lucene-pylucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Petrus Hyvönen <petrus.hyvo...@gmail.com>
Subject Re: Problem using generic types?
Date Sun, 02 Feb 2014 22:26:31 GMT
Hi Andi,

Yes, the confusing thing is that it works well under windows. I compared
the code in __wrap__ and it looks different in the windows version:

 static int t_PointVectorValuePair_init_(t_PointVectorValuePair *self,
PyObject *args, PyObject *kwds)
          {
            switch (PyTuple_GET_SIZE(args)) {
             case 2:
              {
                JArray< jdouble > a0((jobject) NULL);
                JArray< jdouble > a1((jobject) NULL);
                PointVectorValuePair object((jobject) NULL);

                if (!parseArgs(args, "[D[D", &a0, &a1))
                {
                  INT_CALL(object = PointVectorValuePair(a0, a1));
                  self->object = object;
                  break;
                }
              }


So no Parameters[] stuff in windows version.. On mac I'm using java v
1.7.0_45 and 1.6.0_35 under windows. could that make a difference?

I'm using a mvn downloaded version of the apache.math jar, but source is:

public class PointVectorValuePair extends Pair<double[], double[]>
implements Serializable {
    /** Serializable UID. */
    private static final long serialVersionUID = 20120513L;

    /**
     ...
     */
    public PointVectorValuePair(final double[] point,
                                final double[] value) {
        this(point, value, true);
    }

    /**
    ...
     */
    public PointVectorValuePair(final double[] point,
                                final double[] value,
                                final boolean copyArray) {
        super(copyArray ?
              ((point == null) ? null :
               point.clone()) :
              point,
              copyArray ?
              ((value == null) ? null :
               value.clone()) :
              value);
    }

...

full code at
http://commons-math/jacoco/org.apache.commons.math3.optimization/PointVectorValuePair.java.html



Regards
/petrus


On 02 Feb 2014, at 22:54 , Andi Vajda <vajda@apache.org> wrote:


On Feb 2, 2014, at 13:41, Petrus Hyvönen <petrus.hyvonen@gmail.com> wrote:

Hi,

It's part of a class "PointVectorValuePair" and "PointValuePair" in the
apache math commons (
http://commons.apache.org/proper/commons-math/javadocs/api-3.2/org/apache/commons/math3/optimization/PointVectorValuePair.html
)

I have tried to reserve a bunch of things like 'Point', 'Value', 'Pair'
without any difference. If I go through the __wrap__ file, it looks like
this [D object to wrap is a very odd one, the other PY_TYPE wrapping seems
to have 'proper' names without special characters. It's the functions
PointValuePair and PointVectorValuePair that seems to have this strage
class wrapping.  I cannot find a class named D. From the API doc's I would
guess it's trying to set a double [] type.

extract from __wrap__:

under namespace org {
namespace apache {
  namespace commons {
    namespace math3 {
      namespace optimization {

...

        static int t_PointVectorValuePair_init_(t_PointVectorValuePair
*self, PyObject *args, PyObject *kwds)
        {
          switch (PyTuple_GET_SIZE(args)) {
           case 2:
            {
              JArray< jdouble > a0((jobject) NULL);
              JArray< jdouble > a1((jobject) NULL);
              PointVectorValuePair object((jobject) NULL);

              if (!parseArgs(args, "[D[D", &a0, &a1))


It could just be that you found a bug in jcc's handling of generic
parameters that are arrays. I suspect you have some double[] (or worse:
double[][]) somewhere in the java source.
But then this should fail the same way on Windows.
Can you please post the original java source for that class here ?

Andi..

              {
                INT_CALL(object = PointVectorValuePair(a0, a1));
                self->object = object;
                self->parameters[0] = &::PY_TYPE([D);
                self->parameters[1] = &::PY_TYPE([D);
                break;
              }

Regards
/petrus



= &::PY_TYPE([D);
                                                 ^
/Users/petrus/anaconda/envs/_build/lib/python2.7/site-packages/jcc/sources/macros.h:66:23:
note:
    expanded from macro 'PY_TYPE'
#define PY_TYPE(name) name##$$Type
                    ^
build/_orekit/__wrap__.cpp:58329:44: error: use of undeclared identifier
    'D$$Type'
                self->parameters[0] = &::PY_TYPE([D);
                                         ^
/Users/petrus/anaconda/envs/_build/lib/python2.7/site-packages/jcc/sources/macros.h:66:23:
note:
    expanded from macro 'PY_TYPE'
#define PY_TYPE(name) name##$$Type
                    ^
<scratch space>:109:1: note: expanded from here
D$$Type
^
build/_orekit/__wrap__.cpp:58329:55: error: expected ']'
                self->parameters[0] = &::PY_TYPE([D);
                                                    ^
build/_orekit/__wrap__.cpp:58329:52: note: to match this '['
                self->parameters[0] = &::PY_TYPE([D);


On 02 Feb 2014, at 19:48 , Andi Vajda <vajda@apache.org> wrote:


On Feb 2, 2014, at 10:08, Petrus Hyvönen <petrus.hyvonen@gmail.com> wrote:

Hi Andi and Others,

The latest trunk jcc works and builds very fine on my windows machine, and
happy about the extendend support of generics. But I am experiencing the
problem below when building on my mac, using same wrap parameters as on the
windows platform. To me it seems like some compiler problem related to
macros, but don't know.

