Sage uses some metaclasses, such as
sage.misc.nested_class.NestedClassMetaclass, and also
sage.misc.classcall_metaclass.ClasscallMetaclass, which is derived
from the former.
Some people try to cythonise the metaclasses, aiming at a faster class
creation (cf. sage trac #12808).
What works is to rename the ..._metaclass.py files into .pyx files,
but keep the classes non-cdef.
It also works to create a cdef class ClasscallType (implementing the
methods of ClasscallMetaclass in Cython), and then have a non-cdef
class ClasscallMetaclass(ClasscallType, NestedClass):
pass
I tried to go further, and have
cdef class NestedClassMetaclass(type):
...
cdef class (ClasscallMetaclass(NestedClassMetaclass):
...
However, that led to segfaults. Therefore my question: Is it possible
to have a metaclass cdef'ed? Or is there likely to be a different
error?
> Sage uses some metaclasses, such as > sage.misc.nested_class.NestedClassMetaclass, and also > sage.misc.classcall_metaclass.ClasscallMetaclass, which is derived > from the former.
> Some people try to cythonise the metaclasses, aiming at a faster class > creation (cf. sage trac #12808).
> What works is to rename the ..._metaclass.py files into .pyx files, > but keep the classes non-cdef.
> It also works to create a cdef class ClasscallType (implementing the > methods of ClasscallMetaclass in Cython), and then have a non-cdef > class ClasscallMetaclass(ClasscallType, NestedClass): > pass
> I tried to go further, and have > cdef class NestedClassMetaclass(type): > ... > cdef class (ClasscallMetaclass(NestedClassMetaclass): > ...
> However, that led to segfaults. Therefore my question: Is it possible > to have a metaclass cdef'ed? Or is there likely to be a different > error?
> Best regards, > Simon
No, cdef'ed metaclasses aren't supported. Can you provide a minimal example that segfaults that must be a bug?
On 24 Apr., 11:19, Vitja Makarov <vitja.maka...@gmail.com> wrote:
> 2012/4/24 Simon King <simon.k...@uni-jena.de>:
> No, cdef'ed metaclasses aren't supported.
Thank you! So, it will be the slightly more complicated approach with
a metaclass being a non-cdef class that inherits from a cdef class (it
is tested that it works).
> Can you provide a minimal
> example that segfaults that must be a bug?
Well, is it a bug if it is not supported? But I can try to cook
something up.
On 24 Apr., 11:57, Simon King <simon.k...@uni-jena.de> wrote:
> > Can you provide a minimal
> > example that segfaults that must be a bug?
> Well, is it a bug if it is not supported? But I can try to cook
> something up.
Here is a small example.
Put the following into a file meta.pyx:
cdef class CyMetaclass(type):
def __init__(cls, *args):
print cls.__name__
class PyMetaclass(type):
def __init__(cls, *args):
print
cls.__name__
Then, start Sage and do:
sage: attach meta.pyx
Compiling ./meta.pyx...
sage: class A(object):
....: __metaclass__ = PyMetaclass
....:
A
sage: a = A()
sage: a
<__main__.A object at 0x450bd10>
sage: class B(object):
....: __metaclass__ = CyMetaclass
....:
B
*** glibc detected *** python: corrupted double-linked list:
0x0000000004611870 ***
Afterwards, it is not possible to do anything, as neither the sage
prompt nor a shell prompt appears - I have to close the terminal.
I hope that the example helps - in fact, Sage's NestedClassMetaclass
provides an __init__ method, and so the example probably demonstrates
what goes wrong in the "real" application (including the fact that one
does not return to the shell).
------------------------------------------------------------------------
Unhandled SIGSEGV: A segmentation fault occurred in Sage.
This probably occurred because a *compiled* component of Sage has a
bug
in it and is not properly wrapped with sig_on(), sig_off(). You might
want to run Sage under gdb with 'sage -gdb' to debug this.
Sage will now terminate.
