[Numpy-discussion] Error in deallocation ?

Mon Oct 29 05:31:23 EDT 2007

It seems this could be a deadlock somewhere ? The line 835 is where threads
are allowed.
I had another problem today with the same instruction ((x-self.loc) *
(1/self.scale)), the computation just hung. I'm trying to test without the
thread lock.

Matthieu

2007/10/19, Matthieu Brucher <matthieu.brucher at gmail.com>:
>
> Here is an excerpt of the stack on the numpy svn of wednesday :
>
> #0  0x40000402 in __kernel_vsyscall ()
> #1  0x00b8e382 in sem_post at GLIBC_2.0 () from /lib/libpthread.so.0
> #2  0x080fe5b7 in PyThread_release_lock (lock=0x80d0fb2) at
> Python/thread_pthread.h:374
> #3  0x080d0fb2 in PyEval_SaveThread () at Python/ceval.c:299
> #4  0x4064ec7a in dotblas_innerproduct (dummy=0x0, args=0x960522c) at
> numpy/core/blasdot/_dotblas.c:835
> #5  0x0811ecf1 in PyCFunction_Call (func=0x960522c, arg=0x80cfec7,
> kw=0x942068c) at Objects/methodobject.c:73
> #6  0x080cfec7 in call_function (pp_stack=0x0, oparg=1) at
> Python/ceval.c:3564
> #7  0x080cb19a in PyEval_EvalFrameEx. () at Python/ceval.c:2267
> #8  0x080ca848 in PyEval_EvalCodeEx. () at Python/ceval.c:2831
> #9  0x0811de5b in function_call (func=0x0, arg=0x960516c, kw=0x965b148) at
> Objects/funcobject.c:517
> #10 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x960516c, kw=0x8061a66) at
> Objects/abstract.c:1860
> #11 0x08061a66 in instancemethod_call (func=0x95f16ac, arg=0x94f5824,
> kw=0x805bdb6) at Objects/classobject.c:2497
> #12 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x95f16ac, kw=0x80a86e9) at
> Objects/abstract.c:1860
> #13 0x080a86e9 in slot_tp_call (self=0xbf975c78, args=0x0, kwds=0x805bdb6)
> at Objects/typeobject.c:4633
> #14 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x95f16ac, kw=0x80d017b) at
> Objects/abstract.c:1860
> #15 0x080d017b in do_call (func=0x0, pp_stack=0x1, na=1, nk=154955456) at
> Python/ceval.c:3775
> #16 0x080cfd56 in call_function (pp_stack=0x0, oparg=1) at
> Python/ceval.c:3587
> #17 0x080cb19a in PyEval_EvalFrameEx. () at Python/ceval.c:2267
> #18 0x080ca848 in PyEval_EvalCodeEx. () at Python/ceval.c:2831
> #19 0x0811de5b in function_call (func=0x0, arg=0x960520c, kw=0x30) at
> Objects/funcobject.c:517
> #20 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x960520c, kw=0x8061a66) at
> Objects/abstract.c:1860
> #21 0x08061a66 in instancemethod_call (func=0x95ecbac, arg=0x94f57d4,
> kw=0x805bdb6) at Objects/classobject.c:2497
> #22 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x95ecbac, kw=0x80a86e9) at
> Objects/abstract.c:1860
> #23 0x080a86e9 in slot_tp_call (self=0xbf97609c, args=0x0, kwds=0x805bdb6)
> at Objects/typeobject.c:4633
> #24 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x95ecbac, kw=0x80d017b) at
> Objects/abstract.c:1860
> #25 0x080d017b in do_call (func=0x0, pp_stack=0x1, na=1, nk=156957888) at
> Python/ceval.c:3775
> #26 0x080cfd56 in call_function (pp_stack=0x0, oparg=1) at
> Python/ceval.c:3587
> #27 0x080cb19a in PyEval_EvalFrameEx. () at Python/ceval.c:2267
> #28 0x080ca848 in PyEval_EvalCodeEx. () at Python/ceval.c:2831
> #29 0x0811de5b in function_call (func=0x0, arg=0x960518c, kw=0x9548a1) at
> Objects/funcobject.c:517
> #30 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x960518c, kw=0x8061a66) at
