BUG: Make timsort deal with zero length elements. #12944
Conversation
The elements of arrays of unicode, string, and void type may have zero length and such types do not need sorting. This fixes a segfault in timsort due to integer division by zero by checking the element size and returning immediately when it is zero.
| size_t len = PyArray_ITEMSIZE(varr); | ||
| PyArray_CompareFunc *cmp = PyArray_DESCR(varr)->f->compare; | ||
| PyArrayObject *arr = varr; | ||
| size_t len = PyArray_ITEMSIZE(arr); |
There was a problem hiding this comment.
Maybe name it elsize so that the definitions match with the above one? The check can maybe be elsize for both for similarity.
But you did not change it, so I guess that is probably already part of your "style cleanup".
There was a problem hiding this comment.
Yeah, I will probably change it later, but there are so many style nits that I want to clean up that they would just get in the way of a review, so I decided to make that a separate, lower priority, task.
There was a problem hiding this comment.
I may actually do two style cleanups, one just for code style, and another doing a refactor. An instance of the latter would be using npy_intp rather than size_t.
|
Ok, just double checked to see that the other sorts handle this the same way. Style fixups are nice, but it isn't like this is very bad in any case. Thanks Chuck! |
The elements of arrays of unicode, string, and void type may have zero
length and such types do not need sorting. This fixes a segfault in
timsort due to integer division by zero by checking the element size and
returning immediately when it is zero.
This is the first of three PRs dealing with timsort. The next will take care of the broken API compatibility, and the last will be a big coding style cleanup.