view for numpy arrays #987
Conversation
Npyreshape
for more information, see https://pre-commit.ci
* Comment: matching style in file (///), responsibility sentence, consistent punctuation. * Replacing `unsigned char` with `uint8_t` for max consistency. * Removing `1` from `array_view1` because there is only one.
|
I also ran a quick manual leak check to ensure that the It was simply temporarily adding |
There was a problem hiding this comment.
@henryiii, @Skylion007: Looking for a 2nd set of eyes / approval.
include/pybind11/numpy.h
Outdated
| array view(const std::string& dtype) { | ||
| auto& api = detail::npy_api::get(); | ||
| auto new_view = reinterpret_steal<array>(api.PyArray_View_( | ||
| m_ptr, dtype::from_args(pybind11::str(dtype)).release().ptr(), NULL)); | ||
| if (!new_view) { |
There was a problem hiding this comment.
personally prefer leaving braces in,
There was a problem hiding this comment.
Me too, but I think our clang-format config will take them out, that's why I removed them. Let me double check ...
There was a problem hiding this comment.
Sorry it turns out I was wrong (I'm still really confused why I got to believe so strongly that clang-format takes them out).
They are back now. While I was at it, I applied clang-format-diff, except to line 202 in numpy.h, to not make it outstanding there.
(I really hope we'll get to finish establishing pre-commit clang-format sometime soon...)
|
All green, merging! Thanks again @ncullen93 and @Skylion007! |
this function is useful when you have numpy array buffers that need to be reinterpreted as different memory types.
For example, I often get a numpy array buffer from an ITK image that is e.g. size 256x256x4 as uint8 and it needs to be "viewed" as 256x256 float... it'd be nice to do this directly with pybind11.
From numpy:
Not sure if you like the std:string arg input or would rather have py::dtype or something..
Suggested changelog entry:
* Added ``view`` to view arrays with a different datatype.