Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-10 Thread Tobias Hansen 
On 12/09/2016 09:54 PM, Nicolas M. Thiery wrote:
> On Wed, Dec 07, 2016 at 02:50:15PM +, Tobias Hansen wrote:
>> On 12/07/2016 01:26 PM, Jeroen Demeyer wrote:
>> If you look at the patch [1] that was applied, there is a new check for
>> tp_new(). I'm not 100% sure that's what causes the problem, but it looks
>> suspicious.
>>
>> [1] https://bugs.python.org/file45526/update_one_slot2-2.7.patch
> 
> This seems very similar to the issue that Julien Cristau (in CC) had
> stumbled into this Spring while trying to run Sage on top of a then
> recent Python. That version of Python included a patch at exactly the
> same line in Objects/typeobject.c. Julien argued that this patch was
> backward incompatible and it got reverted:
> 
> https://bugs.python.org/issue25731
> 
> It seems it has come back in a new form. Julien: any opinion on that
> new form? Is it a backward incompatible patch in Python that could be
> argued to be reverted / fixed? If yes, is there any chance to make
> this happen given the extremely tight schedule that we have?
> 
> I am going to dig in further tomorrow on the Sage side in search for a
> workaround. As usual, the more eyeballs, the better.
> 
> Cheers,
>   Nicolas
> 
> PS: thanks Tobias for the ping!
> 
> --
> Nicolas M. Thiéry "Isil" 
> http://Nicolas.Thiery.name/
> 

Yeah, it would be good to know if this is a regression in Python or a
bug in Sage. If it's considered a Python regression we could try to get
the change reverted in the Python Debian package.

Of course a workaround in Sage would be even better, because that is
fully under our control.

Best,
Tobias

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-09 Thread Nicolas M. Thiery 
On Wed, Dec 07, 2016 at 02:50:15PM +, Tobias Hansen wrote:
> On 12/07/2016 01:26 PM, Jeroen Demeyer wrote:
> If you look at the patch [1] that was applied, there is a new check for
> tp_new(). I'm not 100% sure that's what causes the problem, but it looks
> suspicious.
>
> [1] https://bugs.python.org/file45526/update_one_slot2-2.7.patch

This seems very similar to the issue that Julien Cristau (in CC) had
stumbled into this Spring while trying to run Sage on top of a then
recent Python. That version of Python included a patch at exactly the
same line in Objects/typeobject.c. Julien argued that this patch was
backward incompatible and it got reverted:

https://bugs.python.org/issue25731

It seems it has come back in a new form. Julien: any opinion on that
new form? Is it a backward incompatible patch in Python that could be
argued to be reverted / fixed? If yes, is there any chance to make
this happen given the extremely tight schedule that we have?

I am going to dig in further tomorrow on the Sage side in search for a
workaround. As usual, the more eyeballs, the better.

Cheers,
Nicolas

PS: thanks Tobias for the ping!

--
Nicolas M. Thiéry "Isil" 
http://Nicolas.Thiery.name/

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Tobias Hansen 
On 12/07/2016 01:26 PM, Jeroen Demeyer wrote:
> What is the problem really? Neither from reading the bug report nor from
> reading this thread do I understand what exactly changed in Python
> 2.7.13 causing this breakage.
> 

If you look at the patch [1] that was applied, there is a new check for
tp_new(). I'm not 100% sure that's what causes the problem, but it looks
suspicious.


[1] https://bugs.python.org/file45526/update_one_slot2-2.7.patch

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Jeroen Demeyer 

I created a ticket for this upgrade:
https://trac.sagemath.org/ticket/22037

But we should first upgrade to Python 2.7.12 before we can tackle 2.7.13:
https://trac.sagemath.org/ticket/19735

Anyone willing to work on the Python 2.7.12 upgrade? It would at least 
make it easier to debug the problems with 2.7.13.


--
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Jeroen Demeyer 
What is the problem really? Neither from reading the bug report nor from 
reading this thread do I understand what exactly changed in Python 
2.7.13 causing this breakage.


--
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Tobias Hansen 
On 12/07/2016 11:07 AM, Tobias Hansen wrote:
> On 12/07/2016 10:57 AM, Francois Bissey wrote:
>>
>>> On 7/12/2016, at 21:15, Jeroen Demeyer  wrote:
>>>
>>> On 2016-12-07 04:17, Francois Bissey wrote:
 But I am not sure how to do the
 __new__ = object.__new__ in cython.
>>>
>>> You certainly cannot do that since Cython's __new__ does non-trivial stuff 
>>> like setting the vtab for c(p)def method calls. This is also partially the 
>>> reason why cannot even implement __new__ for Cython cdef classes.
>>>
>>
>> OK, unless you can do something at the level
>> of tp_call (slot_tp_call may be) like Tobias
>> tried, that makes it a fundamental cython
>> problem. That kind of stuff will likely
>> happen with any cython cdef classes, right?
>>
>> François
>>
> 
> That's what I tried with my cython example (attached), but there the
> error didn't show.
> 
> Tobias
> 

I also just ran the Cython test suite with Python 2.7.13rc1 and there
were no problems.

Tobias

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Tobias Hansen 
On 12/07/2016 10:57 AM, Francois Bissey wrote:
> 
>> On 7/12/2016, at 21:15, Jeroen Demeyer  wrote:
>>
>> On 2016-12-07 04:17, Francois Bissey wrote:
>>> But I am not sure how to do the
>>> __new__ = object.__new__ in cython.
>>
>> You certainly cannot do that since Cython's __new__ does non-trivial stuff 
>> like setting the vtab for c(p)def method calls. This is also partially the 
>> reason why cannot even implement __new__ for Cython cdef classes.
>>
> 
> OK, unless you can do something at the level
> of tp_call (slot_tp_call may be) like Tobias
> tried, that makes it a fundamental cython
> problem. That kind of stuff will likely
> happen with any cython cdef classes, right?
> 
> François
> 

That's what I tried with my cython example (attached), but there the
error didn't show.

Tobias

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.
from distutils.core import setup
from Cython.Build import cythonize

setup(
ext_modules = cythonize("test.pyx")
)

cdef class WithEqualityById:
pass

class CachedRepresentation:
pass

class UniqueRepresentation(CachedRepresentation, WithEqualityById):
pass

cdef class SageObject:
pass

class UnknownClass(UniqueRepresentation, SageObject):
pass

Unknown = UnknownClass()

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Francois Bissey 

> On 7/12/2016, at 21:15, Jeroen Demeyer  wrote:
> 
> On 2016-12-07 04:17, Francois Bissey wrote:
>> But I am not sure how to do the
>> __new__ = object.__new__ in cython.
> 
> You certainly cannot do that since Cython's __new__ does non-trivial stuff 
> like setting the vtab for c(p)def method calls. This is also partially the 
> reason why cannot even implement __new__ for Cython cdef classes.
> 

OK, unless you can do something at the level
of tp_call (slot_tp_call may be) like Tobias
tried, that makes it a fundamental cython
problem. That kind of stuff will likely
happen with any cython cdef classes, right?

François

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-07 Thread Jeroen Demeyer 

On 2016-12-07 04:17, Francois Bissey wrote:

But I am not sure how to do the
__new__ = object.__new__ in cython.


You certainly cannot do that since Cython's __new__ does non-trivial 
stuff like setting the vtab for c(p)def method calls. This is also 
partially the reason why cannot even implement __new__ for Cython cdef 
classes.


--
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-06 Thread Francois Bissey 
I am guessing we should something more fundamental in 
sage/structure/sage_object.pyx
since we are dealing with SageObject classes.
But I am not sure how to do the 
__new__ = object.__new__ in cython.

François

> On 7/12/2016, at 15:25, than...@debian.org wrote:
> 
> When setting __new__ = object.__new__() for UnknownClass it goes on to the 
> next similar error:
> 
>...
>File 
> "/usr/lib/python2.7/dist-packages/sage/categories/sets_cat.py", 
> line 2752, in 
>  cartesian_product = CartesianProductFunctor()
>File "sage/misc/classcall_metaclass.pyx", line 330, in 
> sage.misc.classcall_metaclass.ClasscallMetaclass.__call__ 
> (/sage/misc/classcall_metaclass.c:1413)
>File "sage/misc/cachefunc.pyx", line 1059, in 
> sage.misc.cachefunc.CachedFunction.__call__ 
> (/sage/misc/cachefunc.c:6080)
>File 
> "/usr/lib/python2.7/dist-packages/sage/structure/unique_representation.py",
>  line 1022, in __classcall__
>  instance = typecall(cls, *args, **options)
>File "sage/misc/classcall_metaclass.pyx", line 497, in 
> sage.misc.classcall_metaclass.typecall 
> (/sage/misc/classcall_metaclass.c:1862)
>  TypeError: 
> sage.misc.fast_methods.WithEqualityById.__new__(CartesianProductFunctor) is 
> not safe, use sage.categories.functor.Functor.__new__()

