Repository navigation
Releases
Releases · scikit-hep/awkward
Version 2.15.0
What's Changed
New features
- feat: add type annotations to five array operations by @Paramveersingh-S in #4150
- feat: preserve CUDA nonlocal output range semantics by @ianna in #4464
- feat: preserve attrs through Arrow serialization by @ikrommyd in #4349
- feat: Replace numba-cuda with numba-cuda-mlir in CUDA backend by @shwina in #4358
- feat: support form-only mergeability checks by @ianna in #4401
- feat: convert flat Awkward views for common NumPy calls in Numba by @ianna in #4451
Bug Fixes
- fix: dangling std::string keys in from_rdataframe by @aaronj0 in #4383
- fix: repair the try-it REPL by pinning jQuery and updating jQuery Terminal by @ariostas in #4307
- fix: float16/float128/complex256 reducers, sort, and argsort no longer raise KeyError (#4259) by @ianna in #4270
- fix: pass explicit strides for AwkwardForth outputs and re-enable the cp315t-win32 wheel by @aashirvad08 in #4388
- fix: float128 unique on 32-bit platforms (pyodide) by @henryiii in #4414
- fix: deploy the documentation with rclone instead of the AWS CLI by @ariostas in #4417
- fix: install rclone from its releases instead of the Ubuntu archive by @ariostas in #4419
- fix: preserve sliced nullable list values in Arrow conversion by @renta0426 in #4407
- fix: from_numpy of a masked array with a zero-length inner dimension by @rawsun007 in #4423
- fix: ak.mean/var/std on boolean typetracer arrays by @JoeyTan21 in #4433
- fix: preserve argmin and argmax positions after missing lists by @vera-rubin in #4428
- fix:
ak.from_jsonfails when a null mixes with values of different kinds by @TaiSakuma in #4391 - fix: read float16, float128 and complex256 in type strings by @lgray in #4390
- fix: ignore enclosing-option nulls in UnmaskedArray validity when reading parquet by @aashirvad08 in #4396
- fix: reject a zero slice step in inner dimensions by @ikrommyd in #4352
- fix: support unions in Feather roundtrips by @ariostas in #4441
- fix: migrate to new RDataSource column reader API by @vepadulano in #4048
- fix: remediate the emSign CA incident by @tacaswell in #4491
- fix: to_arrow for unions whose option-type children are longer than the used length by @wolfgang-aura in #4398
- fix: convert IndexedArray indices to int64 for option conversion by @renta0426 in #4427
- fix: slices that select nothing no longer raise negative length errors by @robinbishtt in #4506
- fix: report empty inputs to ak.zip clearly by @ianna in #4448
- fix: serialize CPU fused-op cache compilation by @ianna in #4450
- fix: use 64-bit counters when generating kernel indices by @ianna in #4472
- fix: preserve int64 dataframe indices on 32-bit Windows by @ianna in #4492
- fix: change error to
NotImplementedErrorfor non-NumpyArray inzip_no_broadcastby @ianna in #3743 - fix: compare tuple record type arity by @SidheshwarSarangal in #4385
- fix: preserve pointer provenance in Numba lookups by @ianna in #4462
Performance
- perf: use CuPy's own repeat when it accepts an array of repeats by @ikrommyd in #4354
- perf: build jagged boolean slices without a local-index pass by @ikrommyd in #4334
- perf: hoist wrap_layout's imports to module scope by @ikrommyd in #4346
- perf: cache backend and nplike lookups for unrecognised types by @ikrommyd in #4341
- perf: cheapen NumpyArray construction by @ikrommyd in #4342
- perf: fill missing values without building a UnionArray by @ikrommyd in #4336
- perf: cache compiled kernels and call them through void pointers by @ikrommyd in #4338
- perf: cheapen Index construction and slicing by @ikrommyd in #4340
- perf: restore the nominal isinstance fast path for concrete nplike and backend classes by @ikrommyd in #4339
- perf: delegate Attrs.contains to the backing dict by @ikrommyd in #4347
- perf: avoid building a UnionArray in ak.where for numeric leaves by @ikrommyd in #4335
- perf: build jagged boolean slices with a kernel by @ikrommyd in #4394
- perf: build error-context strings only when an exception is decorated by @ikrommyd in #4331
- perf: write private attributes directly in Array and Record constructors by @ikrommyd in #4333
