Multi-dimensional array slices

[radkoff]

Hi, thanks for the fast response to my last issue. I might have a few more for you this week.

I'm wondering if pythran supports slicing into multi-dimensional arrays with index numpy arrays? This script:

import numpy as np

#pythran export foo(float[][])
def foo(arr):
    arr2 = np.zeros(arr.shape)
    arr2[np.array([1, 2, 3, 4, 5]), np.array([1, 2, 3, 4, 5])] = 1
    return arr2

Is giving me a couple of compilation errors:

/Users/evan/miniforge3/envs/project/lib/python3.9/site-packages/pythran/pythonic/include/types/numpy_gexpr.hpp:549:5: error: static_assert failed due to requirement 'utils::all_of<true, false>::value' "all slices are valid"
    static_assert(
    ^

/var/folders/l5/210k0yvw8xj04j001s6yhlr00000gq/T/tmpj8b8q34j.cpp:103:277: error: no viable overloaded '='
    arr2[pythonic::types::make_tuple(pythonic::numpy::functor::array{}(pythonic::types::make_tuple(1L, 2L, 3L, 4L, 5L), pythonic::numpy::functor::int64{}), pythonic::numpy::functor::array{}(pythonic::types::make_tuple(1L, 2L, 3L, 4L, 5L), pythonic::numpy::functor::int64{}))] = 1L;
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^ ~~

Printing arr2[np.array([1, 2, 3, 4, 5]), np.array([1, 2, 3, 4, 5])] instead of using it for an assignment instead gives the error:

/Users/evan/miniforge3/envs/project/lib/python3.9/site-packages/pythran/pythonic/types/ndarray.hpp:1050:38: error: no type named 'dtype' in '(anonymous namespace)::pythonic::types::numpy_gexpr<(anonymous namespace)::pythonic::types::ndarray<double, (anonymous namespace)::pythonic::types::array_base<long, 2, (anonymous namespace)::pythonic::types::tuple_version>>, (anonymous namespace)::pythonic::types::contiguous_normalized_slice, (anonymous namespace)::pythonic::types::ndarray<long long, (anonymous namespace)::pythonic::types::pshape<std::integral_constant<long, 5>>>>'
    return os << ndarray<typename E::dtype, typename E::shape_t>{e};

I'm using the latest code from master, with compile command:

clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /Users/evan/miniforge3/envs/project/include -arch arm64 -I/Users/evan/miniforge3/envs/project/include -fPIC -O2 -isystem /Users/evan/miniforge3/envs/project/include -arch arm64 -DENABLE_PYTHON_MODULE -D__PYTHRAN__=3 -DPYTHRAN_BLAS_OPENBLAS -I/usr/local/opt/openblas/include -I/Users/evan/miniforge3/envs/project/lib/python3.9/site-packages/pythran -I/Users/evan/miniforge3/envs/project/lib/python3.9/site-packages/numpy/core/include -I/Users/evan/miniforge3/envs/project/include/python3.9 -c /var/folders/l5/210k0yvw8xj04j001s6yhlr00000gq/T/tmpv4d2hdss.cpp -o /var/folders/l5/210k0yvw8xj04j001s6yhlr00000gq/T/tmpbyr82jsz/var/folders/l5/210k0yvw8xj04j001s6yhlr00000gq/T/tmpv4d2hdss.o -std=c++11 -fno-math-errno -Wno-unused-function