Support pointer-to-member-function in def_buffer

[bmerry]

Closes #857. Instead of calling the function object directly, it wraps
it in std::bind and then immediately calls that. This is a poor-man's
form of std::invoke (not available in C++11 or C++14).

[dean0x7d]

The inclusion of <functional> in the main pybind11.h header may be a compilation time issue. At least I assume so, given that pybind11/functional.h is an optional extra and not part of the core headers. cc @wjakob

[jagerman]

The use of <functional> is probably a non-starter (not including the rather heavy-weight <functional> in core pybind was a design decision from way back).

An overload that accepts a member function pointer and converts it into lambda, on the other hand, could work, and shouldn't be noticeably more code.

Also, tests_buffers seems like a better place than test_issues. The latter is quasi-deprecated, with a vague long-term plan to move everything out of it into more appropriate places, and to at least not add anything more into it.

[bmerry]

Ok, I'll give it another go. I guess the simplest thing then would be to provide an overload of def_buffer that takes any PTMF, wraps it in a lambda, and passes it to the original overload?

[bmerry]

I've reworked it and changed the PR title to reflect the new version. I haven't flattened the history at this point.

Do you think it is also worth adding an overload for pointer to data member (so that the class can initialise a buffer_info member when constructed and then we just pass a pointer to that member to def_buffer?

[bmerry]

The pypy build is crashing. Is that expected due to https://bitbucket.org/pypy/pypy/issues/2444 (the other to_python test in the file is disabled because of that, but I don't want to do the same here until someone can confirm that I'm not just hiding a real bug, particularly since that pypy bug seems into indicate a leak rather than a double-free.

[dean0x7d]

Should be OK to skip the test on PyPy. Looks like the same issue as the existing test. The rest looks good to me. I've just noted a minor brace style issue.

[dean0x7d]

Brace style: same line.

[dean0x7d]

Same.

[bmerry]

I've just noted a minor brace style issue.

Fixed, thanks.

[jagerman]

Looks good to me.

[jagerman]

Can you squash the changes and update the commit messages into a single commit?

[bmerry]

Sure. Should I do that as a force-push to the branch?

[jagerman]

Yup.

[bmerry]

Done, and also added the marker to disable the new test in PyPy. Once Travis and Appveyor are happy I think it should be good to go.

[jagerman]

Looks like the last appveyor test got stuck, but all the rest passed so this is good to go.