Documentation
- docs: add TaiSakuma as a contributor for test by @allcontributors[bot] in #4363
- docs: add Soumoditya as a contributor for doc by @allcontributors[bot] in #4362
- docs: add aaronj0 as a contributor for code by @allcontributors[bot] in #4384
- docs: set muon mass in ak.zip in columnar HATS notebook by @henryiii in #4389
- docs: add Paramveersingh-S as a contributor for code by @allcontributors[bot] in #4413
- docs: add renta0426 as a contributor for code by @allcontributors[bot] in #4420
- docs: add llms.txt and Markdown page versions by @henryiii in #4350
- docs: add vepadulano as a contributor for code by @allcontributors[bot] in #4459
- docs: make the to_parquet_row_groups example write real batches by @rawsun007 in #4429
- docs: fix ak.to_parquet_dataset argument docs by @mmporong in #4493
Refactoring
- refactor: share scalar slicing index normalization by @ianna in #4471
- refactor: remove unused sorting-range kernels by @ianna in #4488
- refactor: remove unused CUDA kernel implementations by @ianna in #4461
- refactor: share numeric sort and argsort comparators by @ianna in #4475
- refactor: reuse index normalization in array slicing kernels by @ianna in #4474
Testing
- test: avoid numpy 2.5 warning, stop PR runs from filling the actions cache by @henryiii in #4416
- test: add a property-based test for the
awkward_BitMaskedArray_to_IndexedOptionArraykernel by @TaiSakuma in #4409 - test: widen the #4228 trigger in the Feather roundtrip tests by @TaiSakuma in #4494
- test: add property-based tests for three ByteMaskedArray-mask kernels by @TaiSakuma in #4500
- test: skip the fused-op cache concurrency test on emscripten by @ariostas in #4519
Build & CI
- ci: deploy the documentation to a Cloudflare R2 bucket by @ariostas in #4355
- chore(deps): bump the actions group across 1 directory with 3 updates by @dependabot[bot] in #4357
- chore(deps): bump hypothesis-awkward from 0.20.0 to 0.20.2 by @dependabot[bot] in #4386
- chore(deps): bump codecov/codecov-action from 7.0.0 to 7.1.1 in the actions group by @dependabot[bot] in #4387
- ci: speed up CI and remove redundant jobs by @ariostas in #4393
- chore(deps): bump the actions group across 2 directories with 3 updates by @dependabot[bot] in #4395
- chore(deps): bump hypothesis-awkward from 0.20.2 to 0.20.3 by @dependabot[bot] in #4397
- ci: replace retired macOS 14 test runners by @ianna in #4400
- ci: print Hypothesis statistics in a separate property-test step by @TaiSakuma in #4411
- chore: add Python 3.15 classifiers and test on 3.1...
Version 2.14.0
What's Changed
New features
- feat: add ak.from_cudf and rewrite to_cudf via libcudf by @aashirvad08 in #3949
Bug Fixes
- fix: remove the vectorized union-flattening fast path by @TaiSakuma in #4303
- fix: enforce the plain-function contract in @experimental by @aashirvad08 in #4320
- fix:
to_arrowdatetime conversion by @ianna in #3647 - fix: out-of-bounds write in Cupy.repeat() with trailing zeros by @kmohrman in #4311
Performance
- perf: optimize option-type ListArray to Arrow conversion by @aashirvad08 in #4207
- perf: project option inputs in broadcasting with one nonzero and integer takes by @lgray in #4326
- perf: skip named-axis machinery when no input carries named axes by @ikrommyd in #4332
- perf: cache whether a virtual array's shape is known by @ikrommyd in #4337
- perf: return the input array from asarray when it already has the requested dtype by @ikrommyd in #4343
- perf: test field-name slices before integer-like slices by @ikrommyd in #4344
- perf: fast-path int and numpy integers in is_integer by @ikrommyd in #4345
Documentation
- docs: require the experimental mark when introducing new public APIs by @TaiSakuma in #4299
- docs: render experimental API notices by @Soumoditya in #4315
- docs: add summary lines to 9 field/name/parameter operation docstrings by @TaiSakuma in #4134
- docs: add summary lines to 8 from_* array-library converter docstrings by @TaiSakuma in #4136
- docs: add kmohrman as a contributor for code by @allcontributors[bot] in #4328
- docs: add summary lines to 8 missing-value operation docstrings by @TaiSakuma in #4130
- docs: update CITATION.cff by @TaiSakuma in #4330
- docs: read the Chicago taxi dataset from an R2 mirror by @ianna in #4327
Testing
- test: exclude layouts affected by #4305 from the parquet roundtrip tests by @TaiSakuma in #4306
- test: widen
has_issue_4264to lists under untrimmed regular content by @TaiSakuma in #4309 - test: exclude layouts affected by #4316 from the Arrow
test_roundtripcomparison by @TaiSakuma in #4317 - test: add a no-raise property-test module for
ak.anyby @TaiSakuma in #4314 - test: widen
_has_issue_4228to union option children longer than their used length by @TaiSakuma in #4318 - test: match the experimental warning filter by message, not by class path by @ikrommyd in #4353
Build & CI
- ci: pin numpy below 2.5 in the GPU test environment by @TaiSakuma in #4313
- chore(deps): bump hypothesis-awkward from 0.19.1 to 0.20.0 by @dependabot[bot] in #4294
- chore: generate kernel artefacts from CMake instead of nox -s prepare by @henryiii in #4190
- ci: deploy the documentation to the gh-pages branch by @ariostas in #4323
- chore(deps): bump the actions group across 1 directory with 5 updates by @dependabot[bot] in #4322
- chore: Cap cuda-cccl version at 1.1.1 by @shwina in #4356
New Contributors
- @Soumoditya made their first contribution in #4315
- @kmohrman made their first contribution in #4311
Full Changelog: v2.13.0...v2.14.0
Version 2.13.0
Numba compatibility
numba 0.67.0 and llvmlite 0.49.0, both released on 2026-08-11, rewrote numba.core.cgutils.pointer_add, which broke Awkward Array's Numba integration: indexing an ak.Array inside a jitted function fails to compile with RuntimeError: LLVM IR parsing error (invalid cast opcode for cast from 'i64' to 'ptr'). Awkward Array 2.12.0 and earlier are affected, on every platform and Python version.
This release restores it (#4298). Users staying on an older Awkward Array can pin numba<0.67.
What's Changed
New features
- feat: fused centered sum-of-squares kernel for
ak.var/ak.stdat the innermost axis (CPU + CUDA) by @ianna in #4256 - feat: add the experimental-API mark (functions, methods, properties) by @TaiSakuma in #4279
Bug Fixes
- fix: flatten(axis=None) dropped data or raised on nested unions by @TaiSakuma in #4246
- fix: linear fit overflow by @ianna in #4257
- fix: nested zero-field
RecordArraylength in virtualfrom_buffersby @TaiSakuma in #4293 - fix: empty result for
ak.ravel/ak.flatten(axis=None)with nothing to merge by @TaiSakuma in #4292 - fix: pass pointers, not raw addresses, to numba's
cgutils.pointer_addby @ariostas in #4298
Documentation
- docs: add summary lines to 11 to_* array-library converter docstrings by @TaiSakuma in #4138
- docs: add summary lines to 13 to_* file/format converter docstrings by @TaiSakuma in #4139
- docs: note tests/properties naming in CONTRIBUTING.md by @TaiSakuma in #4273
Testing
- test: add no-raise property tests for
ak.flattenandak.allby @TaiSakuma in #4265 - test: add known-issue predicates for #4278 and #4280 to property tests by @TaiSakuma in #4281
- test: add known-issue predicates for #4255 and #4274 to property tests by @TaiSakuma in #4275
- test: add a no-raise property-test module for ak.ravel by @TaiSakuma in #4284
Build & CI
- chore: use scikit-build-core 1.0 for awkward-cpp by @henryiii in #4236
- chore: drop the over-broad files override on the zizmor hook by @TaiSakuma in #4266
- chore(deps): bump hypothesis-awkward from 0.19.0 to 0.19.1 by @dependabot[bot] in #4271
- chore(deps): bump the actions group with 5 updates by @dependabot[bot] in #4272
- chore: split generated release notes into finer categories by @TaiSakuma in #4267
- ci: limit thread pools in the coverage job by @henryiii in #4276
- ci: set the Dependabot cooldown to zero for hypothesis-awkward by @TaiSakuma in #4290
- chore: remove inert
from __future__ import annotationsimports by @TaiSakuma in #4296 - chore(deps): bump the actions group with 3 updates by @dependabot[bot] in #4295
Full Changelog: v2.12.0...v2.13.0
Version 2.12.0
What's Changed
New features
- feat: write attrs when writing parquet row groups iteratively by @ikrommyd in #4210
- feat: implement
