C API PyObject_Call segfaults with string

[Jen Kris]

Thanks for your reply.  

I eliminated the DECREF and now it doesn't segfault but it returns 0x0.  Same when I substitute pListStrE for pListStr.  pListStr contains the string representation of the fileid, so it seemed like the one to use.  According to  http://web.mit.edu/people/amliu/vrut/python/ext/buildValue.html, PyBuildValue "builds a tuple only if its format string contains two or more format units" and that doc contains examples. 


Feb 9, 2022, 16:52 by songofacandy at gmail.com:

> On Thu, Feb 10, 2022 at 9:42 AM Jen Kris via Python-list
> <python-list at python.org> wrote:
>
>>
>> I have everything finished down to the last line (sentences = gutenberg.sents(fileid)) where I use  PyObject_Call to call gutenberg.sents, but it segfaults.  The fileid is a string -- the first fileid in this corpus is "austen-emma.txt."
>>
>> pName = PyUnicode_FromString("nltk.corpus");
>> pModule = PyImport_Import(pName);
>>
>> pSubMod = PyObject_GetAttrString(pModule, "gutenberg");
>> pFidMod = PyObject_GetAttrString(pSubMod, "fileids");
>> pSentMod = PyObject_GetAttrString(pSubMod, "sents");
>>
>> pFileIds = PyObject_CallObject(pFidMod, 0);
>> pListItem = PyList_GetItem(pFileIds, listIndex);
>> pListStrE = PyUnicode_AsEncodedString(pListItem, "UTF-8", "strict");
>> pListStr = PyBytes_AS_STRING(pListStrE);
>> Py_DECREF(pListStrE);
>>
>
> HERE.
> PyBytes_AS_STRING() returns pointer in the pListStrE Object.
> So Py_DECREF(pListStrE) makes pListStr a dangling pointer.
>
>>
>> // sentences = gutenberg.sents(fileid)
>> PyObject *c_args = Py_BuildValue("s", pListStr);
>>
>
> Why do you encode&decode pListStrE?
> Why don't you use just pListStrE?
>
>> PyObject *NullPtr = 0;
>> pSents = PyObject_Call(pSentMod, c_args, NullPtr);
>>
>
> c_args must tuple, but you passed a unicode object here.
> Read https://docs.python.org/3/c-api/arg.html#c.Py_BuildValue
>
>
>> The final line segfaults:
>> Program received signal SIGSEGV, Segmentation fault.
>> 0x00007ffff6e4e8d5 in _PyEval_EvalCodeWithName ()
>>  from /usr/lib/x86_64-linux-gnu/libpython3.8.so.1.0
>>
>> My guess is the problem is in Py_BuildValue, which returns a pointer but it may not be constructed correctly.  I also tried it with "O" and it doesn't segfault but it returns 0x0.
>>
>> I'm new to using the C API.  Thanks for any help.
>>
>> Jen
>>
>>
>> --
>> https://mail.python.org/mailman/listinfo/python-list
>>
>
> Bests,
>
> -- 
> Inada Naoki  <songofacandy at gmail.com>
>