Does anyone have experience of these failures?

Regards
/Petrus


error: expected unqualified-id
              self->parameters[0] = &::PY_TYPE([D);
                                               ^
/Users/petrus/anaconda/envs/_build/lib/python2.7/site-packages/jcc/sources/macros.h:66:23:
note:
  expanded from macro 'PY_TYPE'
#define PY_TYPE(name) name##$$Type
                  ^
build/_orekit/__wrap__.cpp:2344:44: error: use of undeclared identifier
  'D$$Type'
              self->parameters[0] = &::PY_TYPE([D);


Take a look at the code around line 2344 in that __wrap__.cpp file and
figure out what java code this corresponds to. You might be using a
classname that's a C macro on some platforms - do you have a class named D
for example ?
If you're hitting such a name conflict, add that name to the --reserved
list passed to jcc.

Andi..

                                       ^
/Users/petrus/anaconda/envs/_build/lib/python2.7/site-packages/jcc/sources/macros.h:66:23:
note:
  expanded from macro 'PY_TYPE'
#define PY_TYPE(name) name##$$Type
                  ^
<scratch space>:73:1: note: expanded from here
D$$Type
^
build/_orekit/__wrap__.cpp:2344:55: error: expected ']'
              self->parameters[0] = &::PY_TYPE([D);
                                                  ^
build/_orekit/__wrap__.cpp:2344:52: note: to match this '['
              self->parameters[0] = &::PY_TYPE([D);



On 14 Jan 2014, at 19:02 , Andi Vajda <vajda@apache.org> wrote:


Hi Petrus,

On Jan 14, 2014, at 4:24, Petrus Hyvönen <petrus.hyvonen@gmail.com> wrote:

Dear Andi,

Many many thanks for looking into this!

I am on travel for a week and do not have the computer with the special
case with me to test. I did now however run it on the library that I am
wrapping, and there get some errors in the creation of the wrapped module
(orekit):

build/_orekit/__wrap__.cpp:86504:89: error: no member named
  'HarmonicOscillator$Parametric$$Type' in namespace


I believe I fixed this bug in rev 1557613, header files for classes for
inherited fixed parameters were not included as required.

Andi..

  'org::apache::commons::math3::analysis::function'
...=
&::org::apache::commons::math3::analysis::function::PY_TYPE(HarmonicOs...
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
/Users/petrus/anaconda/lib/python2.7/site-packages/JCC-2.18-py2.7-macosx-10.5-x86_64.egg/jcc/sources/macros.h:66:23:
note:
  expanded from macro 'PY_TYPE'
#define PY_TYPE(name) name##$$Type
                  ^
<scratch space>:63:1: note: expanded from here
HarmonicOscillator$Parametric$$Type
^
build/_orekit/__wrap__.cpp:217752:55: error: no member named
  'OrekitMessages$$Type' in namespace 'org::orekit::errors'
    self->parameters[0] = &::org::orekit::errors::PY_TYPE(OrekitMessages);
                           ~~~~~~~~~~~~~~~~~~~~~~~^
/Users/petrus/anaconda/lib/python2.7/site-packages/JCC-2.18-py2.7-macosx-10.5-x86_64.egg/jcc/sources/macros.h:66:23:
note:
  expanded from macro 'PY_TYPE'
#define PY_TYPE(name) name##$$Type
                  ^
<scratch space>:9:1: note: expanded from here
OrekitMessages$$Type
^
2 errors generated.
error: command 'gcc' failed with exit status 1



The OrekitMessages is defined as a "public enum OrekitMessages implements
Localizable { ..."
and the "public class HarmonicOscillator implements
UnivariateDifferentiable, DifferentiableUnivariateFunction".

Maybe this says you something, I will try to test more and run my test case
in .

Best Regards and again, many thanks again!
/petrus





On 11 Jan 2014, at 14:23 , Andi Vajda <vajda@apache.org> wrote:


Hi Petrus,

On Mon, 30 Dec 2013, Petrus Hyvönen wrote:

Andi, great to hear that you could reproduce it. I'm very thankful if you
could have a look at it, I've been struggling to understand how the
machinery behind this works, but far from it still..


I think I fixed the problem and added an improvement to generics support
along the way. Please check out rev 1557423 from jcc's trunk, rebuild your
module with it and let me know how it's working for you.

The bug had to do with SimpleClass2's wrapper class missing its parameters
slot which it should get from its superclass, SimpleClass. When calling a
method on SimpleClass from SimpleClass2, the missing parameters in the
wrapper's struct would cause memory to get trashed.

In addition to fixing the bug, I also improved support for fixed class
parameters. Now, if you change SimpleClass's return_null() method to return
new Integer(12) instead, when calling it on SimpleClass2, 12 is returned
instead of <Object: 12>.

Andi..

public class SimpleClass<T> { public SimpleClass() {
System.out.println("Created SimpleClass"); }
public T return_null() {
   return (T) new Integer(12);
}

}



Best Regards
/Petrus



On Mon, Dec 30, 2013 at 10:36 AM, Andi Vajda <vajda@apache.org> wrote:


On Sun, 29 Dec 2013, Petrus Hyvönen wrote:

I have distilled the library that I have some trouble with and I think I

have an example that is failing due to same problem I think. I am not good
in java, but have tried to follow the logic from the library I'm wrapping.
The function of the example does not make sense in itself.

public class SimpleClass<T> {
public SimpleClass() {
System.out.println("Created SimpleClass");
}
public T return_null() {
  return  null;
}

}



public class SimpleClass2 extends SimpleClass<Integer>{

public SimpleClass2(){}
public void testInJava(){
System.out.println(this.return_null());
}
}


It seems to me that there is some problem with methods inherited that
returns a generic type, failing in wrapType when this is to be wrapped.

The python script that fails:

a= SimpleClass()
print a.return_null()

b = SimpleClass2()
b.testInJava()

print b.return_null()  #Fails in wrapType

I don't know if the return null is a bad thing to do in java, but the
error
seems very similar to what I experience in the larger library. I have a
skeleton of that this is slightly larger, but not returning null, but
trying to keep the lenght of example low :)

Any comments highly appriciated :)


I've been able to reproduce the problem.
Thank you for providing an isolated test case !

Andi..



best Regards
/Petrus






On Fri, Dec 27, 2013 at 6:25 PM, Andi Vajda <vajda@apache.org> wrote:


On Dec 27, 2013, at 17:36, Petrus Hyvönen <petrus.hyvonen@gmail.com>
wrote:


Dear Andi,

I am working on debugging the failure and try to understand a bit how
JCC
works internally. I haven't gone very far but in case you have some
pointers from these early debugging sessions I would be very thankful. I
know it's complex, and I should try to make some smaller test cases, but

I

don't really have a grasp yet where the problem might be.

Writing this might help me also to get some structure in my thinking :)



The main crash seems to be in the last line, wrapType(), of
__wrap__.cpp:

static PyObject

*t_AbstractReconfigurableDetector_withHandler(t_

AbstractReconfigurableDetector

*self, PyObject *arg)
  {
    ::org::orekit::propagation::events::handlers::EventHandler
a0((jobject) NULL);
    PyTypeObject **p0;
    ::org::orekit::propagation::events::EventDetector
result((jobject) NULL);

    if (!parseArg(arg, "K",

::org::orekit::propagation::events::handlers::

EventHandler::initializeClass,

&a0, &p0,

::org::orekit::propagation::events::handlers::t_

EventHandler::parameters_))

    {
      OBJ_CALL(result = self->object.withHandler(a0));
      return self->parameters[0] != NULL ?
wrapType(self->parameters[0], result.this$) :
::org::orekit::propagation::events::t_EventDetector::wrap_
Object(result);
    }

The parameters[0] does not seem to be null, but neither is it a valid
object, in my debugger it says 0xbaadf00d {ob_refcnt=??? ob_type=???
ob_size=??? ...}     _typeobject *, wrapType is called and when trying
to
access the wrapfn it crashes.

The main python lines are:
tmp1 = ElevationDetector(sta1Frame)                    #

ElevationDetector

is a java public class ElevationDetector extends
AbstractReconfigurableDetector<ElevationDetector>
hand = ContinueOnEvent().of_(ElevationDetector)   # a
java ContinueOnEvent<ElevationDetector> object
elDetector = tmp1.withHandler(hand)     #Crash. withHandler is a method
that is inherited from AbstractReconfigurableDetector to

ElevationDetector


This crashes when interactively entered on the python prompt (or in
other
interactive consoles), but seems to work if executed directly without
interactivity. This difference makes me think that it might be something
with garbage collection, but don't know.

