SWIG newbie questions: R SWIG errors or warnings? is R SWIG functional?

[cottrell]

Doing:

swig -c++ -v -r -o simpletest_wrap.cpp simpletest.i
PKG_LIBS="simpletest.cpp" R CMD SHLIB simpletest_wrap.cpp

yields the following reports

simpletest_wrap.cpp:725:2: warning: #warning "R Swig currently fails on
runtime for 2.6 and higher."
simpletest_wrap.cpp:726:2: warning: #warning "Contact maintainer if you
would like to help fix"

Does this mean that R SWIG is not really functional yet?

My hope was to implement a simulation template class in C++ and
construct wrappers with SWIG for R and octave to plot things and compile
output, but I'm no longer confident that it is possible.

To make this worthwhile I would need to be able to:

1) instantiate wrapped c++ data (parameter) classes from R/octave.  
Wrapped constructors take vector arguments.

2) instantiate wrapped c++ simulation class from R/octave.  Wrapped
constructors take wrapped class instances (from 1) as arguments.

3) call simulation class member functions from R/octave (i.e.
simulation.run and simulation.output type things).

Is what I'm describing possible to do without major pains?  Would it be
way easier with SWIG python?

Any feedback is greatly appreciated ... at this point I'm just about
ready to do the read/write to text file solution or write a wrapper for
R, I've already spent way too long fiddling with SWIG.

david


// simpletest.i
%module simpletest

%{
#include "./simpletest.cpp"
%}

%include "./simpletest.cpp"

%template(something3) something<3>; // this *is* a 3 (I am templating on
a spatial dimension d=1,2,3)



// simpletest.cpp
//#include "/usr/include/boost/numeric/ublas/vector.hpp"
#include <boost/numeric/ublas/vector.hpp>
#include <boost/numeric/ublas/io.hpp>
#include <iostream>

// this is the line which causes problems:
//namespace boost {namespace numeric {namespace ublas {}}}
//using namespace boost::numeric::ublas;

template<int d> class something {
        double thingy;
        boost::numeric::ublas::vector<int> a;
        public:
                something(): a(3){
                        a(1)=1;
                        a(3)=3;
                        a(4)=2;
                        info();
                }
                void info(){
                        std::cout << a;
                }
               
};


--
David Cottrell
http://www.math.mcgill.ca/~cottrell 


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Tue, 2008-04-22 at 10:44 -0400, David Cottrell wrote:
> Doing:
>
> swig -c++ -v -r -o simpletest_wrap.cpp simpletest.i
> PKG_LIBS="simpletest.cpp" R CMD SHLIB simpletest_wrap.cpp

I guess you use swig 1.3.34 and R2.6? Try swig 1.3.35 then!

> yields the following reports
>
> simpletest_wrap.cpp:725:2: warning: #warning "R Swig currently fails on
> runtime for 2.6 and higher."
> simpletest_wrap.cpp:726:2: warning: #warning "Contact maintainer if you
> would like to help fix"
>
> Does this mean that R SWIG is not really functional yet?

YMMV....

> My hope was to implement a simulation template class in C++ and
> construct wrappers with SWIG for R and octave to plot things and compile
> output, but I'm no longer confident that it is possible.
>
> To make this worthwhile I would need to be able to:
>
> 1) instantiate wrapped c++ data (parameter) classes from R/octave.  
> Wrapped constructors take vector arguments.
>
> 2) instantiate wrapped c++ simulation class from R/octave.  Wrapped
> constructors take wrapped class instances (from 1) as arguments.
>
> 3) call simulation class member functions from R/octave (i.e.
> simulation.run and simulation.output type things).
>
> Is what I'm describing possible to do without major pains?  Would it be
> way easier with SWIG python?
>
> Any feedback is greatly appreciated ... at this point I'm just about
> ready to do the read/write to text file solution or write a wrapper for
> R, I've already spent way too long fiddling with SWIG.

my experience so far is that swig + python & octave work nicely... I was
not successful with R, but just try it out ... I would be interested in
learning whether it works for you (or someone).