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-06 Thread than...@debian.org 
Am Mittwoch, 7. Dezember 2016 01:43:09 UTC schrieb tha...@debian.org:
>
> On 12/07/2016 01:27 AM, François Bissey wrote: 
> > On 07/12/16 12:20, tha...@debian.org  wrote: 
> >> Hi sage-devel, 
> >> 
> >> we're almost ready to upload Sage to Debian (in fact we basically have 
> >> to upload it this week to make sure it's included in the next Debian 
> >> release). 
> >> 
> >> However, on Sunday python 2.7.13rc1 was uploaded to Debian and now we 
> >> are facing a bug that I didn't quite manage to work around yet and that 
> >> blocks any development at the moment. 
> >> 
> >> This happens whenever Unknown is imported, meaning during the docbuild 
> >> and when starting sage: 
> >> 
> >> ... 
> >> from sage.misc.unknown import Unknown 
> >>   File 
> >> "/usr/lib/python2.7/dist-packages/sage/misc/unknown.py", line 
> >> 164, in  
> >> Unknown = UnknownClass() 
> >>   File "sage/misc/classcall_metaclass.pyx", line 330, in 
> >> sage.misc.classcall_metaclass.ClasscallMetaclass.__call__ 
> >> (/sage/misc/classcall_metaclass.c:1413) 
> >>   File "sage/misc/cachefunc.pyx", line 1059, in 
> >> sage.misc.cachefunc.CachedFunction.__call__ 
> >> (/sage/misc/cachefunc.c:6080) 
> >>   File 
> >> 
> "/usr/lib/python2.7/dist-packages/sage/structure/unique_representation.py",
>  
>
> >> 
> >> line 1022, in __classcall__ 
> >> instance = typecall(cls, *args, **options) 
> >>   File "sage/misc/classcall_metaclass.pyx", line 497, in 
> >> sage.misc.classcall_metaclass.typecall 
> >> (/sage/misc/classcall_metaclass.c:1862) 
> >> TypeError: 
> sage.misc.fast_methods.WithEqualityById.__new__(UnknownClass) 
> >> is not safe, use object.__new__() 
> >> 
> >> I'm pretty sure it's caused by the change of 
> >> https://bugs.python.org/issue5322 
> >> which is included in python 2.7.13rc1. 
> >> 
> >> I hope you can help me to fix this, or at least provide a workaround. 
> >> 
> > 
> > What did you try so far? 
> > The most obvious thing to try, as far as I can see, is to add 
> > __new__ = object.__new__() 
> > before 
> > def __init__(self): 
> > in sage/misc/unknown.py 
> > 
> > Francois 
> > 
>
> I created a minimal cython example with classes and cdef classes that 
> inherit from each other in the same way as here to see if this is caused 
> by WithEqualityById being a cdef class. In the example everything worked 
> as it should. 
>
> Just because it appears in the error message I also tried replacing the 
> two 
> (type).tp_call(cls, args, kwds) 
> in 
> sage/misc/classcall_metaclass.pyx 
> by 
> type.__call__(cls, *args, **kwds). 
>
> That didn't help either. 
>
> I'll try your suggestion next, thanks. 
>
> Best, Tobias 
>

When setting __new__ = object.__new__() for UnknownClass it goes on to the 
next similar error:

   ...
   File 
"/usr/lib/python2.7/dist-packages/sage/categories/sets_cat.py", 
line 2752, in 
 cartesian_product = CartesianProductFunctor()
   File "sage/misc/classcall_metaclass.pyx", line 330, in 
sage.misc.classcall_metaclass.ClasscallMetaclass.__call__ 
(/sage/misc/classcall_metaclass.c:1413)
   File "sage/misc/cachefunc.pyx", line 1059, in 
sage.misc.cachefunc.CachedFunction.__call__ 
(/sage/misc/cachefunc.c:6080)
   File 
"/usr/lib/python2.7/dist-packages/sage/structure/unique_representation.py",
 
line 1022, in __classcall__
 instance = typecall(cls, *args, **options)
   File "sage/misc/classcall_metaclass.pyx", line 497, in 
sage.misc.classcall_metaclass.typecall 
(/sage/misc/classcall_metaclass.c:1862)
 TypeError: 
sage.misc.fast_methods.WithEqualityById.__new__(CartesianProductFunctor) is 
not safe, use sage.categories.functor.Functor.__new__() 

