Differences between revisions 7 and 8
Revision 7 as of 2012-02-11 18:40:11
Size: 8078
Editor: JeffAllen
Comment: Java line wrap
Revision 8 as of 2013-11-26 21:11:57
Size: 9293
Editor: JeffAllen
Comment:
Deletions are marked like this. Additions are marked like this.
Line 9: Line 9:
{{{gderived.py}}} and {{{gexposed.py}}} are a pair of Python scripts written support the wrapping of classes
written in Java as Python classes. They are in the {{{src/templates folder}}} and they have to be run
with that as the current directory.

The function of {{{gderived.py}}} is to read the source of the Java class and create a another Java
source file in the same package, that extends the original, with highly-stylised content
according to a brief specification. This code is connected with subclassing.

{{{gexposed.py}}} has been superseded by the exposer
({{{org.python.expose.generate.Exposer}}}) and the corresponding Ant task, but {{{gderived.py}}} remains
actively used in the creation of Jython.
{{{gexposed.py}}} is also active code, but only because {{{gderived.py}}} imports it. However, if you use the
new exposer, even if you prohibit subclassing with {{{@ExposedType(isBaseType=false)}}},
it will generate a reference to the sort of class {{{gderived.py}}} creates. The modern exposer is described in the article PythonTypesInJava.

At the time of starting these notes, there is no authoritative user-guide to
{{{gderived.py}}} and what it achieves. These notes stem from use of the tool and a certain amount of
Many of the Java classes that implement Python types have a counterpart class with the same name
but with "Derived" appended. For example `PyString` is paired with `PyStringDerived`, `PyType` with
`PyTypeDerived`, and so on. The `*Derived` classes are each a sub-class of their corresponding
principal class. They come into play when you create a sub-class (in Python) and override (in Python)
one or more methods whose base definition is exposed from the Java implementation. They ensure that
this overriding (Python) method is the version invoked, even when the call is from Java.

There are two parts to this remarkable feature. One is in the implementation of the principal
class itself, where the static exposed `new` method checks to see whether the actual (Python) type
of the object being created is exactly the type that the principal class implements.
If it is, then a new instance of the (Java) class is returned. If it is not exactly that class,
an instance of the counterpart `*Derived` class is returned.

The second part of the feature is in the `*Derived` class. There, each exposed method may be
overridden in a stylised way: it will check for the existence of a (Python) method redefining
(the exposed name of) that method. If it fails to find one, it calls the version in the principal
class (using the `super` keyword in Java). If it finds a Python re-definition, it invokes that
using `PyObject.__call__()`.

The `*Derived` counterpart of each principal class is generated using the script {{{gderived.py}}}
and a brief specification.
The script is in the {{{src/templates}}} directory, together with several modules it imports,
and it has to be run with that as the current directory.
The specification corresponding to each principal class is in the same directory.

One of the imported modules is {{{gexposed.py}}}. This used to have a function in its own right,
but it is superseded by the exposer ({{{org.python.expose.generate.Exposer}}}) and the
corresponding Ant task.
If you use the new exposer, even if you prohibit sub-classing with {{{@ExposedType(isBaseType=false)}}},
it will generate a reference to the sort of class {{{gderived.py}}} creates.
The modern exposer is described in the article PythonTypesInJava.

==== Author's note: ====
At the time of starting these notes, there is no user guide to
{{{gderived.py}}} and what it achieves. These notes stem from use of the tool and a certain amount of
Line 31: Line 49:

Generating the *Derived classes

These are notes on the use of gderived.py, a tool you need when implementing new types in Jython.

Background

Many of the Java classes that implement Python types have a counterpart class with the same name but with "Derived" appended. For example PyString is paired with PyStringDerived, PyType with PyTypeDerived, and so on. The *Derived classes are each a sub-class of their corresponding principal class. They come into play when you create a sub-class (in Python) and override (in Python) one or more methods whose base definition is exposed from the Java implementation. They ensure that this overriding (Python) method is the version invoked, even when the call is from Java.

There are two parts to this remarkable feature. One is in the implementation of the principal class itself, where the static exposed new method checks to see whether the actual (Python) type of the object being created is exactly the type that the principal class implements. If it is, then a new instance of the (Java) class is returned. If it is not exactly that class, an instance of the counterpart *Derived class is returned.

The second part of the feature is in the *Derived class. There, each exposed method may be overridden in a stylised way: it will check for the existence of a (Python) method redefining (the exposed name of) that method. If it fails to find one, it calls the version in the principal class (using the super keyword in Java). If it finds a Python re-definition, it invokes that using PyObject.__call__().

The *Derived counterpart of each principal class is generated using the script gderived.py and a brief specification. The script is in the src/templates directory, together with several modules it imports, and it has to be run with that as the current directory. The specification corresponding to each principal class is in the same directory.

One of the imported modules is gexposed.py. This used to have a function in its own right, but it is superseded by the exposer (org.python.expose.generate.Exposer) and the corresponding Ant task. If you use the new exposer, even if you prohibit sub-classing with @ExposedType(isBaseType=false), it will generate a reference to the sort of class gderived.py creates. The modern exposer is described in the article PythonTypesInJava.

Author's note:

At the time of starting these notes, there is no user guide to gderived.py and what it achieves. These notes stem from use of the tool and a certain amount of reverse-engineering. Please improve on them by correcting misunderstandings and omissions.

