Re: [OctDev] [Changeset] Re: Java calling Octave via JNI

[Jaroslav Hajek]

On Tue, Sep 9, 2008 at 7:44 PM, John W. Eaton <address@hidden> wrote:
> On  5-Sep-2008, David Bateman wrote:
>
> | David Bateman wrote:
> | > MartinHepperle wrote:
> | >> Hello all,
> | >>
> | >> again this nasty Java -> Octave connection.
> | >>
> | >> I also want to embed Octave into a Java program and call Octave
> | >> functions
> | >> from Java.
> | >>
> | >> Initially I started with programming an execution engine which starts
> | >> octave.exe and connected to its standard input and output pipes. This
> | >> concept works (in the meantime I also found two solutions which
> | >> implement
> | >> the same technique) but is somewhat ugly and inefficient due to
> | >> conversions
> | >> to and from strings
> | >>
> | >> What I tried now was to call the Octave DLLs (Windows platform)
> | >> directly via
> | >> a C wrapper and using JNI.
> | >> This technique provides a very elegant solution and I have used this
> | >> in the
> | >> past to call C and FORTRAN codes from Java. I want to define some global
> | >> variables, set the values from Java, execute some user supplied
> | >> functions
> | >> and finally return the results (global variables set inside the user
> | >> functions) to Java.
> | >>
> | >> Now with the Octrave DLLs I stumbled across a problem.
> | >>
> | >> To study the interface to ther DLLs I started with a C++ program.
> | >> Basically I used the following logic:
> | >>
> | >> //-----------------------------
> | >> // start and initialize Octave (once, then keep it alive)
> | >> char * pArgs[] = {"octave.exe","--path", "C:\\Programme\\Octave",
> | >> "--norc",
> | >> NULL};
> | >> int iRet = octave_main (4 pArgs, true);
> | >>
> | >> //-----------------------------
> | >> // execute some commands
> | >> std::string s = std::string("global Output;\nglobal
> | >> Input;\nInput=12;\nOutput=2*Input;\n");
> | >> octave_value x = eval_string (s, false, parse_status);
> | >> // examine the returned "x" with x.isnumeric() etc. works when code
> | >> is in an
> | >> executable (.exe) file
> | >>
> | >> // what does the follwing code do? It works, but where is the
> | >> variable in
> | >> Octave?
> | >> set_global_value (std::string(cName), val);
> | >> octave_value vReturn = get_global_value (std::string(cName));
> | >> // the variable has the correct value, but is_globally_visible
> | >> (std::string(Name)) returns false...
> | >> // what does global mean here?
> | >>
> | >>
> | >>
> | >> // terminate Octave
> | >> do_octave_atexit();
> | >> //-----------------------------
> | >>
> | >> All this works well when I have the code inside an executable C++
> | >> program
> | >> (MSVC8.0).
> | >>
> | >> So far so good.
> | >>
> | >> Now for the interface via JNI I have to put this into a DLL which
> | >> provides a
> | >> thin layer between Java objects and the C interface to Octave.
> | >> If I put the same code inside a DLL I can initialize Octave, but the
> | >> eval_string () and other functions doe not work - they seem to do
> | >> nothing. I found can do somethings with set_global_value () and
> | >> get_global_value (),
> | >> but this is not enough and also these "global" values seem to be
> | >> something
> | >> else than global variables in Octave.
> | >>
> | >> Anyway, I see that eval_string() seems to work when called from a C
> | >> executable, but not when I call it from a DLL. Could this have
> | >> something to
> | >> do with i/o streams which are different in a DLL and in an EXE?
> | >>
> | >> Any ideas where to look and how to test?
> | >>
> | >>
> | >> Thank You,
> | >>
> | >> Martin
> | >>
> | >>
> | >>
> | >>
> | >>
> | > eval_string can only work if the interpreter is initialized.. Check
> | > the mailing lists for how to initialize the interpreter.
> | >
> | > D.
> | >
> | >
> | Seeing as how this question has come up a number of times, I suggest the
> | attached changeset that adds an example of how to initialize the
> | interpreter and use it from a standalone application.
>
> I applied it.
>

I transplanted this "explanationation" (hg description :-) patch to 3.0.x series




-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz