Mailman 3 September 2015 - Cplusplus-sig

Call policy to store reference
by Bruce Merry Sept. 27, 2015

Sept. 27, 2015

Hi I don't know if something like this has been discussed before, but I thought I'd contribute it in case it is useful. I found that with_custodian_and_ward didn't suit my needs, because (a) the association it creates is permanent, even if the underlying association is later broken, leading to handle leaks; (b) I had some issues with destructor ordering when the interpreter exits (sorry, I don't recall the details). I've created an alternative call policy that requires a wrapper custodian class with a handle<> data member, which is set to a reference to the ward object. This is suitable for wrapping functions that set an underlying C++ reference in the custodian, replacing any previous reference rather than adding to it. The code is below - I've only written a _postcall version, but it should be possible to write a non-postcall version too. At the moment it is part of a GPLv3+ project, but I'm working on getting permission from my employer to release this code under the Boost License as well, to make it easier to use in other Boost.Python-based projects (or to incorporate into Boost.Python itself). You can see it in action in spead2 (e.g. https://github.com/ska-sa/spead2/blob/v0.3.0/src/py_send.cpp). template<typename T, boost::python::handle<> T::*handle_ptr, std::size_t custodian, std::size_t ward, class BasePolicy_ = boost::python::default_call_policies> struct store_handle_postcall : BasePolicy_ { static_assert(custodian != ward, "object must not hold reference to itself"); static PyObject* postcall(PyObject *args, PyObject *result) { std::size_t arity = PyTuple_GET_SIZE(args); if (custodian > arity || ward > arity) { PyErr_SetString(PyExc_IndexError, "store_handle_postcall: argument index out of range"); return nullptr; } result = BasePolicy_::postcall(args, result); if (result == nullptr) return nullptr; PyObject *owner = custodian > 0 ? PyTuple_GET_ITEM(args, custodian - 1) : result; PyObject *child = ward > 0 ? PyTuple_GET_ITEM(args, ward - 1) : result; boost::python::extract<T &> extractor(owner); try { T &target = extractor(); target.*handle_ptr = boost::python::handle<>(boost::python::borrowed(child)); } catch (boost::python::error_already_set) { Py_XDECREF(result); return nullptr; } return result; } }; Bruce -- Bruce Merry Senior Science Processing Developer SKA South Africa

1 0

Magic boost conversion unicorn hook?
by Ken Seehart (kseehart) [CONT - Type 2] Sept. 26, 2015

Sept. 26, 2015

I have a python class that wraps a boost.python object. I am using a boost.python library function that requires the raw boost.python object. I want to be able to pass the python object to the boost api function. Is there a hook available on the python side that instructs boost to perform a conversion? Hypothetically, it would be something like this: class EggWrapper(object): def __init__(self, egg): object.__setattr__(self, '_egg', egg) def __getattribute__(self, name): egg = object.__getattribute__(self, '_egg') if name=='_egg': return egg if name in EggWrapper.__dict__: return object.__getattribute__(self, name) return getattr(egg, name) def __setattr__(self, name, value): egg = object.__getattribute__(self, '_egg') setattr(egg, name, value) def magic_boost_conversion_unicorn_hook(self): 'this is automatically called by boost to convert arguments' egg = object.__getattribute__(self, '_egg') return egg import myboostlib egg = EggWrapper(myboostlib.getEgg()) myboostlib.spam(egg) Where the signature for spam is: spam(class boost::python::api::object) And myboostlib.getEgg() returns a boost::python::api::object - I do not have the ability to modify the library or any C/C++ code, the solution must be entirely on the python side. - I can't change the calling convention. The wrapped object must be used (i.e. I can't just say myboostlib.spam(egg._egg), though that is the desired result). So the solution should be incorporated into the EggWrapper class. - I'm only interested in solutions that reflect expert knowledge of Boost.Python, since I believe I have already considered all the trivial solutions. So my question is, does magic_boost_conversion_unicorn_hook() exist, and if so, how is it spelled? Thanks, Ken

3 2

formatting str of exposed type
by Daniel Brake Sept. 18, 2015

Sept. 18, 2015

Hi Boost Python, I would like a C++ class which I am exposing through Boost.Python (initially,the boost::multiprecision::mpfr_float, with expression templates turned off -- eventually a whole pile of types depending on mpfr_float) to be able to respond to the %precision n command, for example, from ipython. Additionally, I would like to be able to use a format spec such as print('{0:.40}'.format(b)) to print a variable b with 40 digits of precision. Even better, I would like to be able to put a standard formatting letter in the format string preceding the .format call. My best guess would be to def("format",...) as a free function for my class_, but I also guess that this is a standard operation, and I would rather use a standard solution. Parsing the formatting string myself doesn't sound like much fun, and am hoping for a little insight from the mailing list. Thanks very much, Daniel Brake University of Notre Dame

3 2

boost python exposing const*
by MM Sept. 9, 2015

Sept. 9, 2015

I have a c++ class T that I succesfully expose via: class_<T, T*> ... later on, I have struct S { const T* underlying; }; class_<S,S*>("S"). .def_readonly("underlying", &S::underlying) ; >From python, when I call: s.underlying where s is a python instance of S, i get the following error: TypeError: No to_python (by-value) converter found for C++ type: T const* Do I need another converter for T const* in particular? or for both T* and T const*? Does the class_<T,T*> above only expose T itself, not pointers to it? Thanks MM

2 1

Why isn't python-exposed boost::gregorian::date available everywhere?
by MM Sept. 8, 2015

Sept. 8, 2015

I thought I'd post this SO question here as well, thanks: I exposed boost::gregorian::date with the following: date_from_python_date{}; to_python_converter<date, date_to_python_date, true>{}; where *date_to_python_date* is a struct with the right convert function: it converts it to a python datetime.date. Some c++ functions return date and calling them from python works. At a later stage, I have a c++ class class F {/// ....public: boost::gregorian::date start;}; which I register with: class_<F, F*, bases<B>>("F") .def_readwrite("start", &F::start, "FTD"); I do this after having python-registered date. I then obtain an instance f of the F wrapper. But then, when I print f.start The error is: No Python class registered for C++ class boost::gregorian::date

1 0