> david
>
>
> // simpletest.i
> %module simpletest
>
> %{
> #include "./simpletest.cpp"
> %}
>
> %include "./simpletest.cpp"
>
> %template(something3) something<3>; // this *is* a 3 (I am templating on
> a spatial dimension d=1,2,3)
>
>
>
> // simpletest.cpp
> //#include "/usr/include/boost/numeric/ublas/vector.hpp"
> #include <boost/numeric/ublas/vector.hpp>
> #include <boost/numeric/ublas/io.hpp>
> #include <iostream>
>
> // this is the line which causes problems:
> //namespace boost {namespace numeric {namespace ublas {}}}
> //using namespace boost::numeric::ublas;
>
> template<int d> class something {
>         double thingy;
>         boost::numeric::ublas::vector<int> a;
>         public:
>                 something(): a(3){
>                         a(1)=1;
>                         a(3)=3;
>                         a(4)=2;
>                         info();
>                 }
>                 void info(){
>                         std::cout << a;
>                 }
>                
> };

Soeren

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[wsfulton]

Soeren Sonnenburg wrote:

> On Tue, 2008-04-22 at 10:44 -0400, David Cottrell wrote:
> > Doing:
> >
> > swig -c++ -v -r -o simpletest_wrap.cpp simpletest.i
> > PKG_LIBS="simpletest.cpp" R CMD SHLIB simpletest_wrap.cpp
>
> I guess you use swig 1.3.34 and R2.6? Try swig 1.3.35 then!
>
> > yields the following reports
> >
> > simpletest_wrap.cpp:725:2: warning: #warning "R Swig currently fails on
> > runtime for 2.6 and higher."
> > simpletest_wrap.cpp:726:2: warning: #warning "Contact maintainer if you
> > would like to help fix"
> >
> > Does this mean that R SWIG is not really functional yet?
>
> YMMV....

...
> my experience so far is that swig + python & octave work nicely... I was
> not successful with R, but just try it out ... I would be interested in
> learning whether it works for you (or someone).
>
SWIG works fine with R <= 2.6, most of the test-suite works well. Not
sure about 2.7, they seem to knock out new versions of R rather quickly
and there were some non-backwards compatible changes going from 2.5 to 2.6.

William

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Thu, 2008-04-24 at 22:50 +0100, William S Fulton wrote:

> Soeren Sonnenburg wrote:
> > On Tue, 2008-04-22 at 10:44 -0400, David Cottrell wrote:
> > > Doing:
> > >
> > > swig -c++ -v -r -o simpletest_wrap.cpp simpletest.i
> > > PKG_LIBS="simpletest.cpp" R CMD SHLIB simpletest_wrap.cpp
> >
> > I guess you use swig 1.3.34 and R2.6? Try swig 1.3.35 then!
> >
> > > yields the following reports
> > >
> > > simpletest_wrap.cpp:725:2: warning: #warning "R Swig currently fails on
> > > runtime for 2.6 and higher."
> > > simpletest_wrap.cpp:726:2: warning: #warning "Contact maintainer if you
> > > would like to help fix"
> > >
> > > Does this mean that R SWIG is not really functional yet?
> >
> > YMMV....
> ...
> > my experience so far is that swig + python & octave work nicely... I was
> > not successful with R, but just try it out ... I would be interested in
> > learning whether it works for you (or someone).
> >
> SWIG works fine with R <= 2.6, most of the test-suite works well. Not
> sure about 2.7, they seem to knock out new versions of R rather quickly
> and there were some non-backwards compatible changes going from 2.5 to 2.6.

Except for

- Checking testcase constants under r
- Checking testcase cpp_broken under r
- Checking testcase namespace_union under r
- Checking testcase nested_comment under r
- Checking testcase template_default_pointer under r
- Checking testcase template_expr under r

all tests run through on R 2.7 too (swig-svn version, my patch applied).
Regarding the changes from R2.5->2.7 the RVERSION25/26/27 makro taking
care of this only changes the way how strings are created - a one liner
change - I don't think that this will break anything.

But still R 2.7. is now the current stable which Rswig doesn't support.

However R & swig did not work for me (already crashing when only loading
the libs) but it might be that this is a bug in R 2.7... I am still
investigating...

