Commits

Author Commit Message Labels Comments Date
ariovistus
cleaned deimos headers of any non-links-to-python stuff. this is a rather half-baked commit, but deimos headers work!!!
ariovistus
deimos headers compile!
ariovistus
improved speed of preprocessor function. hopefully fast enough now
ariovistus
make a deimos-style python header directory. folded in changes from python 2.7 - 3.2. still need to work on diffs of 3.0, 3.1 to 3.2 found many functions that weren't in python.d (and still aren't) plan is to ddoc as many of the python functions as possible.
ariovistus
update d <-> python string conversion to actually use utf encodings. unicode is the python string type of choice now. improved exception messages (ex's __unicode__ tends to know what to do better than I do) python.d was not properly linking some PyUnicode functions, so I wrote a function to simulate the C preprocessor. It's really slow.
ariovistus
merged something
ariovistus
added d -> array.array conversion. moved numpy.ndarray conversion to pyd.extra.
ariovistus
added d -> python for numpy.ndarray. maybe doesn't belong, but it wasn't too complicated.
ariovistus
conversion d ranges <-> python iterators now supported required wrapping a type. so now module 'pyd' exists. come to think of it, this will be murder on multiple extension modules. oh well. Py_XINCREF isn't supposed to throw for non borrowed types changed py_def to take function type rather than return type and arg types
ariovistus
renamed conversion functions. d_type is now python_to_d _py is now d_to_python PydObject.toDItem is now PydObject.to_d python_to_d is now ex_python_to_d d_to_python is now ex_d_to_python parrot is now ex_parrot now enforcing naming convention on pyd: * public facing functions are named like this: foo_bar * types and templates are named like this: FooBar changed…
ariovistus
added some unittests for bytearray, which is a buffer.
ariovistus
New style buffer interface is supported! we have a wrapper around the interface in PydObject.bufferview we have support for for python -> d conversion, for nd arrays (still have to deal with simple buffers) this translates to (hopefully) much faster conversion from e.g. numpy.ndarray to d arrays, but more restrictions on what types can be converted, e.g. numpy ieee754 binary64 matrix to int[][] is not allowed. trying out scons for…
ariovistus
hello scons
ariovistus
doh! Py_buffer got a new field in 2.7 improve make_object docs add some unittests for python_to_d, d_to_python
ariovistus
properly document Borrowed
ariovistus
fixes for embedded * PyDef doesn't pollute its context with python function * improve docs apparently puzzle didn't get in last commit. oops.
ariovistus
hello, puzzle why is buffer.x segfaulting?
ariovistus
more improvements to d_type * fix d_type for null arrays (which aren't really null) * implemented PydInputRange * fix PyEval, PyStmts: don't need to give a module for builtins * replace PyObject_BorrowedRef with template * fix PyEval_GetBuiltins and friends: they return borrowed refs!
ariovistus
improvements to d_type * fix d_type failure to convert python int objects * improve d_type handling of d arrays * add special case to d_type for python's array.array -todo: finish special case for new style buffer interface * let could_not_convert display reason for failure. * some fixes to python.d * added arrayobject to python.d (it doesn't really belong in there, but it probably isn't going to change)
ariovistus
conversion to/fro BigInt now handled. poorly handled: uses string intermediate representation. BigInt won't let you touch its bits. Oh well.
ariovistus
added PyLongObject, which was needed by _PyLong_AsByteArray. python objects can conceivably define their own clear() member functions, so pydobject shouldn't make life difficult if they do.
ariovistus
add _PyLong_AsByteArray and some others.
ariovistus
signature overhaul. * def, Def, StaticDef, Property used to take many parameters most of which were strings and tried to shoehorn as many as possible to be optional via positional default arguments and lots of overloads. That changes today. Now: 1 signature, all parameters optional, poor man's keyword arguments. * class_wrap and struct_wrap get similar treatment * in unrelated news, struct_wrap now works * often, docstring is provided as a runtime argument. I assume this is more shoehorning; I can't see any advantage to such late specification, so now it and modulename are compile time arguments property annotated functions: what a nuisance. * Property can handle them now. * Len can handle them now. * Repr can handle them now. Member works! Todo: should we scrap all these dedicated operator overloads in favor of enabling Def!(Fn, "__op__") ? boost::python seems to roll this way
ariovistus
leave candydoc generation for the web repo
ariovistus
hey, candydoc is nice. Add some doc to code merge dg_convert and lib_abstract into func_wrap got rid of some cruft duplicated in doc/ or the bitbucket wiki.
ariovistus
commit some files
ariovistus
split up python.d, deimos style hello issue 7758
ariovistus
can dmd link to data? yes it can!
ariovistus
Can we link to data? Yes we can!
ariovistus
typo fix
  1. Prev
  2. Next