np.zeros(10, dtype=[]) calls malloc(0) #13645
Comments
|
Closing, fixed by #13691 which both asserts |
|
We probably want to add a test for this somewhere, since that PR does not. |
|
All the code paths in the PR show 100% coverage, implying they are tested, which is why I did not require additional tests before merging. Although one more test could not hurt. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
... and
malloc(0)is implementation-defined. This works fine on my machine, but may throw an exception on other machines.This happens via
PyDataMem_NEW, so we at least have a point we can intercept this. We should probably do one of:PyDataMem_NEW(0)is never called (and perhaps make it fail on all platforms to make this detectable)PyDataMem_NEW(0)to meanPyDataMem_NEW(1), so that it the allocation always succeedsmalloc(0) == NULLperformed inconfigtest, to avoid the cost of the conditional on platforms with the version ofmallocwhich is convenient for us.There are probably a few places where there is a similar problem with
PyDataMem_RENEW(ptr, 0), except that does the wrong thing on all platformsThe text was updated successfully, but these errors were encountered: