A workaround to a JPY issue with long values, fixes #1461

[jmao-denver]

In JPY, PyObject.getObjectValue() seems always to select Java Integer
for a Python int object which is not restricted by the number
of bits in Python 3

[jmao-denver]

DHE suffers from the same problem, added @cpwright as a reviewer to loop him in.

[cpwright]

Should we simply look further up the chain at JPy and fix that. I expect the problem relates to these lines in
JType_ConvertJavaToPythonObject.

} else if (JPy_IS_CLONG(pyArg) && (type == JPy_JObject || ((*jenv)->IsAssignableFrom(jenv, JPy_Integer_JClass, type->classRef)))) {
    return JType_CreateJavaIntegerObject(jenv, type, pyArg, objectRef);
} else if (JPy_IS_CLONG(pyArg) && (type == JPy_JObject || ((*jenv)->IsAssignableFrom(jenv, JPy_Long_JClass, type->classRef)))) {
    return JType_CreateJavaLongObject(jenv, type, pyArg, objectRef);

[devinrsmith]

Should we simply look further up the chain at JPy and fix that. I expect the problem relates to these lines in JType_ConvertJavaToPythonObject.

} else if (JPy_IS_CLONG(pyArg) && (type == JPy_JObject || ((*jenv)->IsAssignableFrom(jenv, JPy_Integer_JClass, type->classRef)))) {
    return JType_CreateJavaIntegerObject(jenv, type, pyArg, objectRef);
} else if (JPy_IS_CLONG(pyArg) && (type == JPy_JObject || ((*jenv)->IsAssignableFrom(jenv, JPy_Long_JClass, type->classRef)))) {
    return JType_CreateJavaLongObject(jenv, type, pyArg, objectRef);

I made this same comment here #1461 (comment). Ultimately, I think the implicit type conversion layer of python objects to to java objects needs to be examined, and we may want to provide the (java) caller with better hooks on how to handle the different cases.

These types of problems may also be better solved by pushing application logic further down into python instead of trying to be as general as we can w/ PythonScope. I've made some similar changes for FetchObject (landing soon) where script session changes are handled with .py logic instead of numerous JNI boundary crossings.

[jmao-denver]

@devinrsmith @cpwright

I think you meant JType_ConvertPythonToJavaObject (not JType_ConvertJavaToPythonObject).

The straightforward changes I could see are:

1. to switch the order of the two blocks (or simply remove the one check assignability of Integer, which amounts to the alternative Devin mentioned about, and with that, potentially double the memory used for a column unnecessarily

2. to examine the actual value and select the appropriate Java type, which is almost the same (hacky) as the solution in this PR but with one fewer JNI/Python boundary crossing

I am not strongly inclined towards any one of the 3 solutions, but this PR is still my favorite simply because it is higher level and less likely to break other things (?)

[devinrsmith]

I'm OK with this approach. Based on @rcaudy feedback thought, he may prefer the pyObject.isLong(), pyObject.getLongValue(), and seeing if it fits into int. It's an extra case, but it fits the pattern established here.

already done