Using ILU With Python

Using ILU with Python

Introduction

This tutorial will show how to use the ILU system with the programming language Python, both as a way of developing software libraries, and as a way of building distributed systems.

In an extended example, we'll build an ILU module that implements a simple four-function calculator, capable of addition, subtraction, multiplication, and division. It will signal an error if the user attempts to divide by zero. The example demonstrates how to specify the interface for the module; how to implement the module in Python; how to use that implementation as a simple library; how to provide the module as a remote service; how to write a client of that remote service; and how to use subtyping to extend an object type and provide different versions of a module. We'll also demonstrate how to use OMG IDL with ILU, and discuss the notion of network garbage collection.

Each of the programs and files referenced in this tutorial is available as a complete program in a separate appendix to this document; parts of programs are quoted in the text of the tutorial.

Specifying the Interface

Our first task is to specify more exactly what it is we're trying to provide. A typical four-function calculator lets a user enter a value, then press an operation key, either +, -, /, or *, then enter another number, then press = to actually have the operation happen. There's usually a CLEAR button to press to reset the state of the calculator. We want to provide something like that.

We'll recast this a bit more formally as the interface of our module; that is, the way the module will appear to clients of its functionality. The interface typically describes a number of function calls which can be made into the module, listing their arguments and return types, and describing their effects. ILU uses object-oriented interfaces, in which the functions in the interface are grouped into sets, each of which applies to an object type. These functions are called methods.

For example, we can think of the calculator as an object type, with several methods: Add, Subtract, Multiply, Divide, Clear, etc. ILU provides a standard notation to write this down with, called ISL (which stands for "Interface Specification Language"). ISL is a declarative language which can be processed by computer programs. It allows you to define object types (with methods), other non-object types, exceptions, and constants.

The interface for our calculator would be written in ISL as:

INTERFACE Tutorial;

EXCEPTION DivideByZero;

TYPE Calculator = OBJECT
  METHODS
    SetValue (v : REAL),
    GetValue () : REAL,
    Add (v : REAL),
    Subtract (v : REAL),
    Multiply (v : REAL),
    Divide (v : REAL) RAISES DivideByZero END
  END;

This defines an interface Tutorial, an exception DivideByZero, and an object type Calculator. Let's consider these one by one.

The interface, Tutorial, is a way of grouping a number of type and exception definitions. This is important to prevent collisions between names defined by one group and names defined by another group. For example, suppose two different people had defined two different object types, with different methods, but both called Calculator! It would be impossible to tell which calculator was meant. By defining the Calculator object type within the scope of the Tutorial interface, this confusion can be avoided.

The exception, DivideByZero, is a formal name for a particular kind of error, division by zero. Exceptions in ILU can specify an exception-value type, as well, which means that real errors of that kind have a value of the exception-value type associated with them. This allows the error to contain useful information about why it might have come about. However, DivideByZero is a simple exception, and has no exception-value type defined. We should note that the full name of this exception is Tutorial.DivideByZero, but for this tutorial we'll simply call our exceptions and types by their short name.

The object type, Calculator (again, really Tutorial.Calculator), is a set of six methods. Two of those methods, SetValue and GetValue, allow us to enter a number into the calculator object, and "read" the number. Note that SetValue takes a single argument, v, of type REAL. REAL is a built-in ISL type, denoting a 64-bit floating point number. Built-in ISL types are things like INTEGER (32-bit signed integer), BYTE (8-bit unsigned byte), and CHARACTER (16-bit Unicode character). Other more complicated types are built up from these simple types using ISL type constructors, such as SEQUENCE OF, RECORD, or ARRAY OF.

Note also that SetValue does not return a value, and neither do Add, Subtract, Multiply, or Divide. Rather, when you want to see what the current value of the calculator is, you must call GetValue, a method which has no arguments, but which returns a REAL value, which is the value of the calculator object. This is an arbitrary decision on our part; we could have written the interface differently, say as

TYPE NotOurCalculator = OBJECT
  METHODS
    SetValue () : REAL,
    Add (v : REAL) : REAL,
    Subtract (v : REAL) : REAL,
    Multiply (v : REAL) : REAL,
    Divide (v : REAL) : REAL RAISES DivideByZero END
  END;

-- but we didn't.

Our list of methods on Calculator is bracketed by the two keywords METHODS and END, and the elements are separated from each other by commas. This is pretty standard in ISL: elements of a list are separated by commas; the keyword END is used when an explicit list-end marker is needed (but not when it's not necessary, as in the list of arguments to a method); the list often begins with some keyword, like METHODS. The raises clause (the list of exceptions which a method might raise) of the method Divide provides another example of a list, this time with only one member, introduced by the keyword RAISES.

