Building error:Unable to load Jamfile.

[GoogleCodeExporter]

What steps will reproduce the problem?
1. download
2. unzip
3. edit config
4. run python setup.py build

What is the expected output? What do you see instead?

running build
running build_py
copying pyopencv/config.py -> build/lib.linux-i686-2.6/pyopencv
running build_ext
building 'pyopencv.pyopencvext' extension
bjam release test link=static
notice: no Python configured in user-config.jam
notice: will use default configuration
/usr/share/boost-build/build/project.jam:270: in find-jamfile from module 
project
error: Unable to load Jamfile.
error: Could not find a Jamfile in directory '/usr/include'.
error: Attempted to find it with pattern '[Bb]uild.jam [Jj]amfile.v2 [Jj]amfile 
[Jj]amfile.jam'.
error: Please consult the documentation at 'http://www.boost.org'.
/usr/share/boost-build/build/project.jam:290: in load-jamfile from module 
project
/usr/share/boost-build/build/project.jam:68: in project.load from module project
/usr/share/boost-build/build/project.jam:718: in project.use from module project
/usr/share/boost-build/build/project.jam:94: in load-used-projects from module 
project
/usr/share/boost-build/build/project.jam:79: in load from module project
/usr/share/boost-build/build/project.jam:170: in project.find from module 
project
/usr/share/boost-build/build-system.jam:248: in load from module build-system
/usr/share/boost-build/kernel/modules.jam:261: in import from module modules
/usr/share/boost-build/kernel/bootstrap.jam:132: in boost-build from module
/home/dan/tmp/pyopencv-2.0.wr1.0.0/boost-build.jam:2: in module scope from 
module

error: command 'bjam' failed with exit status 1

What version of the product are you using? On what operating system?

pyopencv-2.0.wr1.0.0 on Ubuntu 9.10 

Please provide any additional information below.

config.py
---------
import os
from glob import glob

# OpenCV 2.0 library, to be linked against using bjam+gcc
opencv_dir = "/usr/local"
opencv_include_dirs = [opencv_dir+"/include/opencv"]
opencv_library_dirs = [opencv_dir+"/lib"]
opencv_libraries = ["highgui", "ml", "cvaux", "cv", "cxcore"]
opencv_runtime_library_dirs = []
opencv_runtime_libraries_to_be_bundled = []

# Boost.Python library, to be linked against using bjam+gcc and bundled with
boost_include_dirs = ["/usr/include"]
boost_library_dirs = ['/usr/lib']
boost_libraries = []
boost_runtime_library_dirs = []
boost_runtime_libraries_to_be_bundled = []

Original issue reported on code.google.com by rea...@imojo.de on 5 Feb 2010 at 5:38

[GoogleCodeExporter]

Thanks for the info you have provided. It's very helpful. First of all, did you
download a Boost source archive or did you use one of the libboost- packages 
already
packaged on Ubuntu?

Original comment by pmtri80@gmail.com on 5 Feb 2010 at 6:07

• Changed state: Started

[GoogleCodeExporter]

This error only occurs if I'm using the boost install that comes with Ubuntu. 
Now I've tried to build it on my own 
and pyopencv compiles just prefect. So the trick is to build boost from source.

Thanks for your quick help!

Original comment by rea...@imojo.de on 6 Feb 2010 at 4:28

[GoogleCodeExporter]

Great! Thanks for letting me know. Actually, right now I haven't got time to 
tweak
the setup.py file to work with boost distributions other than the source
distribution. But I will work towards achieving that convenience.

Original comment by pmtri80@gmail.com on 6 Feb 2010 at 8:13

• Changed state: Fixed

[GoogleCodeExporter]

Is this issue resolved? 

I'm getting the same error while trying to package pyopencv. I can't continue 
packaging if I can't use the boost packages supplied by ubuntu.

Original comment by gijsmole...@gmail.com on 20 Apr 2010 at 1:44

[GoogleCodeExporter]

Unfortunately, it is not resolved. See my comments on issue 7.