Soeren

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Fri, 2008-04-25 at 08:50 +0200, Soeren Sonnenburg wrote:

> On Thu, 2008-04-24 at 22:50 +0100, William S Fulton wrote:
> > Soeren Sonnenburg wrote:
> > > On Tue, 2008-04-22 at 10:44 -0400, David Cottrell wrote:
> > > > Doing:
> > > >
> > > > swig -c++ -v -r -o simpletest_wrap.cpp simpletest.i
> > > > PKG_LIBS="simpletest.cpp" R CMD SHLIB simpletest_wrap.cpp
> > >
> > > I guess you use swig 1.3.34 and R2.6? Try swig 1.3.35 then!
> > >
> > > > yields the following reports
> > > >
> > > > simpletest_wrap.cpp:725:2: warning: #warning "R Swig currently fails on
> > > > runtime for 2.6 and higher."
> > > > simpletest_wrap.cpp:726:2: warning: #warning "Contact maintainer if you
> > > > would like to help fix"
> > > >
> > > > Does this mean that R SWIG is not really functional yet?
> > >
> > > YMMV....
> > ...
> > > my experience so far is that swig + python & octave work nicely... I was
> > > not successful with R, but just try it out ... I would be interested in
> > > learning whether it works for you (or someone).
> > >
> > SWIG works fine with R <= 2.6, most of the test-suite works well. Not
> > sure about 2.7, they seem to knock out new versions of R rather quickly
> > and there were some non-backwards compatible changes going from 2.5 to 2.6.
>
> Except for
>
> - Checking testcase constants under r
> - Checking testcase cpp_broken under r
> - Checking testcase namespace_union under r
> - Checking testcase nested_comment under r
> - Checking testcase template_default_pointer under r
> - Checking testcase template_expr under r
>
> all tests run through on R 2.7 too (swig-svn version, my patch applied).
> Regarding the changes from R2.5->2.7 the RVERSION25/26/27 makro taking
> care of this only changes the way how strings are created - a one liner
> change - I don't think that this will break anything.
>
> But still R 2.7. is now the current stable which Rswig doesn't support.
>
> However R & swig did not work for me (already crashing when only loading
> the libs) but it might be that this is a bug in R 2.7... I am still
> investigating...

OK, after investigating for some time:

There was actually not only a bug in SWIG but in R 2.7 (and still is -
long input lines cause it to crash, solution posted to Rd mailinglist).
Which means R+swig works in principle. However there are definitely a
couple of (minor/easily fixable) bugs in the R wrapper that need further
fixes before shogun ( http://www.shogun-toolbox.org ) fully works:

1) the dispatcher (called when an object is constructed) wrongly calls
    extends(argtypes[1], 'numeric')) which will never be true for
    a matrix (some magic is needed here).
    Workaround for now is to not use
    RealFeatures(my_matrix), but
    rf=RealFeatures()
    rf$copy_feature_matrix(rf, my_matrix)
2) arguments that take a double* will all be modified via
   as.numeric(arg) which converts matrices -> vectors. IMHO it
   does not make sense to do any conversion
   at all but for compatibility a workaround for now is to use
   %{  if (!is.numeric($input)) $input = as.numeric($input) %}
   in the scoercein typemap in rtype.swg
3) the R2.7 swig mkCharEnc fix (patch submitted here on the ml)

Soeren

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Joseph Wang-2]

Thanks.  I've noticed the R 2.7 release and checked in some fixes before I
checked my e-mail this evenining.  I'll try to get the patches in place in
the next few days.  R 2.7 seems to pass all the regression tests except for
the function pointer one, and I got it to work with QuantLib.

On Friday 25 April 2008 12:40:21 pm Soeren Sonnenburg wrote:

> OK, after investigating for some time:
>
> There was actually not only a bug in SWIG but in R 2.7 (and still is -
> long input lines cause it to crash, solution posted to Rd mailinglist).
> Which means R+swig works in principle. However there are definitely a
> couple of (minor/easily fixable) bugs in the R wrapper that need further
> fixes before shogun ( http://www.shogun-toolbox.org ) fully works:
>
> 1) the dispatcher (called when an object is constructed) wrongly calls
>     extends(argtypes[1], 'numeric')) which will never be true for
>     a matrix (some magic is needed here).
>     Workaround for now is to not use
>     RealFeatures(my_matrix), but
>     rf=RealFeatures()
>     rf$copy_feature_matrix(rf, my_matrix)
> 2) arguments that take a double* will all be modified via
>    as.numeric(arg) which converts matrices -> vectors. IMHO it
>    does not make sense to do any conversion
>    at all but for compatibility a workaround for now is to use
>    %{  if (!is.numeric($input)) $input = as.numeric($input) %}
>    in the scoercein typemap in rtype.swg
> 3) the R2.7 swig mkCharEnc fix (patch submitted here on the ml)
>
> Soeren

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Sat, 2008-04-26 at 04:13 -0400, Joseph Wang wrote:

Hi Joseph,

> Thanks.  I've noticed the R 2.7 release and checked in some fixes before I
> checked my e-mail this evenining.  I'll try to get the patches in place in

Yes I get some conflicts now - I am fine to not have an extra case for
R2.6 :)

> the next few days.  R 2.7 seems to pass all the regression tests except for
> the function pointer one, and I got it to work with QuantLib.