Another standard feature of ISL is separating a name, like v, from a type, like REAL, with a colon character. For example, constants are defined with syntax like

CONSTANT Zero : INTEGER = 0;

Definitions, of interface, types, constants, and exceptions, are terminated with a semicolon.

We should expand our interface a bit by adding more documentation on what our methods actually do. We can do this with the docstring feature of ISL, which allows the user to add arbitrary text to object type definitions and method definitions. Using this, we can write

INTERFACE Tutorial;

EXCEPTION DivideByZero
  "this error is signalled if the client of the Calculator calls
the Divide method with a value of 0";

TYPE Calculator = OBJECT
  COLLECTIBLE
  DOCUMENTATION "4-function calculator"
  METHODS
    SetValue (v : REAL) "Set the value of the calculator to `v'",
    GetValue () : REAL  "Return the value of the calculator",
    Add (v : REAL)      "Adds `v' to the calculator's value",
    Subtract (v : REAL) "Subtracts `v' from the calculator's value",
    Multiply (v : REAL) "Multiplies the calculator's value by `v'",
    Divide (v : REAL) RAISES DivideByZero END
      "Divides the calculator's value by `v'"
  END;

Note that we can use the DOCUMENTATION keyword on object types to add documentation about the object type, and can simply add documentation strings to the end of exception and method definitions. These docstrings are passed on to the Python docstring system, so that they are available at runtime from Python. Documentation strings cannot currently be used for non-object types.

ILU provides a program, islscan, which can be used to check the syntax of an ISL specification. islscan parses the specification and summarizes it to standard output:

% islscan Tutorial.isl
Interface "Tutorial", imports "ilu"
  {defined on line 1
   of file /tmp/tutorial/Tutorial.isl (Fri Jan 27 09:41:12 1995)}

Types:
  real                       {<built-in>, referenced on 10 11 12 13 14 15}

Classes:
  Calculator                 {defined on line 17}
    methods:
      SetValue (v : real);                          {defined 10, id 1}
        "Set the value of the calculator to `v'"
      GetValue () : real;                           {defined 11, id 2}
        "Return the value of the calculator"
      Add (v : real);                               {defined 12, id 3}
        "Adds `v' to the calculator's value"
      Subtract (v : real);                          {defined 13, id 4}
        "Subtracts `v' from the calculator's value"
      Multiply (v : real);                          {defined 14, id 5}
        "Multiplies the calculator's value by `v'"
      Divide (v : real) {DivideByZero};             {defined 16, id 6}
        "Divides the calculator's value by `v'"
    documentation:
      "4-function calculator"
    unique id:  ilu:cigqcW09P1FF98gYVOhf5XxGf15

Exceptions:
  DivideByZero               {defined on line 5, refs 15}
%

islscan simply lists the types defined in the interface, separating out object types (which it calls "classes"), the exceptions, and the constants. Note that for the Calculator object type, it also lists something called its unique id. This is a 160-bit number (expressed in base 64) that ILU assigns automatically to every type, as a way of distinguishing them. While it might interesting to know that it exists (:-), the ILU user never has know what it is; islscan supplies it for the convenience of the ILU implementors, who sometimes do have to know it.

Implementing the True Module

After we've defined an interface, we then need to supply an implementation of our module. Implementations can be done in any language supported by ILU. Which language you choose often depends on what sort of operations have to be performed in implementing the specific functions of the module. Different languages have specific advantages and disadvantages in different areas. Another consideration is whether you wish to use the implementation mainly as a library, in which case it should probably be done in the same language as the rest of your applications, or mainly as a remote service, in which case the specific implementation language is less important.

We'll demonstrate an implementation of the Calculator object type in Python, which is one of the most capable of all the ILU-supported languages. This is just a matter of defining a Python class, corresponding to the Tutorial.Calculator type. Before we do that, though, we'll explain how the names and signatures of the Python functions are arrived at.

What the Interface Looks Like in Python

For every programming language supported by ILU, there is a standard mapping defined from ISL to that programming language. This mapping defines what ISL type names, exception names, method names, and so on look like in that programming language.

The mapping for Python is straightforward. For type names, such as Tutorial.Calculator, the Python name of the ISL type Interface.Name is Interface.Name, with any hyphens replaced by underscores. That is, the name of the interface in ISL becomes the name of the module in Python. So the name of our Calculator type in Python would be Tutorial.Calculator, which is really the name of a Python class.