To be honest, if you were wondering why I have chosen bjam, I would say because 
that
is the only build tool that appears to build a Python extension using 
Boost.Python
correctly. I do not want to use bjam myself because it is rather difficult to 
use.
However, I have not been able to build PyOpenCV successfully using other tools 
like
distutils or scons. Using those tools, I obtained Python modules which crashed 
on my
platforms. I would, however, be grateful if someone could tell me a successful 
case
of building a Python extension using Boost.Python with tools other than bjam.

Cheers,
Minh-Tri

Original comment by pmtri80@gmail.com on 20 Apr 2010 at 2:51

[GoogleCodeExporter]

If the issue isn't fixed, it shouldn't be marked as Fixed.  I'm getting the 
same 
error as well.

Either way, I don't think gijsmolenaar is complaining about bjam.  The only 
point is 
that while pyopencv can't be compiled with the toolkit packaged within 
distributions, 
it won't reach a wider audience.

To be sincere, it looks like pyopencv would be a good replacement for the 
extensions 
which come by default with opencv, but it really has to get to the point where 
it can 
be built with packaged tools.

Original comment by n13m3y3r on 4 May 2010 at 8:11

[GoogleCodeExporter]

Thanks, n13m3y3r, for your comment. I've changed the status back to Accepted. 
It's
not that I don't want pyopencv to reach a wider audience. However, right now 
it's not
easy to fix this issue at all. If someone can help me out at fixing it, or at 
least
point me to relevant sources, I will appreciate it. Otherwise, I'll try to look 
at
the matter more deeply when I manage to get some spare time.

Cheers,
Minh-Tri

Original comment by pmtri80@gmail.com on 4 May 2010 at 8:59

• Changed state: Accepted

[GoogleCodeExporter]

Thanks for your kind answer Minh-Tri.  I will try to see if I can find someone 
else 
with experience with bjam to tell us if the problem is in the packaged version 
of 
Ubuntu, or if there's an alternative way for pyopencv to be doing whatever it 
has to 
do.

I'm finding another compilation issue now.  I'll post it in another issue.

Original comment by n13m3y3r on 5 May 2010 at 6:24

[GoogleCodeExporter]

I just wrote a message to the boost maintainers in Debian.  Let's see if they 
can help 
us.

Original comment by n13m3y3r on 5 May 2010 at 6:46

[GoogleCodeExporter]

Thank you very much for your help, n13m3y3r. On my end, I will try again to see 
if
PyOpenCV can be built with scons. I have seen a successful use case of building
Boost.Python extensions using scons. If we can replace bjam with scons, all our
current issues with bjam will be resolved.

Cheers,
Minh-Tri

PS: Sorry for my late response, I had a very busy day yesterday.

Original comment by pmtri80@gmail.com on 6 May 2010 at 8:56

[GoogleCodeExporter]

I want to vote for using CMake as the build system for PyOpenCV. OpenCV is 
built with
CMake. If PyOpenCV is also built with CMake, there is less new things to learn +
there is a higher chance that PyOpenCV will eventually take over the ctypes 
OpenCV
python wrapper.

Original comment by dattan...@gmail.com on 17 May 2010 at 3:50

[GoogleCodeExporter]

Attached is part of the fix.

* Check out the source code into a src_folder
* Put this file into src_folder/pyopencv
* Use CMake (preferably GUI version) to configure your variables
* Compile

I haven't run the test suite of pyopencv on it.

Remaining things
* Run test suite on the generated .so file
* Add a post-build step to rename the .so file
* Add a pre-build step to run codegen.py so these generated pypp files are 
actually
generated.
* Test on other platforms

Original comment by dattan...@gmail.com on 18 May 2010 at 1:18

Attachments:

• CMakeLists.txt

[GoogleCodeExporter]

That's very cool, Dat. Thank you so much. I wonder why haven't I thought about 
CMake
before. I am rather busy from now until Sep 2010 but I will patch as soon as I 
can.

Cheers,
Minh-Tri

Original comment by pmtri80@gmail.com on 18 May 2010 at 6:47

[GoogleCodeExporter]

I have updated the CMakeLists.txt file so that it GLOB_RECURSE the C++ files 
under
pyopencvext. This will creates a pyopencvext.so. If I put this pyopencvext.so 
file in
the pyopencv folder. Running

import pyopencv

Doesn't give an error. However, when I try some simple code, I will get 
"dimensions
too large error"

from pyopencv import *
a = Mat(3,5,CV_32FC3)
a.setTo(10)

Perhaps Tri will have an idea why this error might happen. I can then fix the
CMakeLists.txt file further so we all could have a CMake version of PyOpenCV 
soon.

