Re: [Boost-build] Access violation patch

Subject: Re: [Boost-build] Access violation patch
From: Steven Watanabe (watanabesj_at_[hidden])
Date: 2011-11-22 09:46:35

AMDG

On 11/22/2011 03:21 AM, Vladimir Prus wrote:
> [Adding list to CC]
> 
> On 21/11/11 19:31, Juraj Ivanèiæ wrote:
>> I produced a patch which fixes access violation from my previous e-mail.
> 
> Thanks for the patch! Comitted in revision 75605.
> 
> - Volodya
> 
>> Index: build/toolset.py
>> ===================================================================
>> --- build/toolset.py    (revision 75578)
>> +++ build/toolset.py    (working copy)
>> @@ -117,8 +117,8 @@
>>                             is specified, then the value of 'feature'
>>                             will be added.
>>       """
>> -    caller = bjam.caller()[:-1]
>> -    if not '.' in rule_or_module and caller.startswith("Jamfile"):
>> +    caller = bjam.caller()
>> +    if not '.' in rule_or_module and caller and
>> caller[:-1].startswith("Jamfile"):
>>           # Unqualified rule name, used inside Jamfile. Most likely
>> used with
>>           # 'make' or 'notfile' rules. This prevents setting flags on
>> the entire
>>           # Jamfile module (this will be considered as rule), but who
>> cares?
>> Index: engine/builtins.c
>> ===================================================================
>> --- engine/builtins.c    (revision 75578)
>> +++ engine/builtins.c    (working copy)
>> @@ -2140,9 +2140,9 @@
>>
>>   PyObject * bjam_caller( PyObject * self, PyObject * args )
>>   {
>> -    PyObject *result = PyString_FromString(
>> -        frame_before_python_call->prev->module->name);
>> -    return result;
>> +    if ( !frame_before_python_call )
>> +        Py_RETURN_NONE;
>> +    return
>> PyString_FromString(frame_before_python_call->prev->module->name);
>>   }
>>
>>   #endif  /* #ifdef HAVE_PYTHON */
>> Index: engine/compile.c
>> ===================================================================
>> --- engine/compile.c    (revision 75578)
>> +++ engine/compile.c    (working copy)
>> @@ -865,6 +865,7 @@
>>
>>       frame_before_python_call = frame;
>>       py_result = PyObject_Call( r->python_function, arguments, kw );
>> +    frame_before_python_call = NULL;

This is still wrong.  The correct fix is
to save and restore frame_before_python_call.
Otherwise, you can get cases like this

bjam:
  python:
    bjam:
      python
    bjam_call -> returns None

>>       Py_DECREF(arguments);
>>       Py_XDECREF(kw);
>>       if ( py_result != NULL )
> 

In Christ,
Steven Watanabe