Python private constructors
There is no such thing as Python private constructors.
However, as I worked on wrapping some C-based code with Cython, I ended up with the following pattern. Some (boring) technical issues do not allow me to just have the make_foo processing in the constructor.
cdef class Foo(object):
"""Documentation for class Foo.
Use `make_foo` to create such object.
"""
cdef long value
def __cinit__(self, value):
self.value = value
def __repr__(self):
return "Foo: #{}".format(self.get_id())
def get_id(self):
return foo_id(self.value)
def make_foo(idx):
"""Build a Foo object from an id.
>>> make_foo(12)
Foo: #12
"""
value = build_foo(idx)
return Foo(value)
So, I expect the user to only call make_foo and I would love that the constructor raises a ValueError when it is called by the user and processes silently when it is called by authorised functions in the module.
Here is a suggestion to reach this goal.
Because of its double-underscore prefix, __SECRET__ is not available from outside the module, and as a basic object(), it is not possible to imitate it and try to fool the constructor of Foo. However, the make_foo function being part of the module knows the __SECRET__ and can safely pass it to the constructor.
__SECRET__ = object()
cdef class Foo(object):
"""Documentation for class Foo.
The constructor raises a ValueError if called directly.
>>> Foo(12)
Traceback (most recent call last):
...
ValueError: Use `make_foo` instead.
"""
cdef long value
def __cinit__(self, value, secret):
if (secret != __SECRET__):
raise ValueError("Use `make_foo` instead.")
self.value = value
def __repr__(self):
return "Foo: #{}".format(self.get_id())
def get_id(self):
return foo_id(self.value)
def make_foo(idx):
"""Build a Foo object from an id.
>>> make_foo(12)
Foo: #12
"""
value = build_foo(idx)
return Foo(value, __SECRET__)