What about the other two problems I am seeing? Why can't we get rid of
the scorecein typemap in rtype.swg at all? I mean it is bad if things
get converted to numeric (or so) although there is a magic (user
written!) typemap doing the work. I still don't understand why ther
should be any conversion at all... wrong type -> sorry wrong type error
no?

Also what about the dispatcher? Any ideas on how to better check if two
types are compatible (extends seems not be the right thing for numeric
types)...

Soeren

> On Friday 25 April 2008 12:40:21 pm Soeren Sonnenburg wrote:
>
> > OK, after investigating for some time:
> >
> > There was actually not only a bug in SWIG but in R 2.7 (and still is -
> > long input lines cause it to crash, solution posted to Rd mailinglist).
> > Which means R+swig works in principle. However there are definitely a
> > couple of (minor/easily fixable) bugs in the R wrapper that need further
> > fixes before shogun ( http://www.shogun-toolbox.org ) fully works:
> >
> > 1) the dispatcher (called when an object is constructed) wrongly calls
> >     extends(argtypes[1], 'numeric')) which will never be true for
> >     a matrix (some magic is needed here).
> >     Workaround for now is to not use
> >     RealFeatures(my_matrix), but
> >     rf=RealFeatures()
> >     rf$copy_feature_matrix(rf, my_matrix)
> > 2) arguments that take a double* will all be modified via
> >    as.numeric(arg) which converts matrices -> vectors. IMHO it
> >    does not make sense to do any conversion
> >    at all but for compatibility a workaround for now is to use
> >    %{  if (!is.numeric($input)) $input = as.numeric($input) %}
> >    in the scoercein typemap in rtype.swg
> > 3) the R2.7 swig mkCharEnc fix (patch submitted here on the ml)
> >
> > Soeren
>
>

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Joseph Wang-2]

On Saturday 26 April 2008 04:57:26 am Soeren Sonnenburg wrote:
> What about the other two problems I am seeing? Why can't we get rid of
> the scorecein typemap in rtype.swg at all? I mean it is bad if things
> get converted to numeric (or so) although there is a magic (user
> written!) typemap doing the work. I still don't understand why ther
> should be any conversion at all...

A lot of R-Swig was written with trial and error.  I just removed the
scorecein typemaps for integers and floats, and the regression tests ran fine
as did my quantlib scripts.  So I just commented them out.

> Also what about the dispatcher? Any ideas on how to better check if two
> types are compatible (extends seems not be the right thing for numeric
> types)...

No ideas, but I'm interested in anything anyone else can think of.....



-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Sat, 2008-04-26 at 15:43 -0400, Joseph Wang wrote:

> On Saturday 26 April 2008 04:57:26 am Soeren Sonnenburg wrote:
> > What about the other two problems I am seeing? Why can't we get rid of
> > the scorecein typemap in rtype.swg at all? I mean it is bad if things
> > get converted to numeric (or so) although there is a magic (user
> > written!) typemap doing the work. I still don't understand why ther
> > should be any conversion at all...
>
> A lot of R-Swig was written with trial and error.  I just removed the
> scorecein typemaps for integers and floats, and the regression tests ran fine
> as did my quantlib scripts.  So I just commented them out.

Why not remove them completely, also the as($input, ...) scoerecein?

Because now on swig-svn still any matrix is converted to a vector due to
these scoerecein typemaps. I just removed them (see attached patch) and
the tests still run through and now also in my project ( shogun )
copy_matrix / get_fm (matrix in/out) work.

> > Also what about the dispatcher? Any ideas on how to better check if two
> > types are compatible (extends seems not be the right thing for numeric
> > types)...
>
> No ideas, but I'm interested in anything anyone else can think of.....

We could make an exception for the numeric type, like instead of asking
for extends(x,'numeric') we ask is.numeric(x) ?

That would be sufficient...