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-06 Thread Tobias Hansen 
On 12/07/2016 01:27 AM, François Bissey wrote:
> On 07/12/16 12:20, than...@debian.org wrote:
>> Hi sage-devel,
>>
>> we're almost ready to upload Sage to Debian (in fact we basically have
>> to upload it this week to make sure it's included in the next Debian
>> release).
>>
>> However, on Sunday python 2.7.13rc1 was uploaded to Debian and now we
>> are facing a bug that I didn't quite manage to work around yet and that
>> blocks any development at the moment.
>>
>> This happens whenever Unknown is imported, meaning during the docbuild
>> and when starting sage:
>>
>> ...
>> from sage.misc.unknown import Unknown
>>   File
>> "/usr/lib/python2.7/dist-packages/sage/misc/unknown.py", line
>> 164, in 
>> Unknown = UnknownClass()
>>   File "sage/misc/classcall_metaclass.pyx", line 330, in
>> sage.misc.classcall_metaclass.ClasscallMetaclass.__call__
>> (/sage/misc/classcall_metaclass.c:1413)
>>   File "sage/misc/cachefunc.pyx", line 1059, in
>> sage.misc.cachefunc.CachedFunction.__call__
>> (/sage/misc/cachefunc.c:6080)
>>   File
>> "/usr/lib/python2.7/dist-packages/sage/structure/unique_representation.py",
>>
>> line 1022, in __classcall__
>> instance = typecall(cls, *args, **options)
>>   File "sage/misc/classcall_metaclass.pyx", line 497, in
>> sage.misc.classcall_metaclass.typecall
>> (/sage/misc/classcall_metaclass.c:1862)
>> TypeError: sage.misc.fast_methods.WithEqualityById.__new__(UnknownClass)
>> is not safe, use object.__new__()
>>
>> I'm pretty sure it's caused by the change of
>> https://bugs.python.org/issue5322
>> which is included in python 2.7.13rc1.
>>
>> I hope you can help me to fix this, or at least provide a workaround.
>>
> 
> What did you try so far?
> The most obvious thing to try, as far as I can see, is to add
> __new__ = object.__new__()
> before
> def __init__(self):
> in sage/misc/unknown.py
> 
> Francois
> 

I created a minimal cython example with classes and cdef classes that
inherit from each other in the same way as here to see if this is caused
by WithEqualityById being a cdef class. In the example everything worked
as it should.

Just because it appears in the error message I also tried replacing the two
(type).tp_call(cls, args, kwds)
in
sage/misc/classcall_metaclass.pyx
by
type.__call__(cls, *args, **kwds).

That didn't help either.

I'll try your suggestion next, thanks.

Best, Tobias

-- 
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.

Re: [sage-devel] Python 2.7.13rc1 breaks UnknownClass

2016-12-06 Thread François Bissey 

On 07/12/16 12:20, than...@debian.org wrote:

Hi sage-devel,

we're almost ready to upload Sage to Debian (in fact we basically have
to upload it this week to make sure it's included in the next Debian
release).

However, on Sunday python 2.7.13rc1 was uploaded to Debian and now we
are facing a bug that I didn't quite manage to work around yet and that
blocks any development at the moment.

This happens whenever Unknown is imported, meaning during the docbuild
and when starting sage:

...
from sage.misc.unknown import Unknown
  File
"/usr/lib/python2.7/dist-packages/sage/misc/unknown.py", line
164, in 
Unknown = UnknownClass()
  File "sage/misc/classcall_metaclass.pyx", line 330, in
sage.misc.classcall_metaclass.ClasscallMetaclass.__call__
(/sage/misc/classcall_metaclass.c:1413)
  File "sage/misc/cachefunc.pyx", line 1059, in
sage.misc.cachefunc.CachedFunction.__call__
(/sage/misc/cachefunc.c:6080)
  File
"/usr/lib/python2.7/dist-packages/sage/structure/unique_representation.py",
line 1022, in __classcall__
instance = typecall(cls, *args, **options)
  File "sage/misc/classcall_metaclass.pyx", line 497, in
sage.misc.classcall_metaclass.typecall
(/sage/misc/classcall_metaclass.c:1862)
TypeError: sage.misc.fast_methods.WithEqualityById.__new__(UnknownClass)
is not safe, use object.__new__()

I'm pretty sure it's caused by the change of
https://bugs.python.org/issue5322
which is included in python 2.7.13rc1.

I hope you can help me to fix this, or at least provide a workaround.



What did you try so far?
The most obvious thing to try, as far as I can see, is to add
__new__ = object.__new__()
before
def __init__(self):
in sage/misc/unknown.py

Francois

--
You received this message because you are subscribed to the Google Groups 
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.