The Python mapping for a method name such as SetValue is the method name, with any hyphens replaced by underscores. The return type of this Python method is whatever is specified in the ISL specification for the method, or None if no type is specified. The arguments for the Python method are the same as specified in the ISL; their types are the Python types corresponding to the ISL types, except that one extra argument is added to the beginning of each Python version of an ISL method; it is an instance of the object type on which the method is defined. An instance is simply a value of that type. Thus the Python method corresponding to our ISL SetValue would have the prototype signature

   def SetValue (self, v):

Similarly, the signatures for the other methods, in Python, are

   def GetValue (self):

   def Add (self, v):

   def Subtract (self, v):

   def Multiply (self, v):

   def Divide (self, v):

Note that even though the Divide method can raise an exception, the signature looks like those of the other methods. This is because the normal Python exception signalling mechanism is used to signal exceptions back to the caller. The mapping of exception names is similar to the mapping used for types. So the exception Tutorial.DivideByZero would also have the name Tutorial.DivideByZero, in Python.

One way to see what all the Python names for an interface look like is to run the program python-stubber. This program reads an ISL file, and generates the necessary Python code to support that interface in Python. One of the files generated is `Interface.py', which contains the definitions of all the Python types for that interface.

% python-stubber Tutorial.isl
client stubs for interface "Tutorial" to Tutorial.py ...
server stubs for interface "Tutorial" to Tutorial__skel.py ...
%

Building the Implementation

To provide an implementation of our interface, we subclass the generated Python class for our Calculator class:

# CalculatorImpl.py

import Tutorial, Tutorial__skel

class Calculator (Tutorial__skel.Calculator):

        def __init__ (self):
                self.the_value = 0.0

        def SetValue (self, v):
                self.the_value = v

        def GetValue (self):
                return self.the_value

        def Add (self, v):
                self.the_value = self.the_value + v

        def Subtract (self, v):
                self.the_value = self.the_value - v

        def Multiply (self, v):
                self.the_value = self.the_value * v

        def Divide (self, v):
                try:
                        self.the_value = self.the_value / v
                except ZeroDivisionError:
                        raise Tutorial.DivideByZero

Each instance of a CalculatorImpl.Calculator object inherits from Tutorial__skel.Calculator, which in turn inherits from Tutorial.Calculator. Each has an instance variable called the_value, which maintains a running total of the `accumulator' for that instance. We can create an instance of a Tutorial.Calculator object by simply calling CalculatorImpl.Calculator().

So, a very simple program to use the Tutorial module might be the following:

# simple1.py, a simple program that demonstrates the use of the
#  Tutorial true module as a library.
#
# run this with the command "python simple1.py NUMBER [NUMBER...]"
#

import Tutorial, CalculatorImpl, string, sys

# A simple program:
#  1)  make an instance of Tutorial.Calculator
#  2)  add all the arguments by invoking the Add method
#  3)  print the resultant value.

def main (argv):

        c = CalculatorImpl.Calculator()
        if not c:
                error("Couldn't create calculator")

        # clear the calculator before using it

        c.SetValue (0.0)

        # now loop over the arguments, adding each in turn */

        for arg in argv[1:]:
                v = string.atof(arg)
                c.Add (v)

        # and print the result

        print "the sum is", c.GetValue()
        sys.exit(0)

main(sys.argv)

This program would be compiled and run as follows:

% python simple1.py 34.9 45.23111 12
the sum is 92.13111
%

This is a completely self-contained use of the Tutorial implementation; when a method is called, it is the true method that is invoked. The use of ILU in this program adds some overhead in terms of included code, but has almost the same performance as a version of this program that does not use ILU.

Checking for Exceptions

Suppose, instead of the Add method, we'd called the Divide method. In that case, we might have had to handle a DivideByZero exception; that is, notice the exception and do something sensible. We do this by establishing a handler for the exception:

        ...
        # now loop over the arguments, Dividing by each in turn */

        try:
                for arg in argv[2:]:
                        v = string.atof(arg)
                        c.Divide (v)
        except:
                print 'exception signalled:  ' + str(sys.exc_type)
                sys.exit(1)
        ...

And here's an example of what we get when it runs:

% python simple2.py 12345 6 7 8 9
the sum is 4.08234126984
% python simple2.py 12345 6 0 8 9
exception signalled:  Tutorial: DivideByZero

Actually, every method may return an exception, as there are a number of standard system exceptions which may be signalled even by methods which have no declared exceptions. So we should check every method to see if it succeeded, even simple ones like GetValue.

Providing the True Module as a Network Service

Now let's see what's involved in providing the calculator functionality as a network service. Basically, there are three things to look at:

• providing a "factory" to build calculator objects;
• publishing the name of the factory; and
• writing a server program.

