Array params from object #39
Conversation
…#1081) This function allows code to query an object's "innate" type and shape, without actually creating a NumPy array out of it. PyArray_FromAny has been changed to use this function, as an assurance that its implementation is correct. This also fixes the known failure in test_regression.py.
|
There are enough new interface functions at this point that it might be worth listing them with a short description and posting on the mailing list. Once they go into a release we will need to support them in perpetuity so a bit of discussion before that fateful step might be a good idea. |
|
Committed in 1d952a5. Do you need permissions so that you can close tickets? |
|
True, it's worth getting the API functions right, and I don't think a rigorous review has consistently been done in the past. I personally would reject calling any function PyArray_2, like PyArray_DescrConverter2, and that's why I came up with CanCastTypeTo as the CanCastTo function with casting parameter. The NpyIter_ functions could certainly use some scrutiny as well. The only one I added that I think needs changing is PyArray_FillWithZero. It should become PyArray_FillWithValue, and take a double for the value, producing an error if the value passed can't be down-cast without changing its value. The current way to fill an array with a value from C is really cumbersome. Since we're not rushing into beta at the moment as far as I can tell, there are a few more things I want to add yet. Maybe wait a bit longer before posting for that review? |
|
Regarding the tickets, yes, I don't have permission to close them currently. |
|
OK, I don't know how to handle the trac permissions, or even if I have that authority, so I put in a request. As to the release schedule, it sounds like Ralf is tied up at least until around June, which I suspect is the end of the school year. No one else has volunteered, so I might need to attempt it myself, which is a bit of a learning experience. IIRC, Ralf took about four months to make his first release. Maybe we should ask Ralf to ask for a volunteer, I know that there was at least one other who did some work on that after David Cournapeau moved on.on. |
* ENH: Expose and use n_children_spawned * DOC: docstring formatting and cleanups. * TST: more test coverage of SeedSequence.
Merge in numpy-hpy from ss/array_assing to labs-hpy-port * commit '7fcca730d29549d7f808b49147619d2cf6eda338': Fix: leaking h_desc in array_assign_subscript Partial port of np.ravel Partial port of array_assign_subscript
refactor: Optimize vext_s8
A new API function for being able to determine the innate data type of an input or output PyObject * before actually converting it to an array or target type.
Also fixes a regression and removes a known failure from test_regression.py.