[Insight-users] Possible misfeature in debian insighttoolkit package

[Steve M. Robbins]

Hi Ben,

I saw your follow-up message that notes we use the cswig python
wrapping rather than WrapITK.  If all the issues are worked out
(mainly removing the dependency of WrapITK on the Review directory) I
think Debian will switch to WrapITK with the next ITK release, which
will enable us to set ITK_USE_PYTHON_NUMARRAY=ON again.

See below for more history.


On Thu, Jan 07, 2010 at 07:18:37PM -0500, Benjamin M. Schwartz wrote:
> While debugging a build problem on Gentoo I read through Debian's build
> config [1], for reference.  I noticed the following comment there:
> 
> """
> +  * Python-Numarray is removed from Debian.
> +    - CMakeCache.txt.debian: Remove ITK_USE_PYTHON_NUMARRAY=ON.
> +    - rules: Remove PYTHON_NUMARRAY_INCLUDE_DIR from CMakeCache.txt.
> +    - control.in: Remove build-depend on python-numarray.
> """
> 
> I believe this is a misunderstanding.  I think the ITK_USE_PYTHON_NUMARRAY
> flag triggers an autodetection step encoded in
> 
> InsightToolkit-3.16.0/Wrapping/WrapITK/ExternalProjects/PyBuffer/FindNUMARRAY.cmake
> 
> That script clearly attempts to detect Numeric, Numarray, or Numpy, and
> will use any of them (according to some order of preference).  
> [...]

> If I am correct, the Debian packages should probably continue to enable
> ITK_USE_PYTHON_NUMARRAY, but with a dependency on numpy.

When I removed ITK_USE_PYTHON_NUMARRAY, I did not try building with
python-numpy [1] (mainly out of ignorance).

Mathieu Malaterre then queried the insight-developers list [2], asking

  1. Does anyone use Numpy ? Is this working ?
  2. What is the difference in between (the patched)
     Wrapping/WrapITK/ExternalProjects/PyBuffer/itkPyBuffer.txx and
     Wrapping/CSwig/CommonA/itkPyBuffer.txx ?

to which Gaëtan Lehmann replied

  1. I do - it works.
  2. I never succeeded to use the later, while the former works great :-)
     The new one is a rework of what was began with the old one, but never  
     polished enough to be production ready.

To my best recollection, we never got a clear answer on why the old
one was "never polished enough to be production ready".  Mathieu did
verify that the old cswig code will compile [3].  However, being busy
with other things, uploading ITK with ITK_USE_PYTHON_NUMARRAY
re-enabled was never a high priority.

Mathieu went on to investigate switching to WrapITK [4] and quickly found
that it depends on code in the Review directory [5], which Debian does not
ship due to potentially unclear license terms.

At this point, I'm waiting to have WrapITK made independent of Review.

Gaëtan, Luis: is this even on the agenda for the next ITK?

Thanks,
-Steve


[1] http://lists.debian.org/debian-med/2009/09/msg00115.html
[2] http://www.itk.org/mailman/private/insight-developers/2009-September/013417.html
[3] http://lists.debian.org/debian-med/2009/09/msg00122.html
[4] http://www.itk.org/pipermail/insight-users/2009-October/033301.html
[5] http://lists.debian.org/debian-med/2009/11/msg00059.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 190 bytes
Desc: Digital signature
URL: <http://www.itk.org/pipermail/insight-users/attachments/20100109/540b5888/attachment.pgp>