Skip to content

Latest commit

 

History

History
75 lines (64 loc) · 9.74 KB

PULL_REQUESTS.md

File metadata and controls

75 lines (64 loc) · 9.74 KB
Raw

To close

  • pybind/pybind11#1138 - Inconsistent use of holder classes mess. Seems superseded by #1161. Either this or #1161 should get a more detailed review, before categorizing.
  • pybind/pybind11#1286 - Pybind GIL API is incompatible with the raw CPython API because they keep track of python state separately. This is a long and contrieved way of making them play nice. My opinion: close as it is a strange way to make these cooperate. Better solution for those that need CPython API is to have a basic pybind class without the "advance GIL use case".
  • pybind/pybind11#1322 - Look at #1286

Fixes

Minor

  • pybind/pybind11#1043 - Something about propagation of return policies. Removes 4 lines, should be safe, but there's a possibility that tests just don't cover what those 4 lines are doing.
  • pybind/pybind11#1089 - Cmake C++ standard version settings. Turned into bikeshedding about cmake versions and whether cmake 3.3 is too new (in 2017). Good candidate for merge.
  • pybind/pybind11#1101 - Slicing with None. Simple enough, but blocked on repetition of code because cast() is not available and @henryiii claims forward declaration to be tricky.
  • pybind/pybind11#1192 - Allows std::enable_shared_from_this to be used for casting and lets pybind avoid a copy in some cases.
  • pybind/pybind11#1210 - Moves optional and variant casters into their own header. Backwards compatible. Bikeshedding made it go stale. Should be merged as is and the bikeshedding discussion, that could be proven useful, taken to a separate issue.
  • pybind/pybind11#1260 - R-value references support for py::cast()
  • pybind/pybind11#1394 - Signed slices
  • pybind/pybind11#1413 - Allow builtins to be used as callbacks
  • pybind/pybind11#1466 - Similar idea to memory views, but with a greater scope. Adds readonly flag to the buffer protocol. The py::array and others (including memory views) would be separate PRs. Looks ready, but should be looked at again.
  • pybind/pybind11#1496 - py::bind_vector can only be extended by another object of the same type, so string_vector.extend(['asd']) doesn't work, but string_vector.extend(StringVector(['asd'])) does. One of the official maintainers called this "an overlooked feature".
  • pybind/pybind11#1767 - Adds __contains__ to py::bind_map for better performance. Without it operator in falls back to __iter__.

Major

  • pybind/pybind11#1131 - Override overload order with py::prepend(). Jakob raised a concern about performance, but without backing it with numbers. Otherwise, no complaints and ready to be merged.
  • pybind/pybind11#1196 - Keyword-only arguments support.

Unfinished

  • pybind/pybind11#781 - Far from finished. Seems very useful. I tried finishing it, but got stuck. Having a "lock into py::class_ after .def" makes it hard to get done.
  • pybind/pybind11#1209 - An attempt to fix a core dump when a C++ thread running a python script gets terminated. Uses some GCC specific black magic, so I don't know what it is about. Tests are needed.
  • pybind/pybind11#1220 - Correctly document that pybind requires cmake 3.0
  • pybind/pybind11#1243 - Nicer error instead of std::bad_alloc in one specific case. Lacks tests.
  • pybind/pybind11#1325 - Allows custom numpy ufuncs to be defined. Work in progress. Stale.
  • pybind/pybind11#1368 - Fixes a segfault with iostream wrappers and releasing GIL.
  • pybind/pybind11#1428 - "Generalize cmake support" was greenlighted by @jakob, but was missing the update to the docs.
  • pybind/pybind11#1463 - py::isinstance<py::str> is too permissive. BUgfix is a WIP.
  • pybind/pybind11#1648 - Minor feature, backwards compatible, but lacks tests.
  • pybind/pybind11#1651 - Implements py::with according to PEP-343, needs to be rebased, otherwise, ready.
  • pybind/pybind11#1687 - Makes py::pythonbuf contain a std::unique_ptr<char[]> instead of char[1024] so that users can set their desired buffer size. No tests. Motivation not compeling. In "Unfinished" category because of the lack of tests and the posibility to be proven wrong about the motivation.
  • pybind/pybind11#1703 - Support for compiling with -fno-exceptions. Motivation: Google wants to use pybind11. Unfinished because I left a few comments that were not addressed.
  • pybind/pybind11#1769 - Interesting use case with two extension modules where one module throws and the other catches the exception. The problem lies in internal/external linkage of some pybind11 types.

To review

  • pybind/pybind11#1098 - Cmake compiler features, by @henryiii. I haven't given it a proper review.
  • pybind/pybind11#1122 - Construct py::enum_ from std::string. A simple feature turned into a discussion about case sensitivity. I'm questioning the usefulness of the PR.
  • pybind/pybind11#1161 - Look at #1138

Issue reproducer