I have the same kind of question, as I would like to read a boost serialized archive from Java, given that we have the corresponding classes in Java. Is there a java package for this or can someone point us to the specifications of the binary or text archive format. Thanks, Gaëtan 2010/10/22 Diederick C. Niehorster :

Not having used the serialization library, the easiest way is probably to write a MATLAB mex file that can open the file (or take a file pointer) and deliver the data to matlab in a format it understands.

Best, Dee

On Fri, Oct 22, 2010 at 04:29, James C. Sutherland wrote:

...

I am just looking into the serialization library.  It seems really nice for archiving using C++.

Is there a way to import an archive generated by that library into another language (e.g. matlab)?  I guess that one would need the rules for IO that the serialization archives use?

Thanks,

James

_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

On Fri, Oct 22, 2010 at 3:08 PM, Robert Valkenburg wrote:

We have done quite a bit of this type of thing, serialising data from a 3D scanner and then reading back into Matlab for analysis. We did exactly what Diederick suggested, wrote mex files. We serialise out of the scanner with boost serialisation and unserialise in the mex files with boost serialiation. Then the problem is really just one of mapping your C++ data structure to an equivalent data structure in Matlab.

The serialisation part is easy, the mex interface to matlab however is not so nice.

Would you be willing to share (on-list or off-list) some of the code that

you wrote to make this happen? I haven't done much with mex and frankly dread the thought. It would be great it I had an example to work from... Thanks, James

On Fri, Oct 22, 2010 at 8:18 PM, Robert Valkenburg wrote:

On Sat, Oct 23, 2010 at 2:41 PM, James C. Sutherland < James.Sutherland@utah.edu> wrote:

...

Would you be willing to share (on-list or off-list) some of the code that you wrote to make this happen? I haven't done much with mex and frankly dread the thought. It would be great it I had an example to work from...

Sure, I'm away from my work for a few days but will send you an example when i get back. I will send it off-list because its really more about Matlab mex than boost serialisation.

Thanks. FWIW, I just hacked together a really simple test case that seems to work. I will post here for the benefit of anyone who may be interested in the future. The serialization library is *really* nice! Three files: - Test.h - Test_mex.cpp - creates a function that can be called from Matlab that un-serializes the object - Test.cpp - creates the archive. ================= ================== #ifndef Test_h #define Test_h #include <ostream> #ifdef BINARY_IO # include # include typedef boost::archive::text_oarchive OutputArchive; typedef boost::archive::text_iarchive InputArchive; #else # include # include typedef boost::archive::binary_oarchive OutputArchive; typedef boost::archive::binary_iarchive InputArchive; #endif struct Test { private: friend std::ostream& operator<<(std::ostream& os, const Test& t ); friend class boost::serialization::access; template<typename Archive> void serialize( Archive& ar, const unsigned int version ) { ar & i; ar & d; ar & *p; } public: int i; double d; double* p; Test() { p=new double; } Test( int _i, double _d ) { i=_i; d=_d; p = new double; *p=d*2; } ~Test(){ delete p; } }; std::ostream& operator<<( std::ostream& os, const Test& t ) { os << t.i << ", " << t.d << ", " << *(t.p) << std::endl; } #endif Test_h ======================== ==================== ======================= =============== #include "mex.h" #include "Test.h" #include <iostream> #include <fstream> using namespace std; void mexFunction( int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[] ) { mexPrintf("Hello.\n"); Test t; ifstream in( "serialize.out",ios_base::in ); InputArchive ia(in); ia >> t; mexPrintf("%i\n",t.i); mexPrintf("%f\n",t.d); mexPrintf("%f\n",*(t.p)); } ================= ================ =============== ================== #include <iostream> #include <fstream> #include "Test.h" using namespace std; int main() { Test t(1,0.1); { std::ofstream ofs("serialize.out"); OutputArchive oa(ofs); oa << t; } cout << "original : " << t; { ifstream ifs("serialize.out",ios_base::in); InputArchive ia(ifs); Test t2; cout << "ready to import..." << flush; ia >> t2; cout << "done" << endl; cout << "restored : " << t2; } } ================== ===================

On Mon, Oct 25, 2010 at 3:27 PM, Daniel Herring wrote:

Isn't matlab capable of reading hdf5 files? That might be a reasonable target for boost::serialization.

True - Matlab has HDF5 IO support. However, HDF5 has a very poor c++ interface. While I would applaud a robust C++ interface for HDF5, I think that for boost to support it as an output target currently would be a painful endeavor. It is probably a better idea to build a MEX interface where the C++ classes are loaded natively and exported to Matlab. James

On Tue, 26 Oct 2010, Jesse Perla wrote:

James C. Sutherland writes:

...

On Mon, Oct 25, 2010 at 3:27 PM, Daniel Herring wrote:Isn't matlab capable of reading hdf5 files?  That might be a reasonable target for boost::serialization.

This was discussed briefly on the ublas list. Someone also told us that boost::serialization is really not meant for proprietary file formats.

While boost::serialization prefers its own file format, HDF5 is a very open standard. "The HDF5 data model, file format, API, library, and tools are open and distributed without charge." http://www.hdfgroup.org/HDF5/whatishdf5.html I understand that a full C++ object to HDF5 mapping is hard (I looked at the problem myself), but the format should not be dismissed lightly. In fact, I mentioned it as a possible format for matlab explicitly because it is a growing standard in scientific computing, whereas other matlab interfaces are vendor-specific. See for example http://www.hdfgroup.org/products/hdf5_tools/ The recent discussion about boost::mpi reawakened my interest in such a possibility... boost::hdf5? A full binding would be hard; but a binding sufficient for one or two datatype might not be. - Daniel