------------------------------------------------------------------------
/home/simon/SAGE/sage-5.0.beta7/spkg/bin/sage: Zeile 308: 3264
Speicherzugriffsfehler sage-ipython "$@" -i
Florent Hivert's solution at #12808 is roughly as follows:
In meta.pyx:
cdef class CyMetaclass:
def __call__(cls, *args, **opts):
print cls.__name__
return type.__call__(cls, *args, **opts)
class PyMetaclass(type):
def __init__(cls, *args):
print cls.__name__
class PyCyMetaclass(CyMetaclass, PyMetaclass): pass
And then in a Sage session:
sage: attach meta.pyx
Compiling ./meta.pyx...
sage: class A(object):
....: __metaclass__ = PyCyMetaclass
....:
A
sage: a = A()
A
sage: a
<__main__.A object at 0x450bf50>
Hence, the metaclass is non-cdef, in particular the init method
belongs to a non-cdef class, but the __call__ method is inherited from
a cdef class.
------------------------------------------------------------------------
Unhandled SIGSEGV: A segmentation fault occurred in Sage.
This probably occurred because a *compiled* component of Sage has a
bug
in it and is not properly wrapped with sig_on(), sig_off(). You might
want to run Sage under gdb with 'sage -gdb' to debug this.
Sage will now terminate.
------------------------------------------------------------------------
/home/simon/SAGE/sage-5.0.beta7/spkg/bin/sage: Zeile 308: 4043
Speicherzugriffsfehler sage-ipython "$@" -i
> On 24 Apr., 12:39, Stefan Behnel <stefan...@behnel.de> wrote: >> Did you notice that you didn't call the supertype's __init__() in your >> first example? Not sure if it matters - just occurred to me.
> Yes, I noticed, and no, calling it does not help.
> Actually in the current code in sage.misc.nested_class.py, > type.__init__ it is not called either.
> Let's try in meta.pyx: > cdef class CyMetaclass(type): > cdef class CyMetaclass(type): > def __init__(cls, *args, **opts): > print cls.__name__ > type.__init__(cls, *args, **opts)
> class PyMetaclass(type): > def __init__(cls, *args): > print cls.__name__
> ------------------------------------------------------------------------ > Unhandled SIGSEGV: A segmentation fault occurred in Sage. > This probably occurred because a *compiled* component of Sage has a > bug > in it and is not properly wrapped with sig_on(), sig_off(). You might > want to run Sage under gdb with 'sage -gdb' to debug this. > Sage will now terminate. > ------------------------------------------------------------------------ > /home/simon/SAGE/sage-5.0.beta7/spkg/bin/sage: Zeile 308: 4043 > Speicherzugriffsfehler sage-ipython "$@" -i
> Best regards, > Simon
Hmm, it seems that Cython doesn't support type objects:
cdef class CyMetaclass(type): def __init__(cls, *args): print 'haha' print cls.__name__
CyMetaclass('foo', (object,), {})
print cls.__name__ causes segfault here and glibc's SIGABRT if removed.
> On 24 Apr., 12:39, Stefan Behnel <stefan...@behnel.de> wrote:
> > Did you notice that you didn't call the supertype's __init__() in your
> > first example? Not sure if it matters - just occurred to me.
> Yes, I noticed, and no, calling it does not help.
> Actually in the current code in sage.misc.nested_class.py,
> type.__init__ it is not called either.
> Let's try in meta.pyx:
> cdef class CyMetaclass(type):
> cdef class CyMetaclass(type):
> def __init__(cls, *args, **opts):
> print cls.__name__
> type.__init__(cls, *args, **opts)
> class PyMetaclass(type):
> def __init__(cls, *args):
> print cls.__name__
I think you should add the following three line at the beginning of you file:
cdef extern from "object.h":
ctypedef class __builtin__.type [object PyHeapTypeObject]:
pass
I'm not exactly sure what it does if you don't do that but it solve the
segfault for me.
Cheers,
Florent
Meta2.pyx::
============
cdef extern from "object.h":
ctypedef class __builtin__.type [object PyHeapTypeObject]:
pass
cdef class CyMetaclass(type):
def __init__(cls, *args, **opts):
print cls.__name__
type.__init__(cls, *args, **opts)
============
sage: attach Meta2.pyx
Compiling ./Meta2.pyx...
sage: class A(object):
....: __metaclass__ = CyMetaclass
....: A
On 24 Apr., 15:02, Florent Hivert <Florent.Hiv...@lri.fr> wrote:
> cdef extern from "object.h":
> ctypedef class __builtin__.type [object PyHeapTypeObject]:
> pass
That's very cool! It even works in the "real" application.
I will see (at #12808) whether it has an effect on the time spent for
creating classes. I think the speedup could be considerable: After
all, NestedClassMetaclass occurs even more often in Sage than
ClasscallMetaclass.