awkward_argsorton the CUDA (cuda.compute) backend by @ianna in #4240 - feat: overflow-safe & numerically-stable ak.var/std/moment/mean/covar/corr via fused two-pass sum-of-squares/powers reducers by @ianna in #4232
Bug-fixes and performance
- fix: Record field assignment mutates shared layouts and stale cppyy cache by @henryiii in #4095
- fix: AwkwardForth VM and libawkward correctness issues by @henryiii in #4105
- fix: undo an assignment that awkward_validation rejects by @ikrommyd in #4212
- fix: overflow-safe stats without needless copies by @ianna in #4223
- fix: Fix Numba lowering for
ArrayBuilder.append()with Unicode strings by @ianna in #4249 - perf: Implement native string lowering for Numba layouts and fix silent data corruption by @ianna in #4250
Other
- test: add property-based roundtrip test for
to_numpy/from_numpyby @TaiSakuma in #4215 - docs: add summary lines to 5 math/numeric operation docstrings by @TaiSakuma in #4133
- docs: add summary lines to 11 from_* file/format converter docstrings by @TaiSakuma in #4137
- test: add property-based roundtrip test for
to_parquet/from_parquetby @TaiSakuma in #4218 - chore: remove Gitter chat and StackOverflow links from README by @ianna in #4213
- docs: add summary lines to 8 misc operation docstrings by @TaiSakuma in #4140
- test: add stability roundtrip test for
to_numpy/from_numpyby @TaiSakuma in #4225 - docs: add summary lines to 19 reducer operation docstrings by @TaiSakuma in #4129
- docs: add summary lines to 3 *_like operation docstrings by @TaiSakuma in #4131
- docs: add summary lines to 5 sorting/indexing operation docstrings by @TaiSakuma in #4132
- docs: add summary lines to 10 type/validity operation docstrings by @TaiSakuma in #4135
- test: add property-based roundtrip tests for
to_arrow/from_arrowby @TaiSakuma in #4231 - test: add property-based roundtrip tests for
to_feather/from_featherby @TaiSakuma in #4239 - test: add property-based roundtrip tests for
to_json/from_jsonby @TaiSakuma in #4245 - docs: move description text out of Examples: sections in merged files by @TaiSakuma in #4234
- chore: stop requiring
from __future__ import annotationsby @TaiSakuma in #4253
Full Changelog: v2.11.0...v2.12.0
Version 2.11.0
What's Changed
New features
- feat: add lazy IR scheduling layer by @ianna in #4173
- feat: allow row group size in bytes as a string in the to_parquet family of functions by @ikrommyd in #4202
Bug-fixes and performance
- fix: include kernel-specification.yml in source distribution by @5U5H4NTH006 in #4179
- fix: default row group size is very big and not what pyarrow defaults to by @ikrommyd in #4191
- fix: handle
FutureWarningfor deprecated feather by @ianna in #4198 - fix: nan-variant reducers drop behavior and attrs by @henryiii in #4093
- fix: validation no-ops and broken error paths in contents by @henryiii in #4102
- fix: form and type equality bugs by @henryiii in #4101
- perf: vectorize union flattening and cache nplike feature probes by @henryiii in #4109
- fix: assorted operation bugs (softmax, enforce_type, merge_union_of_records, and more) by @henryiii in #4107
- fix: categorical hashing, delayed-backend detection, and pretty-printing by @henryiii in #4106
- fix: ensure
lengthis a numpy scalar for CUDA stability by @ianna in #4194 - fix:
any_backend_is_delayedshould only be iterating over sequences of arrays for now by @ikrommyd in #4208 - fix: transient attrs should not be written to parquet by @ikrommyd in #4209
Other
- chore(deps): bump hypothesis-awkward from 0.18.0 to 0.18.1 by @dependabot[bot] in #4175
- docs: add 5U5H4NTH006 as a contributor for code by @allcontributors[bot] in #4180
- chore(deps): bump the actions group with 2 updates by @dependabot[bot] in #4176
- chore(deps): bump the actions group with 2 updates by @dependabot[bot] in #4196
- chore(deps): bump hypothesis-awkward from 0.18.1 to 0.19.0 by @dependabot[bot] in #4195
- chore: bump default parquet version to 2.6 by @ikrommyd in #4192
- chore: remove dead code and outdated guards by @henryiii in #4103
- chore(deps): bump the actions group with 2 updates by @dependabot[bot] in #4206
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #4182
- chore: update reviewer list by @pfackeldey in #4200
New Contributors
- @5U5H4NTH006 made their first contribution in #4179
Full Changelog: v2.10.0...v2.11.0
Version 2.10.0
Highlights
- Much faster list reductions (
ak.sum,ak.max, âĶ) â about 5à faster on average (geometric mean), 10Ã+ on long lists #4056- Kernels rewritten: CPU (
awkward-cpp), GPU (cuda.compute) - Exact floating-point results may differ from the previous release
- The same arrays may have different internal layouts than before
- Removed the undocumented
ak.index.EmptyIndex/ZeroIndex/LazyIndex
- Kernels rewritten: CPU (
- Many bug fixes and improvements from code review by Claude Fable 5 (PRs 4089â4108) #4085
What's Changed
New features
- feat: complete CPU kernel migration from
parentstooffsetsby @ianna in #4056 - feat: migrate
UnionArraykernels tocuda.computeby @maxymnaumchyk in #4083 - feat: combine cuda.compute PRs into one by @maxymnaumchyk in #4124
- feat: migrate kernels that were using
parentsto cuda.compute by @maxymnaumchyk in #4159
Bug-fixes and performance
- fix: add missing .attrs support in parquet serialisation by @albert-physics in #4079
- fix:
test_pandas_attr_serialisationshould have unique parquet path per thread by @ikrommyd in #4110 - fix: passing a non-Collection iterable to pytest.parametrize is deprecated by @maxymnaumchyk in #4144
- fix: datetime.time conversion and binary buffer copy in bindings by @henryiii in #4089
- fix: properly error on conversion of
IndexedOptionArraywith anEmptyArraycontent andlength > 0to aByteMaskedArrayby @ikrommyd in #4127 - fix: named-axis context sharing and ak.where broadcasting crash by @henryiii in #4108
- fix: require pyarrow>=17.0.0 by @henryiii in #4152
- fix: wrong results from string sort/unique kernels by @henryiii in #4091
- fix: int64 precision loss in sort comparators and per-comparison branching by @henryiii in #4090
- fix: LayoutBuilder Union form JSON and Record map-type support by @henryiii in #4094
- fix: misc integration bugs (hist, avro, numexpr, RDataFrame) by @henryiii in #4096
- fix: modify cache key for macOS deployment in workflow by @ianna in #4158
- fix: reject non-integer float indices instead of truncating by @henryiii in #4097
- fix: typetracer correctness issues affecting dask-awkward by @henryiii in #4100
- fix: BitMaskedArray mask inversion and placeholder detection by @henryiii in #4092
- fix: Arrow conversion without pandas, empty chunked arrays, and multi-file Parquet attrs by @henryiii in #4099
- fix: BitMasked builder use-after-free and GrowableBuffer copy bugs by @henryiii in #4098
Other
- ci: fix deploy job by @ariostas in #4082
- chore: add an AGENTS file by @henryiii in #4086
- chore: don't refuse to install awkward_cpp when using uv by @henryiii in #4087
- chore(deps): bump the actions group with 3 updates by @dependabot[bot] in #4121
- docs: add albert-physics as a contributor for code by @allcontributors[bot] in #4122
- ci: automate release-note categorization with type labels by @TaiSakuma in #4125
- docs: document kernel spec invariants and CPU/CUDA equivalence in AGENTS by @henryiii in #4088
- ci: run property-based tests nightly with a large max_examples by @TaiSakuma in #4128
- test: add a property-based kernel test with CPU and GPU backends by @TaiSakuma in #4142
- ci: show Hypothesis statistics in the nightly property tests by @TaiSakuma in #4146
- chore: modernize header-only code to C++17 and remove dead code by @henryiii in #4104
- ci: run the GPU kernel property tests nightly at the nightly profile by @TaiSakuma in #4147
- ci: enable pyodide build by @ryanking13 in #4149
- ci: fix docs-preview PR comment failing without a checkout by @TaiSakuma in #4155
- chore(deps): bump the actions group across 1 directory with 3 updates by @dependabot[bot] in #4153
- docs: describe the returned object in Returns: for already-merged operations by @TaiSakuma in #4148
- chore(deps): update python version for mypy by @maxymnaumchyk in #4172
New Contributors
- @albert-physics made their first contribution in #4079
- @ryanking13 made their first contribution in #4149
Full Changelog: v2.9.1...v2.10.0
Version 2.9.1
New features