Original comment by dattan...@gmail.com on 19 May 2010 at 3:40

Attachments:

• CMakeLists.txt

[GoogleCodeExporter]

"ValueError: dimensions too large" is an exception raised by numpy when an 
ndarray
with a very large size is created. In your example, the only part that involves 
a
creation of an ndarray is when Python invokes repr() to obtain a representation
string for the returning Mat object of the 'a.setTo(10)' function call. I 
suspect
something is wrong at the part to create an ndarray view of a Mat object.

To confirm this, you could try 'a.setTo(10);' instead of 'a.setTo(10)' and see 
if no
error is raised. 

Original comment by pmtri80@gmail.com on 19 May 2010 at 6:50

[GoogleCodeExporter]

Ok, No error is raised. -_- I am confused. I thought that ; is optional in 
Python.

Original comment by dattan...@gmail.com on 19 May 2010 at 1:10

[GoogleCodeExporter]

It is definitely optional at the end of a line in terms of parsing.

What difference is it making exactly, and in which context?  Who's executing 
these 
lines, and what's the output in both cases?

Original comment by n13m3y3r on 19 May 2010 at 1:35

[GoogleCodeExporter]

The ';' character is optional in Python. But I thought you were using IPython. 
In
IPython, ';' means don't print out the result of the current expression.

Original comment by pmtri80@gmail.com on 19 May 2010 at 2:25

[GoogleCodeExporter]

I believe that the problem lies with the (auto) conversion into Numpy. It seems 
that
in order to display the matrix via print command (__repr__), the matrix is 
converted
to Numpy.

When I run a.ndarray I will also get the same (i.e. ValueError: dimensions too 
large).

Tri, can you explain briefly what is suppose to be built into the 
pyopencvext.so?

Original comment by dattan...@gmail.com on 19 May 2010 at 3:48

[GoogleCodeExporter]

I have made my CMakeLists.txt a direct translation from the Jamroot file. The 
problem
still remains. I have a hunch that this might have something to do with my 
system
being 64-bit.

For this CMakeLists.txt file to work, put it in the pyopencv-2.1.0.wr1.0.2 
folder.

Original comment by dattan...@gmail.com on 20 May 2010 at 1:46

Attachments:

• CMakeLists.txt

[GoogleCodeExporter]

There's nothing special about building pyopencvext.so. All I did was telling 
bjam to
compile every source file in the pyopencvext folder (recursively) and to link 
them
altogether. What I am afraid is that bjam may include some platform-specific
compilation flags and linking flags, which are hidden from us, in order to 
build C++
code properly. 

To get to know which flags have been provided to gcc, try adding the '-v' flags 
to
the compiler. To do this with PyOpenCV, all you have to do is to insert
"<compileflags>-v" to line 53 (i.e. the line containing "using gcc :  :  g++ :
<compileflags>-O3 ...") in the 'bjamcompiler.py' file, and run 'setup.py' again.

Original comment by pmtri80@gmail.com on 20 May 2010 at 1:59

[GoogleCodeExporter]

Oh, just saw your last post. Thanks.

Original comment by pmtri80@gmail.com on 20 May 2010 at 2:00

[GoogleCodeExporter]

Starting from r1025, PyOpenCV can be build via CMake (thanks Dat) as well as 
via setuptools. bjam is no longer required. If you have time, please feel free 
to try it out on your platform and let me know if there is any problem. 

To install PyOpenCV via CMake:
mkdir build
cd build
cmake ..
make
make install

To install PyOpenCV via setuptools (PyOpenCV will automatically invoke CMake to 
self-configure if the file 'config.py' is not found):
python setup.py install 

For now, I'd like to close this issue. Thank you all very much for your 
feedbacks.

Best regards,
Minh-Tri

Original comment by pmtri80@gmail.com on 27 Aug 2010 at 7:53

• Changed state: Fixed