The bigger picture is that CPython's core team simply does not care too much about performance. Performance has never been a fundamental requirement, but merely an afterthought.
The biggest reminder of this was Python 3, which for me was a complete disappointment. They could have limited python's super-dynamic behavior (e.g. changing builtin functions, patching classes on the fly, etc.) or made them optional. They could've added optional typing annotations a la Cython. Or even changing the builtins and language syntax to allow more inplace operations and preallocations, so that temporary results wouldn't have to be allocated on the heap over and over again. All of these changes would have made python faster and more JIT-able. None of these things happened. Performance-wise, python 3 is no step forward.
Python+Cython is still a powerful combination, but eventually Julia or similar languages will eat python's lunch with respect to scientific computing.
The biggest reminder of this was Python 3, which for me was a complete disappointment. They could have limited python's super-dynamic behavior (e.g. changing builtin functions, patching classes on the fly, etc.) or made them optional. They could've added optional typing annotations a la Cython. Or even changing the builtins and language syntax to allow more inplace operations and preallocations, so that temporary results wouldn't have to be allocated on the heap over and over again. All of these changes would have made python faster and more JIT-able. None of these things happened. Performance-wise, python 3 is no step forward.
Python+Cython is still a powerful combination, but eventually Julia or similar languages will eat python's lunch with respect to scientific computing.