- feat: improve and test big-endian support by @ikrommyd in #3676
- feat: implement array compatibility/interface protocols for virtual array and placeholder array (for better errors) by @ikrommyd in #3839
- feat: reimplement reducers kernels using cccl by @maxymnaumchyk in #3840
- feat: implement byteswap as an nplike function by @aashirvad08 in #3845
- feat: lazy parents allocation by @ianna in #3860
- feat: extend Python reducers with
offsetsupport (precursor toparentremoval) by @ianna in #3865 - feat: pass 'offsets' to reducer kernels (precursor to parent removal) by @maxymnaumchyk in #3883
- feat: reducer max for
axis==Noneby @ianna in #3884 - feat: improve and fix delayed length calculations and add delayed length calculation for layouts that need it by @ikrommyd in #3889
- feat: pickle in native byteorder and pass byteorder through pickling and unpickling by @ikrommyd in #3893
- feat: customly overload the reducers that call cuda-compute kernels by @maxymnaumchyk in #3894
- feat: add
axis=Nonespecialization forMinreducer + small refactoring for existing specializatons by @ikrommyd in #3896 - feat: cuda.compute combinations kernel study by @ianna in #3908
- feat: add
missing_repeatkernel implementation using cuda.compute by @maxymnaumchyk in #3922 - feat: add
index_rpad_and_clipkernels implementations using cuda.compute by @maxymnaumchyk in #3923 - feat: allow generator expressions in highlevel operations that accept iterable of arrays as input by @ikrommyd in #3938
- feat: add awkward.* aliases to objects.inv for intersphinx by @TaiSakuma in #3958
- feat: properly handle numpy functions that return default containers of numpy arrays in
__array_function__by @ikrommyd in #3991 - feat: lazy concatenation of virtual arrays by @ikrommyd in #4010
- feat: migrate awkward_index_rpad_and_clip_axis0 to cuda.compute by @danielfrg in #4012
- feat: migrate awkward_reduce_sum_complex to cuda.compute by @danielfrg in #4016
- feat: migrate awkward_reduce_max_complex to cuda.compute by @danielfrg in #4018
Bug-fixes and performance
- fix: temporarily use unary_transform instead of segmented_reduce by @maxymnaumchyk in #3814
- fix: C++ errors in
ak.from_rdataframeby @bendavid in #3851 - fix: Fix multiple CUDA out-of-bounds accesses and kernel memory management issues by @ianna in #3858
- fix: harden CUDA/CPU kernels against out-of-bounds writes in
ListOffsetArrayby @ianna in #3859 - fix: add offsets for UnmaskedArray reducer by @ianna in #3874
- fix: GrowableBuffer cleanup compiler warnings by @ianna in #3881
- fix: fix typetracer's
maximumimplementation by @ikrommyd in #3895 - fix: fix ASAN detected buffer overflows in awkward-cpp by @ikrommyd in #3898
- fix: Numba-CUDA refactoring --
CUDATargetContextis spinned off toFunctionDescriptorby @ianna in #3902 - fix: return shallow copy of fields list when
.fieldsis being accessed from highlevel array/record by @ikrommyd in #3906 - fix: handle union array simplification returning back indexed arrays in
ak.almost_equalby @ikrommyd in #3909 - fix:
ak.array_equalwith datetimes and timedeltas with NaTs by @ikrommyd in #3921 - fix: consistent byteorder settings by @ikrommyd in #3954
- fix: couple of fixes in tests that arose from running them in parallel threads by @ikrommyd in #3961
- fix: proper comparison in
ak.almost_equalbetween record array fields by @ikrommyd in #3962 - fix: restore docstring cross-references lost in autoapi migration by @TaiSakuma in #3972
- fix: race condition in record array CUDA kernel by @ianna in #3982
- fix: ensure jax backend uses arrays on cpu only by @ikrommyd in #3990
- fix: array attrs not being validated at creation and being of inconsistent type by @ikrommyd in #3996
- perf: optimize CPU kernels and fix segfault when concatenating arrays with None values by @ianna in #3875
- perf: restructure CUDA unit tests by @ianna in #3935
Other
- chore: ignore jax deprecation warning in pytest by @ikrommyd in #3864
- chore: Update Python version from 3.10 to 3.14 in ROOT workflow by @ianna in #3876
- chore: pin ROOT to 6.36.06 in workflow by @ianna in #3886
- chore: Increase coverage target from 20% to 98% by @ianna in #3897
