feat(Init): Initial upload of array + interoperability package by Simpag · Pull Request #1 · team-decent/decent-array

Simpag · 2026-05-10T18:22:51Z

Initial upload of array + interoperability modules, including tests, docs, type-checking and compilation. Better docs and a README is needed.

Copilot

Pull request overview

Initial import of the decent_array package, providing an Array wrapper and a single-active-backend interoperability layer with NumPy/PyTorch/JAX/TensorFlow backends, plus CI, docs, benchmarks, and a comprehensive pytest suite.

Changes:

Added backend manager + abstract backend contract, plus concrete backends (NumPy/PyTorch/JAX/TensorFlow) and module-level iop functions/RNG coordination.
Added Array wrapper implementing operators, indexing, and common properties via the active backend.
Added tox/CI/Sphinx/ReadTheDocs scaffolding, benchmarks, and extensive tests.

Reviewed changes

Copilot reviewed 49 out of 52 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
tests/test_iop_rng.py	Tests for RNG seeding, snapshot/restore, and distribution helpers.
tests/test_iop_functions.py	Tests for module-level interoperability functions (creation, math, linalg, indexing).
tests/test_decorators.py	Tests for the cost-method autodecorator argument unwrapping/rewrapping behavior.
tests/test_backend_manager.py	Tests backend activation, registration, listener behavior, and reset semantics.
tests/test_array.py	Tests `Array` operators/dunders, indexing, coercion, and properties across backends.
tests/conftest.py	Parametrized backend/device fixture with skip logic for unavailable combinations.
readthedocs.yaml	Read the Docs build configuration.
pyproject.toml	Project metadata, dependencies/extras, tox envs, ruff/mypy/pytest config, mypyc build hook.
LICENSE	Adds AGPL-3.0 license text.
docs/sphinx_theme.txt	Sphinx theme requirements list.
docs/source/user.rst	Initial user guide stub + installation snippet.
docs/source/index.rst	Sphinx landing page and toctree.
docs/source/developer.rst	Developer guide covering tox workflows, mypyc usage, and contribution process.
docs/source/conf.py	Sphinx configuration (theme, intersphinx, missing-reference fixups, sidebar config).
docs/source/background.rst	Background page stub.
docs/source/author.rst	Contributors page.
docs/source/api/decent_array.types.rst	API stub for `decent_array.types`.
docs/source/api/decent_array.rst	API toctree root.
docs/source/api/decent_array.interoperability.rst	API stub for `decent_array.interoperability`.
docs/source/api/decent_array.array.rst	API stub for `decent_array` (exports `Array`).
docs/source/_templates/package.rst.jinja	Custom sphinx-apidoc package template.
docs/source/_templates/module.rst.jinja	Custom sphinx-apidoc module template.
docs/source/_static/custom.css	Small theme CSS tweak for logo sizing.
docs/Makefile	Sphinx Makefile (unix).
docs/make.bat	Sphinx make script (windows).
decent_array/types.py	Shared type aliases + supported framework/device enums.
decent_array/interoperability/_tensorflow/tensorflow_backend.py	TensorFlow backend implementation + registration side-effect.
decent_array/interoperability/_tensorflow/init.py	TensorFlow backend package initializer/export.
decent_array/interoperability/_pytorch/pytorch_backend.py	PyTorch backend implementation + registration side-effect.
decent_array/interoperability/_pytorch/init.py	PyTorch backend package initializer/export.
decent_array/interoperability/_numpy/numpy_backend.py	NumPy backend implementation + registration side-effect.
decent_array/interoperability/_numpy/init.py	NumPy backend package initializer/export.
decent_array/interoperability/_jax/jax_backend.py	JAX backend implementation + registration side-effect.
decent_array/interoperability/_jax/init.py	JAX backend package initializer/export.
decent_array/interoperability/_iop/rng.py	Cross-backend RNG coordination (python `random`, numpy, backend) + distributions.
decent_array/interoperability/_iop/functions.py	Module-level iop function surface delegating to the active backend.
decent_array/interoperability/_iop/init.py	iop package marker/init (empty).
decent_array/interoperability/_decorators.py	Decorator to unwrap `Array` args for backend-native cost-method implementations.
decent_array/interoperability/_backend_manager.py	Backend registry/activation/reset + auto-import and listener mechanism.
decent_array/interoperability/_abstracts/backend.py	Abstract `Backend` API contract (creation/manipulation/linalg/math/RNG).
decent_array/interoperability/_abstracts/init.py	Abstracts package export (`Backend`).
decent_array/interoperability/init.py	Public interoperability API exports (functions + RNG helpers + decorator).
decent_array/_array.py	`Array` wrapper implementation (operators, indexing, properties) tied to active backend.
decent_array/init.py	Package exports (`Array`, `interoperability`, `types`).
benchmarks/profile_hotpath.py	cProfile-based hot-path profiler for wrapper vs function dispatch.
benchmarks/bench_iop.py	Microbenchmark for iop function-call overhead vs native frameworks.
benchmarks/bench_common.py	Shared benchmark helpers (backend discovery/activation + timing utilities).
benchmarks/bench_array.py	Microbenchmark for `Array` operator overhead vs native frameworks.
.gitignore	Python/build/dev artifact ignores.
.github/workflows/ci.yaml	CI pipeline running tox envs (mypy/pytest/ruff/sphinx/mypyc) on Py3.13 across OSes.
.github/CODEOWNERS	Sets default code owners for the repo.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+[project]
+name = "decent-array"
+version = "0.1.0"
+authors = [{name = "Elias Ram"}, {name = "Simon Granström"}, {name = "Adriana Rodriguez"}, {name = "Nicola Bastianello"}]
+maintainers = [{name = "Team Decent"}]
+description = "A library of array operations and linear algebra primitives for interoperability across ML frameworks."
+readme = "README.md"
+requires-python = ">=3.13"
+classifiers = [
+  "Programming Language :: Python :: 3.13",
+  "Operating System :: OS Independent",
+]
+license = "AGPL-3.0-only"
+dependencies = [
+  "numpy",
+]