Soeren

[rtype.diff]

---

Index: Lib/r/rtype.swg
===================================================================
--- Lib/r/rtype.swg (revision 10394)
+++ Lib/r/rtype.swg (working copy)
@@ -36,6 +36,7 @@
 */
 
 
+/*
 %typemap(scoercein) int, int *, int &
   %{  $input = as($input, "integer");     %}
 %typemap(scoercein) ptrdiff_t, ptrdiff_t *, ptrdiff_t &
@@ -78,11 +79,9 @@
  %}
 
 
-/* **************************************************************** */
 
 %typemap(scoercein) bool, bool *, bool &
     "$input = as.logical($input) ";
-/*
 %typemap(scoercein) int,
                     int *,
     int &,
@@ -119,11 +118,11 @@
                     long*,
                     long long *
 %{  $input = as.numeric($input) %}
-*/
 
 %typemap(scoercein) char *, string, std::string,
 string &, std::string &
 %{  $input = as($input, "character") %}
+*/
 
 %typemap(scoerceout) enum SWIGTYPE
   %{  $result = enumFromInteger($result, "$R_class") %}

---

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Sat, 2008-04-26 at 15:43 -0400, Joseph Wang wrote:
> On Saturday 26 April 2008 04:57:26 am Soeren Sonnenburg wrote:
[...]
> > Also what about the dispatcher? Any ideas on how to better check if two
> > types are compatible (extends seems not be the right thing for numeric
> > types)...
>
> No ideas, but I'm interested in anything anyone else can think of.....

OK, I came up with a patch that solves this problem (for me):

I save the args in a argv list and if use is.numeric(argv[[1]]) for
numeric types ...

Another issue I've noticed is with 'garbage collection': There seems to
be a problem when I use

%feature("ref")  and %feature("unref"), could it be a problem that the
C++ functions I've bound to this are named ref() and unref()?

Soeren

[r_cxx.diff]

---

Index: Source/Modules/r.cxx
===================================================================
--- Source/Modules/r.cxx (revision 10394)
+++ Source/Modules/r.cxx (working copy)
@@ -1604,6 +1604,7 @@
   int   nfunc = Len(dispatch);
   Printv(f->code,
  "argtypes <- mapply(class, list(...))\n",
+ "argv <- list(...)\n",
  "argc <- length(argtypes)\n", NIL );
 
   Printf(f->code, "# dispatch functions %d\n", nfunc);
@@ -1637,10 +1638,17 @@
  if(tm) {
   replaceRClass(tm, Getattr(p, "type"));
  }
+ if (DohStrcmp(tm,"numeric")==0) {
+ Printf(f->code, "%sis.numeric(argv[[%d]])",
+       j == 0 ? "" : " && ",
+       j+1);
+ }
+ else {
  Printf(f->code, "%sextends(argtypes[%d], '%s')",
        j == 0 ? "" : " && ",
        j+1,
        tm);
+ }
  p = Getattr(p, "tmap:in:next");
       }
       Printf(f->code, ") { f <- %s%s }\n", sfname, overname);

---

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Joseph Wang-2]

On Sunday 27 April 2008 01:24:58 am Soeren Sonnenburg wrote:

About scoercein, I commented them in the Lib/r files which means that it won't
emit any of the coercion code.

> OK, I came up with a patch that solves this problem (for me):
>
> I save the args in a argv list and if use is.numeric(argv[[1]]) for
> numeric types ...

The patch you put in doesn't seem to break any of the regression tests or my
QuantLib demos, so I've checked it into trunk.

> Another issue I've noticed is with 'garbage collection': There seems to
> be a problem when I use
>
> %feature("ref")  and %feature("unref"), could it be a problem that the
> C++ functions I've bound to this are named ref() and unref()?

What's the exact nature of the problem?  I haven't much work with making sure
that memory management works correctly so there might be a lot of things that
are broken.



-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Sun, 2008-04-27 at 14:29 -0400, Joseph Wang wrote:
> On Sunday 27 April 2008 01:24:58 am Soeren Sonnenburg wrote:
>
> About scoercein, I commented them in the Lib/r files which means that it won't
> emit any of the coercion code.

OK, for shogun I will start porting all the examples I have for
octave/python to R. And play around a bit. Lets see whether I notice a
problem (should be of minor nature anyway, worst that could happen is
that something has to be recast to be numeric :)

> > OK, I came up with a patch that solves this problem (for me):
> >
> > I save the args in a argv list and if use is.numeric(argv[[1]]) for
> > numeric types ...
>
> The patch you put in doesn't seem to break any of the regression tests or my
> QuantLib demos, so I've checked it into trunk.