Any comments appriciated, I know this is likely very difficult to
comment
on as it's not very encapsulated.


Right, so unless you can isolate this into something I can reproduce, I'm
afraid there isn't much I can comment.
It is quite likely that by the time you have that reproducible test case
ready, you also have the solution to the problem. Or I might be able to
help then...

Andi..


Regards
/Petrus






On Sun, Dec 15, 2013 at 10:36 AM, Andi Vajda <vajda@apache.org> wrote:


On Dec 15, 2013, at 5:43, Petrus Hyvönen <petrus.hyvonen@gmail.com>

wrote:

Hi Andi,

I see your point and have now kept in the pure python domain.

If I run my script from the shell by "python script.py" it does not

crash. However if I execute it line-by-line in python it crashes (or in
other tools such as ipython notebook).

All classes used are non-wrapped java classes, but I get the same

effect


with classes made for python subclassing.


I am getting this on both MacOSX 64-bit python and Windows 7 32-bit

python.



elDetector =

elDetector.withHandler(ContinueOnEvent().of_(ElevationDetector))


#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x000000010005da1a, pid=3318, tid=1287
#
# JRE version: Java(TM) SE Runtime Environment (7.0_45-b18) (build

1.7.0_45-b18)

# Java VM: Java HotSpot(TM) 64-Bit Server VM (24.45-b08 mixed mode

bsd-amd64 compressed oops)

# Problematic frame:
# C  [libpython2.7.dylib+0x5aa1a]  PyObject_GetAttr+0x1a
#


from the stack it seems like there is somthing happening in "wrapType"
Stack: [0x00007fff5fb80000,0x00007fff5fc00000],
sp=0x00007fff5fbff470,

free space=509k

Native frames: (J=compiled Java code, j=interpreted, Vv=VM code,

C=native code)

C  [libpython2.7.dylib+0x5aa1a]  PyObject_GetAttr+0x1a
C  [_orekit.so+0xa80878]  wrapType(_typeobject*, _jobject*
const&)+0x58
C  [_orekit.so+0x554400]


org::orekit::propagation::events::t_AbstractReconfigurableDetector

_withHandler(org::orekit::propagation::events::t_
AbstractReconfigurableDetector*,

_object*)+0x1c0



First, is the generic class assignment correct as if to write "new

ContinueOnEvent<ElevationDetector>()" in java? And is it ok to use

regular


java objects/types?



Any other comments to move forward highly appriciated.. Is it somehow

possible to get more log what is going wrong?

You could compile the whole thing for debugging, by adding --debug
after
'build' in the jcc invocation and run it with gdb.

If you can isolate a reproducible crash into a small test case, I can

also


take a look at it.


Andi..


WIth best regards
/Petrus



On 15 Dec 2013, at 2:40 , Andi Vajda <vajda@apache.org> wrote:


On Dec 14, 2013, at 19:14, Petrus Hyvönen <petrus.hyvonen@gmail.com
wrote:



Hi,

I'm having a problem with I think might be related to generic types,

but not sure at all.



I'm wrapping a orbit calculation library, which has been working
well

but in latest version is using generic types and I'm getting some

problems.


The script works when executed in plain python, but fails in ipython

notebook on this last line when executed as a couple of cells.


What is an 'ipython notebook' ?

Andi.,


The section with problem in my script is:
elDetector =

ElevationDetector(sta1Frame).withConstantElevation(math.

radians(5.0))

elDetector =

elDetector.withHandler(ContinueOnEvent().of_(ElevationDetector))



In Java it would typically look something like:

ElevationDetector detector = new ElevationDetector(topo)
                                   .withConstantElevation(x)
                                   .withHandler(new

ContinueOnEvent<ElevationDetector>());



It produces correct results in plain python, but crashes the kernel

in


ipython if executed as cells, and in exection from spyder I get an error

message:


" elDetector =

elDetector.withHandler(ContinueOnEvent().of_(ElevationDetector))


AttributeError: 'str' object has no attribute 'wrapfn_' "


As I have been using this setup stabely with lots of other functions

it feels like there is something with the generic type line, but I

don't
really know how to get any further? I'm confused by that the pauses in

the


execution could seem to affect the result.



Any comments highly appriciated...

Best Regards
/Petrus



--
_____________________________________________
Petrus Hyvönen, Uppsala, Sweden
Mobile Phone/SMS:+46 73 803 19 00



--
_____________________________________________
Petrus Hyvönen, Uppsala, Sweden
Mobile Phone/SMS:+46 73 803 19 00



-- 
_____________________________________________
Petrus Hyvönen, Uppsala, Sweden
Mobile Phone/SMS:+46 73 803 19 00

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message