Simpag · 2026-05-10T21:31:51Z

I will try to remember to upload benchmark results tomorrow evening

nicola-bastianello

first batch of comments, mostly on docstrings and docs. a few of the comments on adding stuff to docs we can copy-paste to a new issue and address them later

nicola-bastianello · 2026-05-14T13:40:32Z

@@ -0,0 +1,71 @@
+"""


it would be nice to add instructions on how to run the benchmarks and interpret the results; maybe we add this when preparing the docs

nicola-bastianello · 2026-05-14T13:41:29Z

+"""
+Abstract :class:`Backend` contract.
+
+All abstract methods live in this single class rather than across six mixin ABCs. The


this intro can be edited to remove mention of mixin, and the rest of the discussion should be included in/moved to the docs

nicola-bastianello · 2026-05-14T13:43:22Z

@@ -0,0 +1,301 @@
+"""
+JAX backend for interoperability_2.


some mentions of interoperability_2 still present

nicola-bastianello · 2026-05-14T13:44:33Z

+package can be auto-loaded on the first ``set_backend("tensorflow")`` call.
+
+TF eager Tensors are immutable, so :meth:`set_item` round-trips through numpy and the
+in-place math operations rebind the wrapper's underlying value.


a discussion on this should be included in the docs; let's open an issue about docs to keep track of what we should include

nicola-bastianello · 2026-05-14T13:48:12Z

+
+    Args:
+        backend: A :class:`~decent_array.types.SupportedFrameworks` value, its canonical string (e.g.
+            ``"numpy"``, ``"pytorch"``), or any alias declared by the backend at


mention of alias should be removed

nicola-bastianello · 2026-05-14T14:11:51Z

+decent\_array
+=============
+
+.. toctree::


I would change the order: array, interoperability, types

nicola-bastianello · 2026-05-14T14:13:30Z

+============
+
+decent-array is developed by `Simon Granström <https://github.com/Simpag/>`_ and 
+`Adriana Rodriguez <https://github.com/adrianardv/>`_, under the supervision of 


if Adriana has not directly contributed, we can remove her name for now. here I'd keep only people who have committed

nicola-bastianello · 2026-05-14T14:22:49Z

+
+
+
+Compiled hot path (mypyc)


when running tox -e mypyc for the first time I got error: Microsoft Visual C++ 14.0 or greater is required. Get it with "Microsoft C++ Build Tools": https://visualstudio.microsoft.com/visual-cpp-build-tools/