Using Factories to Build Objects

When one program uses code from another address space, it has to get its hands on an instance of an ILU object, to be able to call methods. In our library application, we simply made a call into the true module, to create an instance of the calculator object. In the networked world, we need to do the same kind of thing, but this time the call into the true module has to be a method on an object type. In short, we need to have some object type which exports a method something like

  CreateCalculator () : Calculator

There are several ways to provide this. The standard way of doing it is to add an object type to our Tutorial interface, which contains this method. This kind of object type is sometimes called a factory, because it exists only in order to build instances of other object types. We'll add the following type definition to our `Tutorial.isl':

TYPE Factory = OBJECT
  METHODS
    CreateCalculator () : Calculator
  END;

Then we need to provide an implementation of the Factory object type, just as we did with the Calculator type:

import Tutorial, Tutorial__skel, CalculatorImpl

class Factory (Tutorial__skel.Factory):

        # have the __init__ method take handle and server args
        # so that we can control which ILU kernel server is used,
        # and what the instance handle of the Factory object on
        # that server is.  This allows us to control the object ID
        # of the new Factory instance.

        def __init__(self, handle=None, server=None):
                self.IluInstHandle = handle
                self.IluServer = server
                
        def CreateCalculator (self):
                return (CalculatorImpl.Calculator())

Now, to provide other programs a way of creating calculator objects, we'll just create just one instance of Tutorial.Factory, and let programs call the CreateCalculator method on that at will, to obtain new calculator objects.

Publishing a Well-Known Instance

The question then arises, how does a program that wants to use the Factory object get its hands on that one well-known instance? The answer is to use the simple binding system built into ILU. Simple binding allows a program acting as a "server" to publish the location of a well-known object, and allows programs acting as "clients" of that server to look up the location, given the object's name.

The name of an ILU object instance has two parts, which are the instance handle of the object, and the name of its kernel server, called the server ID. (The kernel server is a data structure maintained by the ILU kernel which takes care of all communication between different address spaces.) These two combined must form a universally unique ID for the object. Usually you can simply let the ILU system choose names for your objects automatically, in which case it takes care to choose names which will never conflict with names in use by others. However, for objects which we wish to publish, we need to specify what the name of an object will be, so that users of the well-known object can find it.

When working with the Python programming language, this act of explicitly specifying an object name is divided into two steps. First, we create a kernel server with a specified server ID. Secondly, we create an instance of an object on this new server, with a specified instance handle. Together, the server ID and the instance handle form the name of the instance.

For instance, we might use a server ID of Tutorial.domain, where domain is your Internet domain (typically something like department.company.com, or department.univerity.edu). This serves to distinguish your server from other servers on the net. Then we can use a simple instance handle, like theFactory. The name, or object ID, of this object would then be theFactory@Tutorial.domain, where domain would vary from place to place. Note that this implies that only one instance of this object is going to exist in the whole domain. If you have many people using different versions of this object in your domain, you should introduce more qualifiers in the server ID so that your kernel server can be distinguished from that run by others.

The Server Program

Given this information, we can now write a complete program that will serve as a provider of calculator objects to other programs. It will create a single Factory instance with a well-known name, publish that instance, then hang out servicing methods invoked on its objects. Here's what it looks like:

# server.py -- a program that runs a Tutorial.Calculator server
#

import ilu, FactoryImpl, sys

def main(argv):

        if (len(argv) < 2):
                print "Usage:  python server.py SERVER-ID"
                sys.exit(1)

        # Create a kernel server with appropriate server ID, which
        #  is passed in as the first argument

        theServer = ilu.CreateServer (argv[1])

        # Now create an instance of a Factory object on that server,
        #  with the instance handle "theFactory"

        theFactory = FactoryImpl.Factory ("theFactory", theServer)

        # Now make the Factory object "well-known" by publishing it.

        theFactory.IluPublish()

        # Now we print the string binding handle (the object's name plus
        # its location) of the new instance.

        print "Factory instance published."
        print "Its SBH is '" + theFactory.IluSBH() + "'."

        handle = ilu.CreateLoopHandle()
        ilu.RunMainLoop (handle)


main(sys.argv)

When we run this program, we'll see something like:

% python server.py Tutorial.dept.company.com &
Factory instance published.
Its SBH is 'theFactory@Tutorial.dept.company.com@somegibberish'.
% 

This indicates that the object known as theFactory@Tutorial.dept.company.com is being exported in a particular way, which is encoded in the somegibberish part of the string binding handle. Your specific numbers will vary, but it should look similar.