homepage

initproc is declared to return an int, but what returned values mean is not documented. Noddy_init in http://docs.python.org/3/extending/newtypes.html?highlight=initproc#adding-data-and-methods-to-the-basic-example can be seen to return 0 on success and -1 on error, but that's about it.
Also, when I wrote a function which return 1 on error, on every second invocation the exception would be ignored:
static int Reader_init(Reader *self, PyObject *args, PyObject *keywds)
{
 ...
 if (flags && path) {
 PyErr_SetString(PyExc_ValueError, "cannot use both flags and path");
 return 1;
 }
 ...
}
>>> obj(123, '/tmp')
>>> obj(123, '/tmp')
...
ValueError
>>> obj(123, '/tmp')
>>> obj(123, '/tmp')
...
ValueError
I'm not sure how to interpret this since I couldn't find the documentation for the expected value.

The return value for error conditions should be -1.
- typeobject.c checks with "< 0"
- in _iomodule.c, there is "== -1"
- and pygobject/gobject/gobjectmodule.c just does::
 if (...tp_init(...))
 PyErr_Print();

On Fri, Mar 08, 2013 at 02:30:18PM +0000, Amaury Forgeot d'Arc wrote:
> 
> Amaury Forgeot d'Arc added the comment:
> 
> The return value for error conditions should be -1.
> 
> - typeobject.c checks with "< 0"
> - in _iomodule.c, there is "== -1"
> - and pygobject/gobject/gobjectmodule.c just does::
> if (...tp_init(...))
> PyErr_Print();
That's not very nice. Would it make sense to unify those
checks, e.g. for "initproc() < 0"? Than the documentation
could be updated.
Zbyszek

Note that all these cases are compatible with "tp_init returns 0 on success and -1 on error".

See attached patch to clarify this in the docs.

New changeset c6dc1e0db7f0 by R David Murray in branch '3.4':
#17380: Document tp_init return value in extending docs.
https://hg.python.org/cpython/rev/c6dc1e0db7f0
New changeset d74ede4bbf81 by R David Murray in branch 'default':
Merge: #17380: Document tp_init return value in extending docs.
https://hg.python.org/cpython/rev/d74ede4bbf81 

Thanks, James.