I think here we should mention this possibility, and how to solve it for windows, I don't know about linux/max: 1. go to that website, 2. click download build tools, run the .exe, 3. select "C++ build tools" and install. this is going to take a while and install >6GB of stuff (not ideal, but it looks like there's no way around this storage footprint)

Yes, you need a C compiler to make this work. Can add it to the docs

nicola-bastianello · 2026-05-14T14:23:55Z

@@ -0,0 +1,12 @@
+User Guide
+==========
+This user guide shows you different examples of how to use decent-array.


sentence can be removed for now maybe? and we add to the docs todo list adding examples here, but I'd say it's lower priority

All docs are pure boilerplate. I copied them from decent-bench

nicola-bastianello · 2026-05-14T14:53:18Z

@@ -0,0 +1,7 @@
+decent\_array.interoperability


we should probably find a way to divide the interop api in different pages, each collecting a category of functions (math, linalg, rng, etc). we can discuss the best division

nicola-bastianello

second batch of comments on repo set-up

nicola-bastianello · 2026-05-14T15:00:30Z

@@ -0,0 +1,11 @@
+**/__pycache__


maybe we should add .pyd?

on my system at least they are not ignored by vscode source control

I did not get those on linux, I'll add more python metadata to the ignore list

nicola-bastianello · 2026-05-14T15:00:50Z

+[project]
+name = "decent-array"
+version = "0.1.0"
+authors = [{name = "Elias Ram"}, {name = "Simon Granström"}, {name = "Adriana Rodriguez"}, {name = "Nicola Bastianello"}]


to be changed to you and me only

nicola-bastianello · 2026-05-14T15:04:27Z

+Issues = "https://github.com/team-decent/decent-array/issues"
+
+[project.optional-dependencies]
+dev = [


I'm not sure about the division of optional dependencies here; maybe we should have separate items for torch, tensorflow and jax? or also put torch in dev-cpu and dev-gpu instead of in dev (to be "democratic")

also, "scipy-stubs", "types-networkx", "types-tabulate", might not be needed

the dev, dev-cpu and dev-gpu are interconnected. A dev environment will have dev + dev-cpu or dev + dev-gpu. The split between dev-cpu and dev-gpu is only performed because they require different sub-packages to be installed properly. Torch can install on cpu only devices and gpu devices using the same method. We could move torch to dev-cpu but we'd have to force torch to only install cpu components.

nicola-bastianello · 2026-05-14T15:08:39Z

+mypy-args = ["--ignore-missing-imports"]
+
+[tool.tox]
+envlist = ["dev", "mypy", "pytest", "ruff", "sphinx"]


during my own dev work, I've found that I like having time-consuming checks later; maybe we could change this order to: dev, ruff, mypy, pytest, sphinx? this way if ruff and mypy fail early, one can cancel execution of the checks, instead of waiting for pytest to finish before seeing if ruff is failing

of course one can argue about priorities, that pytest is more important.. let's discuss

I know that one can run the checks individually, but sometimes I'm overly optimistic and run tox directly thinking all checks will be green

The full tox command should generally only be performed right before commit, we might want to add a lint environment that calls ruff + mypy

nicola-bastianello

third batch of comments, this time on aligning to the array api

some of the signatures proposed by array api are a bit more complicated than what we currently have. so for now we could just make small changes (like introducing / where needed), and leave bigger changes on the type annotations for later

nicola-bastianello · 2026-05-14T15:20:24Z

+        return self.transpose
+
+    @property
+    def any(self) -> bool:


all and any are not properties in the array api, I would remove them. transpose should be exposed only as T

https://data-apis.org/array-api/latest/API_specification/array_object.html#attributes

All and any are properties of numpy, torch and jax. I tend to use them a lot because its convenient. Only tensorflow does not have it because of its unique way to handle arrays.