> Objects/abstract.c:1860
> #31 0x08061a66 in instancemethod_call (func=0x960526c, arg=0x95fc874,
> kw=0x805bdb6) at Objects/classobject.c:2497
> #32 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x960526c, kw=0x80a86e9) at
> Objects/abstract.c:1860
> #33 0x080a86e9 in slot_tp_call (self=0xbf9764c0, args=0x0, kwds=0x805bdb6)
> at Objects/typeobject.c:4633
> #34 0x0805bdb6 in PyObject_Call (func=0x0, arg=0x960526c, kw=0x80d017b) at
> Objects/abstract.c:1860
> #35 0x080d017b in do_call (func=0x0, pp_stack=0x1, na=1, nk=157640136) at
> Python/ceval.c:3775
> #36 0x080cfd56 in call_function (pp_stack=0x0, oparg=1) at
> Python/ceval.c:3587
> #37 0x080cb19a in PyEval_EvalFrameEx. () at Python/ceval.c:2267
> #38 0x080ca848 in PyEval_EvalCodeEx. () at Python/ceval.c:2831
> #39 0x0811de5b in function_call (func=0x0, arg=0x95f260c, kw=0x0) at
> Objects/funcobject.c:517
> #40 0x0805bdb6 in PyObject_Call (func=0x9609934, arg=0x95f260c,
> kw=0x8061a66) at Objects/abstract.c:1860
>
> Seems that the bug could be somewhere in the handling of the dot blas
> module ?
>
> Matthieu
>
> 2007/10/15, Travis E. Oliphant <oliphant at enthought.com >:
> >
> > Matthieu Brucher wrote:
> > >
> > >     The problem is that there is a data-type reference counting error
> > some
> > >     where that is attempting to deallocate the built-in data-type 'l'
> > >
> > >
> > >
> > > That's what I supposed, but I couldn't find the reason why it wanted
> > > to do this
> > >
> > >
> > >     It's not really a Python error but a logging.  The code won't let
> > you
> > >     deallocate the built-ins, but it will tell you that something
> > >     tried to.
> > >
> > >     Reference counting on data-types is easy to get wrong
> > >     (particularly with
> > >     Pyrex extension modules) because most calls consume a reference to
> > the
> > >     data-type (if they return an object that  contains a reference to
> > the
> > >     data-type).
> > >
> > >     It is a bug, and it would be nice to figure it out, but that would
> > >     require the code that caused it.
> > >
> > >
> > > I've updated my numpy version to the latest svn, the behaviour seems
> > > to be different (more warnings), I'll try to give more information
> > > about the error, but giving the whole code will not be simple (it uses
> > > a big data file that seems to trigger the error as with other data
> > > files, the error didn't show up :()
> > >
> >
> > There are two types of errors that can occur with reference counting on
> > data-types.
> >
> > 1) There are too many DECREF's --- this gets us to the error quickly and
> >
> > is usually easy to reproduce
> > 2) There are too many INCREF's (the reference count keeps going up until
> > the internal counter wraps around to 0 and deallocation is attempted)
> > --- this error is harder to reproduce and usually takes a while before
> > it happens in the code.
> >
> >
> > -Travis
> >
> > _______________________________________________
> > Numpy-discussion mailing list
> > Numpy-discussion at scipy.org
> > http://projects.scipy.org/mailman/listinfo/numpy-discussion
> >
>
>

-- 
French PhD student
Website : http://miles.developpez.com/
Blogs : http://matt.eifelle.com and http://blog.developpez.com/?blog=92
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20071029/f10289a8/attachment.html>