Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.maint.python > #14697 > unrolled thread

Fwd: change in the extention importation with 3.11

Back to article view | Back to linux.debian.maint.python

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Fwd: change in the extention importation with 3.11 Alastair McKinstry <alastair.mckinstry@sceal.ie> - 2022-12-06 18:00 +0100

#14697 — Fwd: change in the extention importation with 3.11

On 06/12/2022 13:47, picca wrote:
> Hello, I am trying to fix this bug
>
I'm debugging something similar, 
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1024911

> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1024859
>
> I find the error message not very informative...
>
> :~/$ python3.11
> Python 3.11.0+ (main, Nov  4 2022, 09:23:33) [GCC 12.2.0] on linux
> Type "help", "copyright", "credits" or "license" for more information.
>>>> import scitbx_linalg_ext
> Traceback (most recent call last):
>   File "<stdin>", line 1, in <module>
> SystemError: initialization of scitbx_linalg_ext raised unreported
> exception
>>>>
>
The problem is that python is catching a C/C++  exception.

You can debug with:

gdb /usr/bin/python3.11

...

(gdb) catch throw

(gdb) run

>> import scitbx_linalg_ext

...

> I would like your opinion and some help in order to fix this or at 
> least understand what is going on.
>
> ther extension was build with boost_python, maybe something is wrong 
> also with boost_python and Python3.11.
>
> I do not know if other packages using boost_python are affected like this.
>
> thanks for considering
>
> Frederic
>

Both cctbx and ecflow are breaking on generating new enums.

A simple test case is:

```

#include <boost/python.hpp>
     using namespace boost::python;



//  enum has flags 0x42000, no traversefunction set



class CheckPt  {
public:
    /// NEVER   - the check pt file is never saved
    /// ON_TIME - the check pt file is saved periodically. specified by 
checkPtInterval.
    /// ALWAYS  - the check pt file is saved after any state change
    /// UNDEFINED   - Internal use only
    enum Mode { NEVER, ON_TIME, ALWAYS, UNDEFINED};

};


void inc_enum () {
    enum_<CheckPt::Mode>("CheckPt",
             "CheckPt is enum that is used to control check pointing in 
the `ecflow_server`_\n\n"
             "- NEVER  : Switches of check pointing\n"
             "- ON_TIME: `check point`_ file is saved periodically, 
specified by checkPtInterval. This is the default.\n"
             "- ALWAYS : `check point`_ file is saved after any state 
change, *not* recommended for large definitions\n"
             "- UNDEFINED : None of the the above, used to provide 
default argument\n"
    )
    .value("NEVER",  CheckPt::NEVER)
    .value("ON_TIME",CheckPt::ON_TIME)
    .value("ALWAYS", CheckPt::ALWAYS)
    .value("UNDEFINED", CheckPt::UNDEFINED)
    ;
}

char const *greet()
{
   return "Hello world\n";
}

BOOST_PYTHON_MODULE(wrapper)
{
     def ("greet", greet);
     inc_enum();
}

```


What appears to be happening is a change in python3.11: 
https://docs.python.org/ja/3.11/whatsnew/3.11.html

     The PyType_Ready() function now raises an error if a type is 
defined with the Py_TPFLAGS_HAVE_GC flag set but has no traverse 
function         (PyTypeObject.tp_traverse). (Contributed by Victor 
Stinner in bpo-44263.)

and indeed, the enum types are created without a traverse function but 
with the HAVE_GC flag set.

It doesn't at first glance look like the code for boost1.80 is different.

A Python/boost expert is now needed.


Regards

Alastair McKinstry

mckinstry@debian.org



-- 
Alastair McKinstry,
GPG: 82383CE9165B347C787081A2CBE6BB4E5D9AD3A5
ph: +353 87 6847928 e: alastair@sceal.ie, im: @sceal.ie:mckinstry

[toc] | [standalone]

Back to top | Article view | linux.debian.maint.python

csiph-web