All these frameworks also expose T + transpose. While I don’t care mind removing it, I tend to use .transpose for readability over .T

nicola-bastianello · 2026-05-14T15:24:45Z

+        """Return the sum of the array and another array or a scalar."""
+        return Array(self.value + (other.value if type(other) is Array else other))
+
+    def __radd__(self, other: float) -> Array:


as per array api: The results of applying reflected operators must match their non-reflected equivalents. https://data-apis.org/array-api/latest/API_specification/array_object.html#reflected-operators

this means that type of other should be expanded to include Array

Its not needed because if we add two arrays the __add__ will be called over __radd__. The other argument will never be of type Array in __radd__ so I'm not sure what they are talking about, might be poor wording on their end

nicola-bastianello · 2026-05-14T15:25:31Z

+        """Return the sum of the array and a scalar."""
+        return Array(other + self.value)
+
+    def __sub__(self, other: Array | float) -> Array:


array.__sub__(other: int | float | complex | array, /) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.__sub__.html#array_api.array.__sub__

According to PEP 484, float encapsulates int so int | float is redundant (our linting also highlights this). We could discuss if we want to support complex numbers or not. I believe that it is safe so we could switch, then complex would include int and float so no need to explicitly type this.

nicola-bastianello · 2026-05-14T15:26:20Z

+        """Return the subtraction of the array from a scalar."""
+        return Array(other - self.value)
+
+    def __mul__(self, other: Array | float) -> Array:


array.__mul__(other: int | float | complex | array, /) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.__mul__.html#array_api.array.__mul__

nicola-bastianello · 2026-05-14T15:26:52Z

+        """Return the product of the array and a scalar."""
+        return Array(other * self.value)
+
+    def __truediv__(self, other: Array | float) -> Array:


array.__truediv__(other: int | float | complex | array, /) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.__truediv__.html#array_api.array.__truediv__

nicola-bastianello · 2026-05-14T15:35:47Z

+        """Element-wise greater-than."""
+        return Array(self.value > (other.value if type(other) is Array else other))
+
+    def __ge__(self, other: Array | float) -> Array:


array.__ge__(other: int | float | array, /) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.__ge__.html#array_api.array.__ge__

nicola-bastianello · 2026-05-14T15:37:11Z

+    # which fails TF's ``1 & bool_tensor`` rejection. Native operator semantics on
+    # the wrapped tensor enforce the actual dtype contract.
+
+    def __and__(self, other: Array | int) -> Array:


array.__and__(other: int | bool | array, /) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.__and__.html#array_api.array.__and__

nicola-bastianello · 2026-05-14T15:39:26Z

+
+    # Indexing -------------------------------------------------------------
+
+    def __getitem__(self, key: ArrayKey) -> Array:


I've tried more complex key types but it very easily collapses or doesn’t compile

nicola-bastianello · 2026-05-14T15:40:24Z

+        """Return the item at ``key``."""
+        return self._backend.get_item(self, key)
+
+    def __setitem__(self, key: ArrayKey, value: Array | float) -> None:


nicola-bastianello · 2026-05-14T15:42:50Z

+
+    # Containers / iteration ----------------------------------------------
+
+    def __len__(self) -> int:


__len__ is not in the array api, and I couldn't find an explanation. since all backeneds allow for this operation, let's keep it

nicola-bastianello

other comments, on private vs. public

nicola-bastianello · 2026-05-15T08:24:32Z

+            listener(_BACKEND_INSTANCE)
+
+
+def register_backend_listener(listener: Callable[[Backend | None], None]) -> None:


register_backend_listener maybe should be private? I'm not sure I see a situation where a user would need this function

These are not exposed to the public api but are used within the codebase which is why they are not explicitly "private". Generally private methods are only used within the same script, not cross modules

nicola-bastianello · 2026-05-15T08:25:48Z

+    _COORDINATOR.set_rng_state(state)
+
+
+def derive_seed() -> int:


derive_seed should maybe be private? I think it would just be confusing for users.

but if we decide to keep it public, then the docs should explain in some detail how the seed is derived

derive seed is needed for users to create their own rng system in a "deterministic random" way when a seed is set using iop.set_seed. For example in cost functions, datasets or schemes

nicola-bastianello · 2026-05-15T08:27:46Z

+    def __init__(self) -> None:
+        self._global_seed: int | None = None
+
+    def set_seed(self, seed: int, *, set_global_seed: bool = True) -> None:


set_global_seed is an option that might be confusing to users. we could consider spinning off a _set_local_seed function and remove the option from set_seed

No methods in the _RngCoordinator are exposed. See set_seed futher down

nicola-bastianello · 2026-05-15T08:30:53Z

+Run with::
+
+    python benchmarks/bench_array.py
+"""


it would be nice to have benchmarks that directly compare compiled and uncompiled versions; or at least have an option at the top to say which version should be used

in any case all the benchmarks look good to me. I run them and the results of compilation are fantastic!

The benchmarks print if they are compiled or not. I dont think there's an easy way to compare them in the same script. Could possibly do it using sub-processes but it would be rather intrusive because we'd need to delete compile files

nicola-bastianello

comments on the iop api

my priority for now is to ensure compatibility with the array api. this means changing the names and signatures of many functions, but at this stage we don't need to change the behavior (e.g. we can ignore arguments that are not currently supported). my idea is that aligning the api right now is easier than doing it later, and that changing behavior is easier to do at a later time

nicola-bastianello · 2026-05-15T08:41:18Z

+# Array creation
+
+
+def zeros(shape: tuple[int, ...]) -> Array:


array api suggests the following signature: zeros(shape: int | Tuple[int, ...], *, dtype: dtype | None = None, device: device | None = None) -> array https://data-apis.org/array-api/latest/API_specification/generated/array_api.zeros.html#array_api.zeros

they make device a kwarg; in the current setup, however, device choice is part of the backend. some options: 1) we avoid an explicit device kwarg, 2) we introduce it with default None and always ignore it (like numpy does), 3) we implement creating/moving arrays between different devices. I'm leaning towards 2), which would allow us to go for 3) in the future if we think it's interesting

in any case, I think we should type as int | Tuple[int, ...] like they do and expose dtype if possible

same discussion applies to zeros_like, ones, ones_like, eye, etc..

I dont see a reason to expose the device param if we want to keep it simple. It could easily cause device conflicts and users would have to think about device management

nicola-bastianello · 2026-05-15T08:41:52Z

+    return _BACKEND_INSTANCE.zeros(shape)
+
+
+def zeros_like(array: Array) -> Array:


zeros_like(x: array, /, *, dtype: dtype | None = None, device: device | None = None) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.zeros_like.html#array_api.zeros_like

nicola-bastianello · 2026-05-15T08:44:17Z

+    return _BACKEND_INSTANCE.zeros_like(array)
+
+
+def ones(shape: tuple[int, ...]) -> Array:


ones(shape: int | Tuple[int, ...], *, dtype: dtype | None = None, device: device | None = None) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.ones.html#array_api.ones

nicola-bastianello · 2026-05-15T08:44:40Z

+    return _BACKEND_INSTANCE.ones(shape)
+
+
+def ones_like(array: Array) -> Array:


ones_like(x: array, /, *, dtype: dtype | None = None, device: device | None = None) → array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.ones_like.html#array_api.ones_like

nicola-bastianello · 2026-05-15T08:45:15Z

+    return _BACKEND_INSTANCE.ones_like(array)
+
+
+def eye(n: int) -> Array:


eye(n_rows: int, n_cols: int | None = None, /, *, k: int = 0, dtype: dtype | None = None, device: device | None = None) -> array

https://data-apis.org/array-api/latest/API_specification/generated/array_api.eye.html#array_api.eye

nicola-bastianello · 2026-05-15T09:48:50Z

+    return _BACKEND_INSTANCE.sign(array)
+
+
+def maximum(array1: Array | float, array2: Array | float) -> Array:


maximum(x1: array | int | float, x2: array | int | float, /) -> array https://data-apis.org/array-api/latest/API_specification/generated/array_api.maximum.html#array_api.maximum

nicola-bastianello · 2026-05-15T09:49:12Z

+    return _BACKEND_INSTANCE.maximum(array1, array2)
+
+
+def argmax(array: Array, axis: int | None = None, keepdims: bool = False) -> Array:


argmax(x: array, /, *, axis: int | None = None, keepdims: bool = False) -> array https://data-apis.org/array-api/latest/API_specification/generated/array_api.argmax.html#array_api.argmax

nicola-bastianello · 2026-05-15T09:49:30Z

+    return _BACKEND_INSTANCE.argmax(array, axis, keepdims)
+
+
+def argmin(array: Array, axis: int | None = None, keepdims: bool = False) -> Array:


argmin(x: array, /, *, axis: int | None = None, keepdims: bool = False) -> array https://data-apis.org/array-api/latest/API_specification/generated/array_api.argmin.html#array_api.argmin

nicola-bastianello · 2026-05-15T09:52:37Z

+    return _BACKEND_INSTANCE.argmin(array, axis, keepdims)
+
+
+def set_item(array: Array, key: ArrayKey, value: Array) -> None:


nicola-bastianello · 2026-05-15T10:01:15Z

+    return (current_seed + random_data) % (2**32)
+
+
+def normal(mean: float = 0.0, std: float = 1.0, shape: tuple[int, ...] = ()) -> Array:


random functions for array creation are not currently covered by the array api, so let's make our best guess at a good api, knowing that in the future we might need to break it if we want compatibility

normal, uniform, choice look good; but normal_like and uniform_like are not universally exposed. I would remove the two *_like functions, since it's not difficult to grab the shape from an array. if we decide to keep them, array arg should be renamed x

we could also include the dtype arg that e.g. torch, jax, tf expose

nicola-bastianello

comments on conversion to/from numpy, and on devices

nicola-bastianello · 2026-05-15T10:34:52Z

+    return _BACKEND_INSTANCE.device_to_native(device)
+
+
+def device_of(array: Array) -> SupportedDevices:


the array api suggests exposing device as a property of array objects. I think we should do the same, and make this function into a private utility. (as context: pytorch and tf also expose .device, in jax one needs .devices())

https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.device.html

the array api also suggests exposing:

default_device() -> device https://data-apis.org/array-api/latest/API_specification/generated/array_api.info.default_device.html#array_api.info.default_device

devices() -> Tuple[device, ...] https://data-apis.org/array-api/latest/API_specification/generated/array_api.info.devices.html#array_api.info.devices

I think we could already add default_device to return the backend device; devices we can discuss later

another thing to discuss later is array.to_device(device: device, /, *, stream: int | Any | None = None) -> array https://data-apis.org/array-api/latest/API_specification/generated/array_api.array.to_device.html#array_api.array.to_device

this is part of the discussion of allowing device as an arg in array creation functions

nicola-bastianello · 2026-05-15T10:39:42Z

+    return _BACKEND_INSTANCE.eye_like(array)
+
+
+def device_to_native(device: SupportedDevices) -> Any:  # noqa: ANN401


this should probably be private

nicola-bastianello · 2026-05-15T10:51:10Z

+    return _BACKEND_INSTANCE.copy(array)
+
+
+def to_numpy(array: Array) -> NDArray[Any]:


both torch and tf use tensor.numpy() for conversion to numpy. we could change this function to private and call it from Array.numpy(). jax is a bit different but I think it's fair to use .numpy() as the api choice

related #3, which would require implementing __array__ and __array_wrap__, but that's a discussion for later

nicola-bastianello · 2026-05-15T10:52:15Z

+    return _BACKEND_INSTANCE.to_numpy(array)
+
+
+def from_numpy(array: NDArray[Any]) -> Array:


all frameworks seem to do this differently, so let's keep the name from_numpy. only change: array arg to x as in the rest of the api

nicola-bastianello · 2026-05-16T08:15:15Z

another comment: I like that with the new version there is no need for _NoBackend class that replicates the same api but always raises. however, having a is None check in every iop function might add a slight overhead (probably very little). so I've thought of an alternative: we go back to a custom class, and use __getattribute__ to raise with the custom error message

class _NoBackend:
    def __getattribute__(self, name):
        raise RuntimeError(...)

