Array API compliance

[lshaw8317]

The following is a summary of the failing tests in the array-api-tests suite. When you have managed to make the test pass, mark the box next to the file.

There are three main groups of tests: 1 - lazy expression and chunking related things (unique_counts, cumulative_sum etc.) 2 - indexing related things (chiefly implement fancy indexing for .slice method) 3 - improving numexpr (add op_codes such as ge_bcd etc.)

• [ ] test_utility_functions.py : add blosc2.diff function, squeeze needs to be modified to return a view (at C-level, follow expand_dims)
• [ ] test_statistical_functions.py : add cumulative_sum and cumulative_prod
• [ ] test_special_cases.py: everything
• [ ] test_sorting_functions.py: add argsort
• [ ] test_signatures.py : everything
• [ ] test_set_functions.py: add unique_counts, unique_inverse, unique_all, unique_values
• [ ] test_searching_functions.py: add argmax, argmin, where attribute to NDArray, nonzero, count_nonzero
• [ ] test_operators_and_elementwise_functions.py: everything
• [ ] test_manipulation_functions.py: SERIOUS (Fatal Python error: Floating-point exception)
• [ ] test_linalg.py: haven't implemented most things (see #476), which is fine, but some tests also fail because numexpr needs a ge_bcd opcode.
• [x] ~test_inspection_functions: passes all~
• [ ] test_indexing_functions.py: need to improvetake and take_along_axis (latter fails tests)
• [ ] test_has_names : fails 23 tests, need to add functions listed in comment below
• [ ] test_fft.py: skips all tests
• [ ] test_data_type_functions.py: a few different things (11 tests), add broadcast_arrays
• [ ] test_creation_functions.py: add reshape for N-D arrays.
• [x] ~test_constants.py: passes all~
• [ ] test_array_object.py: one test takes too long for some reason sometimes

In order to run the tests, follow these steps:

1. Git clone array-api test suite https://github.com/data-apis/array-api-tests
2. Navigate to array-api-tests and $ pip install -r requirements.txt into environment.
3. Return to home directory, git clone blosc2, navigate to resulting directory and pip install -e . in your environment.
4. Run tests from python-blosc2 directory via ARRAY_API_TESTS_MODULE=blosc2 pytest ../array-api-tests/array_api_tests/test_has_names.py

[lshaw8317]

Seems like the remaining functions we need are:

• [ ] tril
• [ ] triu
• [ ] flip
• [ ] repeat
• [ ] roll
• [ ] tile
• [ ] argmax/argmin
• [ ] searchsorted
• [ ] cumulative_sum/cumulative_prod
• [ ] argsort
• [ ] diff
• [ ] unique_counts/unique_inverse/unique_values/unique_all
• [ ] broadcast_arrays
• [ ] moveaxis
• [ ] unstack
• [ ] from_dlpack/__dlpack__/__dlpack_device__