Hi Luc, Thanks for this. I thought that a limited use of python's own abc module would be one of the safer uses of metaclasses. But apparently even this is not without pain. My google-fu is slowly warming up this new year, and eventually I did find a message (from Ralf) suggesting that a boost python wrapper's __metaclass__ is its __class__, depending how it is constructed. I still couldn't work out from that how to proceed though, so I'm happy to put this aside. Your explicit alternative produces exactly the behaviour I was after, in a more obvious way. Cheers -- David On 8 January 2014 10:05, Luc Bourhis wrote:

Hi David,

I am afraid this is a dead end. Indeed scitbx.lstbx.normal_eqns.non_linear_ls is a Boost Python wrapper. What you could do instead is write:

class Refinery(object):

def __new__(cls, *args, **kwds): assert cls.__bases__ return object.__new__(cls)

class AdaptLstbx(Refinery, normal_eqns.non_linear_ls): pass

It is then impossible to construct an instance of Refinery. Only heirs of Refinery can be instantiated. You could make the assert more subtle to suit your needs of course.

Generally speaking, i.e. it would not apply to you, David, if you do not plan to commit metaclass-based code to the cctbx, I would strongly advise cctbx developers against using metaclasses at all. This is one of the surest way to make the code impossible to understand to your fellow programmers as it introduces highly implicit behaviour: explicit is better than implicit in the long run.

As any rule, it has exceptions: there are 4 uses of metaclasses in the cctbx as I write this. But they are highly specialised and self-contained uses.

Best wishes,

Luc J. Bourhis

...

Hi,

Does anyone know how to use __metaclass__ specification alongside inheritance from boost python classes? This code illustrates the problem:

import abc from scitbx.lstbx import normal_eqns

class Refinery(object): __metaclass__ = abc.ABCMeta pass

class AdaptLstbx(Refinery, normal_eqns.non_linear_ls): pass

The result is

Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/usr/lib/python2.7/abc.py", line 87, in __new__ cls = super(ABCMeta, mcls).__new__(mcls, name, bases, namespace) TypeError: Error when calling the metaclass bases metaclass conflict: the metaclass of a derived class must be a (non-strict) subclass of the metaclasses of all its bases

The real Refinery is a class that has one method that must be supplied by a concrete subclass. I like the idea of making it impossible to instantiate a base Refinery rather than relying on e.g. raise NotImplementedError() in that method definition. This is what ABCMeta would give me, if only I could get it to work here. Google found me some answers that require me to know the __metaclass__ of the bases (as suggested by the error message), but normal_eqns.non_linear_ls.__metaclass__ does not exist.

I don't terribly mind the raise NotImplementedError() alternative, but I thought that if I can easily use ABCMeta, I'd like to.

Cheers

-- David _______________________________________________ cctbxbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/cctbxbb

_______________________________________________ cctbxbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/cctbxbb

import abc from scitbx.lstbx import normal_eqns class Refinery(object): class __metaclass__(normal_eqns.non_linear_ls.__class__, abc.ABCMeta): pass

class AdaptLstbx(Refinery, normal_eqns.non_linear_ls): pass

Except that it does not prevent r = Refinery() after all since the __metaclass__ has a valid instantiation through normal_eqns.non_linear_ls.__class__. which gives me a flashback: I've already burnt myself with that! Luc