The work was done on a Windows 7 system, using Python 2.7 (without trying later versions, because of the vintage of the code). The choice of OS shows sometimes in the direction of slashes in pathnames, but that shouldn't confuse anyone. Although the motivation was to add a serious Python type (bytearray) to Jython, illustrations will be drawn from a facetiously-named type (piranha), with a Java implementation in src/org/python/ethel/the/frog/Piranha.java.

gderived.py as a Command

The 2-argument Forms

The most transparent form of the command is:

python [--lazy] gderived.py <derived-spec> <output-file>

When using gderived.py in that way one is working with three user files:

<derived-spec>, the specification for the contents of the derived Java class. By convention, this has the extension .derived, and the Python name of the type being defined. The files for Jython types are in the src/templates folder along with the scripts, but anywhere seems to work with this form of the command, so we'll use src/org/python/ethel/the/frog/piranha.derived (note lower case type name piranha).

<output-file>, the file in which the generated class will be written. This has to be in the Jython source tree under src/org/python. If your code is not there, gderived.py seems to run correctly, but will get the Java package statement wrong. You can supply any filename you like, but the class it writes will be named by adding "Derived" to the name of the input class. For our example the output file is src/org/python/ethel/the/frog/PiranhaDerived.java.

And last but not least, the class file that implements your type. The input file is identified from the directory of the output file and the class name given in the text of <derived-spec> (see below). This therefore also has to be in the Jython source tree under src/org/python. For our example the input file is src/org/python/ethel/the/frog/Piranha.java.

The --lazy option causes gderived.py only to generate the output file if the input file is newer.

The 1-argument and 0-argument Forms

A second form of the command is:

python gderived.py [--lazy] [<derived-spec>]

When using gderived.py in that way one is working with the same three user files as above, and a configuration file src/templates/mappings. The entries in that file look like this:

int.derived:org.python.core.PyIntegerDerived
object.derived:org.python.core.PyObjectDerived
random.derived:org.python.modules.random.PyRandomDerived
ast_Assert.derived:org.python.antlr.ast.AssertDerived

In effect, this file allows gderived.py to look up the second argument given the first, although this second argument is now given in dotted notation. In this form, the specification file <name>.derived has to be in src/templates and the input and output classes will be found relative to src.

Finally, the <derived-spec> argument is optional. In the zero-argument form, gderived.py will process all of the entries in src/templates/mappings. It is essentially this form, with the --lazy option, that implements the template Ant target in build.xml.

The Specification file <name>.derived

Available Directives

base_class

Define the name of the input class. Do not qualify the class name with the package: the script will work it out from the output file path, relative to src. E.g. base_class: Piranha

want_dict
Request creation of a dictionary in the derived class. If not specified, only a slots array is created.
require
define
ctr
incl
Include all the methods from this base class,
unary1
binary
ibinary
rest
no_toString

The generated class will not be given a custom toString() method. Use this when you already have a satisfactory one in the base.

Examples of Use

Minimal Case

Input Piranha.java

Here is an example of a type defined in Java for access as a built-in in Jython. For information on the annotations and structure see PythonTypesInJava.

   1 package org.python.ethel.the.frog;
   2 
   3 import org.python.core.PyObject;
   4 import org.python.core.PyString;
   5 import org.python.core.PyType;
   6 import org.python.expose.ExposedMethod;
   7 import org.python.expose.ExposedNew;
   8 import org.python.expose.ExposedType;
   9 
  10 @ExposedType(name="piranha")
  11 public class Piranha extends PyObject {
  12 
  13     public static final PyType TYPE = PyType.fromClass(Piranha.class);
  14     
  15     public Piranha() { this(TYPE); }
  16     
  17     public Piranha(PyType subType) { super(subType); }
  18     
  19     @ExposedNew
  20     final void newPiranha(PyObject[] args, String[] keywords) {}
  21     
  22     @ExposedMethod(names={"theOperation", "theOtherOperation"})
  23     public int operation(int payment) { return payment*2; }
  24 }

Specification piranha.derived

base_class: Piranha

Output PiranhaDerived.java

The command

python gderived.py ..\org\python\ethel\the\frog\piranha.derived  ..\org\python\ethel\the\frog\PiranhaDerived.java

issued with current directory src/templates produces

   1 /* Generated file, do not modify.  See jython/src/templates/gderived.py. */
   2 package org.python.ethel.the.frog;
   3 
   4 import java.io.Serializable;
   5 import org.python.core.*;
   6 
   7 public class PiranhaDerived extends Piranha implements Slotted {
   8 
   9     public PyObject getSlot(int index) {
  10         return slots[index];
  11     }
  12 
  13     public void setSlot(int index,PyObject value) {
  14         slots[index]=value;
  15     }
  16 
  17     private PyObject[]slots;
  18 
  19     public String toString() {
  20         PyType self_type=getType();
  21         PyObject impl=self_type.lookup("__repr__");
  22         if (impl!=null) {
  23             PyObject res=impl.__get__(this,self_type).__call__();
  24             if (!(res instanceof PyString))
  25                 throw Py.TypeError(
  26                     "__repr__ returned non-string (type "+
  27                     res.getType().fastGetName()+")");
  28             return((PyString)res).toString();
  29         }
  30         return super.toString();
  31     }
  32 
  33 }

Additional Features

Input Piranha.java

Specification piranha.derived

Output PiranhaDerived.java

GeneratedDerivedClasses (last edited 2017-08-26 21:16:41 by JeffAllen)