rdkit-devel

[Rdkit-devel] "minimal" version of the RDKit

[Greg L. <gre...@gm...>]

Dear all,

In connection with the MolCore project (still in the larval stage:
http://sourceforge.net/apps/trac/molcore/) I've put together a branch
of the RDKit that includes only the most basic molecular functionality
(reading and writing from smiles and SDF, substructure searching,
reactions): http://rdkit.svn.sourceforge.net/viewvc/rdkit/branches/minimal_1Aug2009/

One potentially interesting thing about this branch to the rest of you
is that it is built using cmake instead of bjam. Build instructions
are in the INSTALL file.

Getting cmake working turned out to be relatively straightforward,
particularly now that boost supports it, and it's much easier to get
started with than bjam. I will at some point (in October) port the
trunk of the RDKit over to cmake as well (if anyone else wants to try
doing this: I think it should be pretty easy based on what's in the
minimal branch).

-greg

Re: [Rdkit-devel] "minimal" version of the RDKit

[Riccardo V. <ric...@gm...>]

Hi,

On Wed, Sep 2, 2009 at 6:39 AM, Greg Landrum<gre...@gm...> wrote:
> [...]
>
> One potentially interesting thing about this branch to the rest of you
> is that it is built using cmake instead of bjam. Build instructions
> are in the INSTALL file.
>
> [...]. I will at some point (in October) port the
> trunk of the RDKit over to cmake as well (if anyone else wants to try
> doing this: I think it should be pretty easy based on what's in the
> minimal branch).

sounds very good to me, I think that under linux cmake will allow a
better integration of the build process with the rest of the
distribution. Here's some feedback:

I've just finished compiling the minimal branch against the boost
libraries already available on my linux box (Fedora 11) and almost
everything worked out-of-the box and required a minimal configuration
of the build environment (just setting $RDBASE for the build,
$LD_LIBRARY_PATH and $PYTHONPATH to run the tests).

The only issue I faced has been related to the inclusion of a header
file belonging to numpy (the numpy include directory here is
/usr/lib/python2.6/site-packages/numpy/core/include/).

I can't say if the problem is just mine, but I worked around it with a
FindNumpy.cmake file belonging to the avogadro git repository that was
easily returned by a web search and then some trial-and-error changes
to it and the interested CMakeLists.txt files (until now my experience
with cmake was just from the end-user point of view, so the minor
fixes I made are not probably very "clean"..).

Best regards,
Riccardo

Re: [Rdkit-devel] "minimal" version of the RDKit

[Greg L. <gre...@gm...>]

Dear Riccardo,

Thanks for the feedback.

On Thu, Sep 3, 2009 at 1:40 AM, Riccardo
Vianello<ric...@gm...> wrote:
> On Wed, Sep 2, 2009 at 6:39 AM, Greg Landrum<gre...@gm...> wrote:
>> [...]
>>
>> One potentially interesting thing about this branch to the rest of you
>> is that it is built using cmake instead of bjam. Build instructions
>> are in the INSTALL file.
>>
>> [...]. I will at some point (in October) port the
>> trunk of the RDKit over to cmake as well (if anyone else wants to try
>> doing this: I think it should be pretty easy based on what's in the
>> minimal branch).
>
> sounds very good to me, I think that under linux cmake will allow a
> better integration of the build process with the rest of the
> distribution.

Yeah, I think it's really going to help; and not just on linux.
Windows developers are going to have a much easier time of it when the
RDKit is more easily used with visual studio projects.

> Here's some feedback:
> I've just finished compiling the minimal branch against the boost
> libraries already available on my linux box (Fedora 11) and almost
> everything worked out-of-the box and required a minimal configuration
> of the build environment (just setting $RDBASE for the build,
> $LD_LIBRARY_PATH and $PYTHONPATH to run the tests).

Glad to hear it.

> The only issue I faced has been related to the inclusion of a header
> file belonging to numpy (the numpy include directory here is
> /usr/lib/python2.6/site-packages/numpy/core/include/).
>
> I can't say if the problem is just mine, but I worked around it with a
> FindNumpy.cmake file belonging to the avogadro git repository that was
> easily returned by a web search and then some trial-and-error changes
> to it and the interested CMakeLists.txt files (until now my experience
> with cmake was just from the end-user point of view, so the minor
> fixes I made are not probably very "clean"..).

This was a mistake on my part. I had forgotten to check in a version
of FindNumPy.cmake that I added to get things building on the Mac.
That's been fixed in svn now.

-greg