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #3900
- chore: delete nvidia channel from GPU ci as it is outdated by @ikrommyd in #3903
- chore: upgrade hypothesis-awkward to >=0.10 by @TaiSakuma in #3936
- chore: pin hypothesis-awkward to exact version by @TaiSakuma in #3942
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #3951
- chore: prepare for numpy 2.5 compatibility by @ikrommyd in #3976
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #3993
- chore: Use kw-only arguments in all cuda.compute invocations by @shwina in #3994
- chore: bump hypothesis-awkward from 0.16.0 to 0.18.0 by @TaiSakuma in #3995
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #4076
- chore(ci): harden ci config by @ariostas in #4075
- chore(deps): bump aws-actions/configure-aws-credentials from 5 to 6 in the actions group by @dependabot[bot] in #3857
- chore(deps): bump the actions group with 3 updates by @dependabot[bot] in #3899
- chore(deps): bump pypa/cibuildwheel from 3.3 to 3.4 in the actions group by @dependabot[bot] in #3905
- chore(deps): bump the actions group across 1 directory with 3 updates by @dependabot[bot] in #3940
- chore(deps): bump hypothesis-awkward from 0.10.0 to 0.12.0 by @dependabot[bot] in #3945
- chore(deps): bump hypothesis-awkward from 0.12.0 to 0.13.0 by @dependabot[bot] in #3952
- chore(deps): bump pypa/gh-action-pypi-publish from 1.13.0 to 1.14.0 in the actions group by @dependabot[bot] in #3963
- chore(deps): bump hypothesis-awkward from 0.13.0 to 0.14.0 by @dependabot[bot] in #3968
- chore(deps): bump hypothesis-awkward from 0.14.0 to 0.15.0 by @dependabot[bot] in #3974
- chore(deps): bump the actions group with 2 updates by @dependabot[bot] in #3979
- chore(deps): bump hypothesis-awkward from 0.15.0 to 0.16.0 by @dependabot[bot] in #3984
- chore(deps): bump the actions group with 2 updates by @dependabot[bot] in #3987
- chore(deps): bump aws-actions/configure-aws-credentials from 6.1.1 to 6.1.3 in the actions group by @dependabot[bot] in #4077
- ci: windows ARM build by @henryiii in #3855
- ci: remove xfail from tests that work with cupy 14, pin cupy >= 14 in CI, and split cuda tests with virtual arrays in a separate CI task by @ikrommyd in #3890
- ci: use self-hosted CI for codecov by @ariostas in #3901
- ci: fix issues with self-hosted CI by @ariostas in #3933
- ci: add dependabot config for hypothesis-awkward by @TaiSakuma in #3944
- ci: use reusable workflows by @ariostas in https:/...
Version 2.9.0
New features
Bug-fixes and performance
- fix: pass the right input type to
awkward_NumpyArray_subrange_equal*kernels by @ikrommyd in #3818 - fix: define behavior class within the test function in
test_1511_set_attribute.pyto make it thread-local by @ikrommyd in #3826 - fix: properly respect requested byteorder in
ak.from_buffersfor ndarray buffers by @ikrommyd in #3820 - fix: use
set_maskinstead ofset_base_maskinto_cudfby @ikrommyd in #3847
Other
- chore: use "M8" dtype instead of "<M8" in tests by @ikrommyd in #3819
- chore(deps): bump actions/attest-build-provenance from 3.1.0 to 3.2.0 in the actions group by @dependabot[bot] in #3828
- chore: move array definitions inside test functions in
test_1490_jax_reducers_combinations.pyby @ikrommyd in #3829 - chore: xfail
tests-cuda/test_3051_to_cuda.pyonly in the latest cudf by @ikrommyd in #3836 - chore: do not skip
test_recordarray_argmax_y_fieldand make the xfail reason consistent fortest_listarray_sortintests-cuda/test_3459_virtualarray_with_cuda.pyby @ikrommyd in #3837 - chore: update pre-commit hooks by @pre-commit-ci[bot] in #3834
- chore: delete old
VirtualArrayclass by @ikrommyd in #3848 - build: drop support for PyPy by @ikrommyd in #3835
- ci: drop 3.13t by @henryiii in #3843
- ci: drop cp/pp build selectors by @henryiii in #3854
- docs: add automated contributions policy to CONTRIBUTING.md by @ianna in #3831
- docs: document named-axis support for axis in high-level operations by @X0708a in #3823
- refactor: improve
ak.to_cudferrors and documentation by @ikrommyd in #3850
Full Changelog: v2.8.12...v2.9.0
Version 2.8.12
New features
- feat: improve merging of
