|I l@ve RuBoard|
5.2 Overriding a Built-In Method
Credit: Dave Haynes
In most cases, overriding special methods of built-in objects when you inherit from those objects (or wrap them with automatic delegation, which is not technically an override) poses no special challenge. When inheriting in Python 2.2, you can call the special method of the superclass with the usual unbound-method syntax. When wrapping, use the syntax that is specific to the operation, such as self.data[someindex] for indexing.
Slicing is harder, because while slicing should go through the same special method _ _getitem_ _ as indexing (since Python 2.0), lists and tuples still implement an older approach: the more limited special method _ _getslice_ _ (and similarly for _ _setitem_ _ versus _ _setslice_ _ and _ _delitem_ _ versus _ _delslice_ _). So, you must provide a remedy, normally with a try/except:
class SliceTester: def _ _init_ _(self): self.data = ['zero', 'one', 'two', 'three', 'four'] def _ _getitem_ _(self, indexOrSlice): try: return self.data[indexOrSlice] except TypeError: return self.data[indexOrSlice.start:indexOrSlice.stop]
When a user-defined class wraps (or, in Python 2.2, inherits from) a list or tuple, it often needs to define the _ _set*_ _ and _ _get*_ _ special methods and delegate part or all of their operation to the wrapped (or inherited) built-in object to provide the correct access to the data.
The documentation for Python 2.0 and later deprecates the use of _ _getslice_ _ and _ _setslice_ _. Instead, it suggests providing suitably extended versions of _ _getitem_ _ and _ _setitem_ _. This is a truly excellent idea because it enables the use of the extended-form slicing approaches (including step, ellipsis, and so on) that Numeric Python has made so deservedly popular among its regular users. Unfortunately, if you try to pass a slice object to the item-oriented special methods of a list or tuple object, you get a TypeError; the underlying C API still insists on receiving integer parameters, not slice objects in all their glory, whatever the documentation may say.
Fortunately, working around this problem isn't as dramatic as all that. You just need to trap the TypeError you get from trying to index an old-fashioned sequence with a slice, and remedy it suitably. Here's the typical self-test code that you can append to the recipe's module and execute when it is run as a main script:
if _ _name_ _ == "_ _main_ _": theSlice = SliceTester( ) a = theSlice b = theSlice[:3] print a print b
In the recipe's SliceTester example class, the remedy is pretty minimal; it's just an attempt to use start and stop attributes of the noninteger index (presumably an instance of the slice built-in type). You may want to do a lot more (implement step, ellipsis, and so on).
Note that this recipe doesn't cover all of the cases in which slices can be used. There is a third argument to the slice operator that defines the step, or stride, of the slicing. For example, if data is a Numeric Python array (the only widely used software that supports slicing in all its glory), data[0:101:10] returns the sequence data, data, data梪p to data. Similarly, data[::-1] returns a sequence containing the contents of data reversed. The third argument to the slice operator is stored in the step attribute of slice objects and is set to None if a step isn't specified (as in list[start:end]). Given this, it shouldn't be a surprise that the recipe shown earlier will not magically add support for steps to objects that don't support new-style slices.
The point of this recipe is that you must be aware of these limitations and take precautionary measures. Also, don't type-test for an index of type slice. If normal indexing refuses the index, you are better off catching the TypeError in an except clause and entering another try/except in which you try to use the index as the slice you now expect it to be. This lets client code pass you objects that are polymorphic to slice objects.
5.2.4 See Also
The section of the Language Reference on slicing; the description of the slice built-in function in the Library Reference.
|I l@ve RuBoard|