this would allow removing all checks from the iop functions. again, I'm not sure this is much of an improvement in terms of performance, but it would at least cut out a lot of repeated code

Simpag · 2026-05-16T15:20:56Z

another comment: I like that with the new version there is no need for _NoBackend class that replicates the same api but always raises. however, having a is None check in every iop function might add a slight overhead (probably very little). so I've thought of an alternative: we go back to a custom class, and use __getattribute__ to raise with the custom error message
class _NoBackend:
    def __getattribute__(self, name):
        raise RuntimeError(...)
this would allow removing all checks from the iop functions. again, I'm not sure this is much of an improvement in terms of performance, but it would at least cut out a lot of repeated code

I've already tried this and it doesnt compile. _NoBackend would not be of type Backend so it is not assignable to backend instance. So we'd need to implement every abstract method. The current solution was the best I could come up with without having some very complex code

Simpag added 6 commits May 10, 2026 03:22

Initial upload

1a482dc

fix docs

49226bc

fix broken links

ce74739

fix pytest

7ff0352

update ci

6c617e7

add logical operators

4633489

Copilot AI review requested due to automatic review settings May 10, 2026 18:22

Copilot started reviewing on behalf of Simpag May 10, 2026 18:23 View session

Copilot AI reviewed May 10, 2026

View reviewed changes

Simpag added 4 commits May 10, 2026 20:34

Merge branch 'main' into dev1

19b5060

fix(PR): PR comments and fix MPS tests

8852fdb

Merge branch 'dev1' of github.com:Simpag/decent-array into dev1

f12d604

fix(tests): Fix float64 error on MPS for torch

609b939

Simpag requested a review from nicola-bastianello May 10, 2026 21:33

nicola-bastianello reviewed May 14, 2026

View reviewed changes

nicola-bastianello mentioned this pull request May 14, 2026

Docs #2

Open

nicola-bastianello reviewed May 15, 2026

View reviewed changes


		# Indexing -------------------------------------------------------------

		def __getitem__(self, key: ArrayKey) -> Array:


		# Containers / iteration ----------------------------------------------

		def __len__(self) -> int:

		listener(_BACKEND_INSTANCE)


		def register_backend_listener(listener: Callable[[Backend \| None], None]) -> None:

		return _BACKEND_INSTANCE.zeros(shape)


		def zeros_like(array: Array) -> Array:

		return _BACKEND_INSTANCE.zeros_like(array)


		def ones(shape: tuple[int, ...]) -> Array:

		return _BACKEND_INSTANCE.ones(shape)


		def ones_like(array: Array) -> Array:

		return _BACKEND_INSTANCE.ones_like(array)


		def eye(n: int) -> Array:

		return _BACKEND_INSTANCE.sign(array)


		def maximum(array1: Array \| float, array2: Array \| float) -> Array:

		return _BACKEND_INSTANCE.maximum(array1, array2)


		def argmax(array: Array, axis: int \| None = None, keepdims: bool = False) -> Array:

		return _BACKEND_INSTANCE.argmax(array, axis, keepdims)


		def argmin(array: Array, axis: int \| None = None, keepdims: bool = False) -> Array:

		return _BACKEND_INSTANCE.argmin(array, axis, keepdims)


		def set_item(array: Array, key: ArrayKey, value: Array) -> None:

		return (current_seed + random_data) % (2**32)


		def normal(mean: float = 0.0, std: float = 1.0, shape: tuple[int, ...] = ()) -> Array:

		return _BACKEND_INSTANCE.device_to_native(device)


		def device_of(array: Array) -> SupportedDevices:

		return _BACKEND_INSTANCE.eye_like(array)


		def device_to_native(device: SupportedDevices) -> Any: # noqa: ANN401

		return _BACKEND_INSTANCE.copy(array)


		def to_numpy(array: Array) -> NDArray[Any]:

		return _BACKEND_INSTANCE.to_numpy(array)


		def from_numpy(array: NDArray[Any]) -> Array:

Conversation

Simpag commented May 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Simpag commented May 10, 2026

Uh oh!

nicola-bastianello left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nicola-bastianello left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nicola-bastianello left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Simpag commented May 10, 2026 •

edited

Loading