PyXB is developed using git. You can check out the development branch with:
git clone git://pyxb.git.sourceforge.net/gitroot/pyxb/pyxb
Public development is mostly done on the
next branch, which is the
default for checkouts from SourceForge. The
master branch contains
material integrated into the release, and follows
next. Tags for each
release are in the format
Bug fixes with unit tests are pushed to the
next branch as soon as they
are fixed. Users whose reported issues have been fixed are encouraged to
use the development branch until the fix has been made available in a tagged
and packaged release.
The practices described herein are intended to be used, but evolved over time, and not all are followed in every situation.
PyXB follows generally uses a coding style consistent with those described in the Python coding standard in PEP 8. Specific exceptions are listed below.
The pet peeves of PyXB’s maintainer differ from the BDFL‘s when it comes to whitespace. Spaces are used around operators to reveal precedence assumptions, and sometimes around parenthesis to distinguish tuple value boundaries:
hypot2 = x*x + y*y pair_of_pairs = ( (a, b), (b, c) )
Definitions of methods and classes always use a space before the parenthesis; invocations never do:
def declaration (self, recurse=False): if recurse: self.declaration(False)
PyXB heavily uses a single leading underscore as an indication of protected/friend status, as suggested in PEP 8. Wherever possible, double leading underscores are used to hide class member fields, restricting access to them through protected or public accessor methods.
Class members that are specifically class (as opposed to instance) members begin with a capital letter.
An underscore is used to separate the descriptive class or function name from a suffix that indicates its use. Standard suffixes are:
mixin: A class that does not stand alone, but is a superclass of one or more classes
csc: A method that uses cooperative super calling
vb: An indication that the method is expected to be overridden in a subclass.
PyXB provides a standard exception hierarchy that extends the one built into Python.
Where an execution branch has been identified that requires behavior that has not yet been implemented, raise an IncompleteImplementationError. At the current stage of development, these should be extremely rare.
Where the system detects that a precondition is not satisfied, processing must stop. If the precondition failure is due to improper use of the PyXB internal or public API, a LogicError should be raised. If the precondition failure is due to invalid input from the schema, a SchemaValidationError should be raised.
If the precondition is inexplicably false, Python’s
may be used. Use of
assert should be rare, and only in places that are
guaranteed to be exercised during the course of testing.
The exception behavior of methods SHOULD be documented. Documentation of asserts is not required.
Use decorators (PEP 318) to mark class methods. Note that this restricts us to Python 2.4 or later. Sigh with disappointment and move on.
The term “attribute” has different meanings in Python and XML.
tag : Refers to the text that opens an XML element
instance : [as an adjective] Refers to a characteristic of an instance of a Python class.
class : [as an adjective] Refers to a characteristic of a Python class itself, shared among all instances.
field : Refers to a named attribute of a Python object (class). When the attribute holds a value, it is an “instance (class) variable” or “instance (class) field”. When it holds a reference to an invokable object, it is an “instance (class) method”.
Use of new-style classes¶
Too many things, such as clean hooking into the pickling system, require the use of new-style classes. Namespaces, schema components, and types (simple and complex) all use new-style classes.
For this to work properly, if you implement an
__init__ method, it must take
arbitrary args and keywords, invoke
and extract any arguments it needs from the keywords. If you do not need to
do anything in the init method, leave it out. See
this commentary for a detailed
description of the intricacies of
PyXB makes heavy use of multiple inheritance through mix-in classes. If there are constraints on where the mix-in must appear in the method resolution order (mro), note that clearly in the mix-in documentation.
Invoking Superclass Instances¶
Cooperative super calling is a
pattern where a class may inherit multiple implementations of the same
method, and want to call all of them. Normally this is done by invoking the
parent class implementation before or after the subclass implementation. In
non-trivial inheritance hierarchies (as result from using many mix-ins),
it’s not obvious who the next parent to call is, so the Python
function is used. However, at some point a class will be reached that has
no more definitions for the called method, and attempting to invoke one
would produce an
Use the following idiom to conditionally invoke superclass methods when you are not sure the superclass has such a method.
def method_csc (self, *args, **kw): super_fn = getattr(super(ThisClass, self), 'method_csc', lambda *a,**kw: self) return super_fn(*args, **kw)
Note the use of the
_csc suffix to highlight that this method uses
cooperative super calling.
Class Variables At Multiple Levels¶
There are several cases where we want to store information in a class, but allow subclasses (not instances) to override it. An example is in the simpleTypeDefinition hierarchy where each class maintains a set of ConstrainingFacet instances that are available for constraining values of the class. In many cases, a subclass will not change the set of facets that affect instances, so we want to be able to inherit the parent class map; but in other cases we may need to add constraints that only affect the new class and its descendents.
This sort of thing is supported by implementing a private class method in
the base class which combines the dynamically-determined actual class name
with a constant identifier and uses
setattr to access
the class-specific value.
Type declarations may extend a type of the same name in a different namespace. If they do so, their binding classes will likely have the same name, but in different modules, while also inheriting (in exactly the situation where we want different values). For this reason, the constructed attribute name should also incorporate the module or namespace, something normally not done with the double-underscore feature of Python.