. It seemed a widespread problem and not specific to any
one application or library, which makes me think it is something in a Python
include file.
I appreciate any help that anyone can provide. Please let me know if I need
to clarify or add any details.
-Kirk
On Mon, Nov 15, 2010 at 10:11 PM, <[email protected]> wrote:
> No, I made sure to include the address-model=64 switch to bjam.
>
> Tomorrow when I get in I'll update the thread with all the steps I've
> followed.
>
> -Kirk
>
>
>
>
> On Nov 15, 2010 9:52pm, Greg Landrum <[email protected]> wrote:
> > Kirk,
> >
> >
> >
> > On Tue, Nov 16, 2010 at 12:38 AM, Robert DeLisle [email protected]>
> wrote:
> >
> > > Yes, that is also true.
> >
> > >
> >
> > > The error in my most recent messages stems from the default build of
> Python
> >
> > > supporst Unicode UCS2, but apparently boost expects UCS4. A rebuild of
> >
> > > Python with UCS4 enabled fixed that problem.
> >
> > >
> >
> > > Now I get a similar error related to Py_InitModule4 not being defined.
> From
> >
> > > what I can find, this is a 32-bit - 64-bit problem in which this was
> defined
> >
> > > as Py_InitModule4_64 in the 64-bit Python libraries but that change may
> not
> >
> > > have cascaded to all necessary parts of the build process. Most of the
> >
> > > changes involve some substantial changes to the accessing code, but I'm
> >
> > > still looking for a better option.
> >
> >
> >
> > Could it be that the boost libraries you are using were not built in
> >
> > 64bit mode? I've managed to force a 64bit build in the past with the
> >
> > following command line:
> >
> > ./bjam address-model=64 cflags=-fPIC cxxflags=-fPIC install
> >
> >
> >
> > Best Regards,
> >
> > -greg
> >
>
------------------------------------------------------------------------------
Beautiful is writing same markup. Internet Explorer 9 supports
standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3.
Spend less time writing and rewriting code and more time creating great
experiences on the web. Be a part of the beta today
http://p.sf.net/sfu/msIE9-sfdev2dev
_______________________________________________
Rdkit-discuss mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