Yeah that shouldn't do any harm.

> > Another issue I've noticed is with 'garbage collection': There seems to
> > be a problem when I use
> >
> > %feature("ref")  and %feature("unref"), could it be a problem that the
> > C++ functions I've bound to this are named ref() and unref()?
>
> What's the exact nature of the problem?  I haven't much work with making sure
> that memory management works correctly so there might be a lot of things that
> are broken.

Well I've noticed a double free when running a toy example while
ref/unref were used (these work in octave/python)... so I guess
something is not inc_ref'd or dec_ref'd twice...

Anyway this is (for now) the least important thing. Anyway if both
quantlib and shogun work using swig then Rswig has some serious+big
enough applications and the regression tests are complex enough such
that one could actually recommend it...

Thanks for your help, I will ping back when I discover an issue (apart
from ref/unref).

Soeren





-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user

[Soeren Sonnenburg-9]

On Sun, 2008-04-27 at 14:29 -0400, Joseph Wang wrote:
> On Sunday 27 April 2008 01:24:58 am Soeren Sonnenburg wrote:

Hi Joseph,

I've now managed to get all examples from octave/python ported to R,
however I needed some tiny adjustment to rtype.swg get them to work (see
attached patch).

> About scoercein, I commented them in the Lib/r files which means that it won't
> emit any of the coercion code.

This seems not enough, as as($input, "numeric") is still turning the
matrix into a flat vector. Simply commenting removing the
$input = as($input, "numeric"); command from scoercein makes things work
for me.

Another issue I stumbled accross: I have some typemap working on a
templated structure, which is correctly wrapped by R. However again the
call to function tries to do a coercion where none is necessary. The fix
here was again to remove all the SWIGTYPE, SWIGTYPE*, SWIGTYPE&
coerceIfNotSubclass calls.

I've ran the whole testsuite and still every example (except for one
that didn't compile) works fine :-)

So please try with quantlib and apply or fix and apply :)

> > Another issue I've noticed is with 'garbage collection': There seems to
> > be a problem when I use
> >
> > %feature("ref")  and %feature("unref"), could it be a problem that the
> > C++ functions I've bound to this are named ref() and unref()?
>
> What's the exact nature of the problem?  I haven't much work with making sure
> that memory management works correctly so there might be a lot of things that
> are broken.

I am for now ignoring garbage collection. What I still don't understand
is the ref warning I am getting (even though I have my ref/unref code
commented out):

> x=LDA()
> x
An object of class "_p_CLDA"
Slot "ref":
Error in slot(object, what) :
  no slot of name "ref" for this object of class "_p_CLDA"
>

where is that ref coming from? And what should it be? Is this about
reference counting? I am not sure what I should do to get rid of this...
any ideas?

Soeren

[rtype.diff]

---

Index: Lib/r/rtype.swg
===================================================================
--- Lib/r/rtype.swg (revision 10412)
+++ Lib/r/rtype.swg (working copy)
@@ -45,9 +45,9 @@
 %typemap(scoercein) unsigned int, unsigned int *, unsigned int &
   %{  $input = as($input, "integer");     %}
 %typemap(scoercein) double, double *, double &
-  %{  $input = as($input, "numeric");     %}
+  %{   %}
 %typemap(scoercein) float, float *, float &
-  %{  $input = as($input, "numeric");     %}
+  %{   %}
 %typemap(scoercein) char, char *, char &
   %{  $input = as($input, "character");     %}
 %typemap(scoercein) string, string *, string &
@@ -61,7 +61,10 @@
 %typemap(scoercein) enum SWIGTYPE *
   %{  $input = enumToInteger($input, "$R_class") %}
 
-%typemap(scoercein) SWIGTYPE *
+%typemap(scoercein) SWIGTYPE, SWIGTYPE *, SWIGTYPE &
+  %{  %}
+
+/*%typemap(scoercein) SWIGTYPE*
   %{ $input = coerceIfNotSubclass($input, "$R_class") %}
 
 %typemap(scoercein) SWIGTYPE &
@@ -69,7 +72,7 @@
 
 %typemap(scoercein) SWIGTYPE  
   %{ $input = coerceIfNotSubclass($input, "$&R_class") %}
-
+*/
 
 %typemap(scoercein) SWIGTYPE[ANY]  
  %{

---

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Swig-user mailing list
Swig-user@...
https://lists.sourceforge.net/lists/listinfo/swig-user