NumpyArrays and simplification ofUnionArrays + fixak.almost_equalforUnionArrays by @ikrommyd in #3773 - feat: Add
ak.sort()for CUDA backend by @shwina in #3750 - feat: Make
ak.combinationsfaster on GPU by usingcp.searchsortedto compute output list indexes by @shwina in #3795 - feat: add argmax implementation using CCCL by @maxymnaumchyk in #3777
- feat: make
ak.combinationsfaster on GPU by usingcp.searchsortedto compute output list indexes by @ianna in #3798 - feat: add argmin implementation using CCCL by @maxymnaumchyk in #3811
Bug-fixes and performance
- fix: incorrect type passed down for the
mergeastableargument by @ikrommyd in #3779 - fix: fix
ak.validity_errorfor typetracer-backed layouts with unknown length by @ikrommyd in #3784 - fix: make
ak.enforce_typework with typetracers by @ikrommyd in #3764 - fix:
ArrayBuilder.show()formatter forwarding by @X0708a in #3796 - fix: fix and test
Form.length_zero/one_array()by @ikrommyd in #3810 - fix: make named axis thread-safe and start testing using multiple threads in CI by @ikrommyd in #3805
Other
- ci: use codecov token by @ariostas in #3769
- chore(ci): use macOS to test ML dependencies by @ariostas in #3774
- chore(deps): bump the actions group across 1 directory with 5 updates by @dependabot[bot] in #3778
- ci: xfail failing cudf tests until fixed by @ikrommyd in #3781
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #3785
- chore: request for tensorflow in requirements only if python < 3.14 by @ikrommyd in #3787
- chore: fix noxfile on Intel macOS by @henryiii in #3815
- docs: add X0708a as a contributor for code by @allcontributors[bot] in #3797
- docs: add a comparison of different argmax implementations (using unary_transform, segmented_reduce and cuda-kernels) by @maxymnaumchyk in #3800
- docs: Document GPU backend and cuDF compatibility limitations by @X0708a in #3813
New Contributors
- @X0708a made their first contribution in #3796
Full Changelog: v2.8.11...v2.8.12
Version 2.8.11
New features
- feat: add behavior cls validation hook by @pfackeldey in #3710
- feat: add a playground for cccl studies by @pfackeldey in #3734
- feat: add cuda.compute implementation to CCCL study by @shwina in #3748
- feat: ability to optionally disable the caching of virtual arrays on themselves by @ikrommyd in #3741
Bug-fixes and performance
- fix: string array numpy conversion fails with int32 offsets from parquet by @DylanModesitt in #3697
- fix: only check length against empty form in
ak.from_buffersif it is known by @ikrommyd in #3714 - fix: typestr respects now limit_cols for .show and repr_mimebundle by @pfackeldey in #3712
- fix:
ArrayBuilder.__bool__raisesTypeErrorwhen builder length is 1 by @T90REAL in #3738 - fix: prevent teardown from crashing GPU test suite by @ianna in #3747
- fix: hide awkward.ak_* and awkward.str.akstr_* from 'awkward' and 'awkward.str' module by @pfackeldey in #3752
- fix: align multi-level indices before merge in
to_dataframe()by @TaiSakuma in #3733 - fix: numba test failures on Python 3.14 by @ianna in #3757
- fix: incorrect passing of
Indexinside nplike function into_IndexedOptionArray64by @ikrommyd in #3727 - fix: weights should not be raised to the power n in the n-th moment calculation by @ikrommyd in #3729
- fix:
ak.Recordcreation from iterable with nested dicts by @ikrommyd in #3728 - fix: keep track of buffer key instead of field path for VirtualNDArrays & PlaceholderArrays by @pfackeldey in #3751
- fix: typetracer nplike's
allfunction leads to infinite recursion foraxis=Noneby @ikrommyd in #3765
Other
- ci: fix compilation on self-hosted CI and test with multiple versions of CUDA by @ariostas in #3704
- chore: update acknowledgements to include NumFOCUS affiliation by @ianna in #3708
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #3709
- chore: pytest log_level is better than log_cli_level by @henryiii in #3719
- chore: update requirements files for latest possible python versions by @ikrommyd in #3739
- chore(deps): bump the actions group with 2 updates by @dependabot[bot] in #3744
- chore: update pre-commit hooks by @pre-commit-ci[bot] in #3754
- chore: Remove CUDA 11 support by @shwina in #3758
- chore(ci): use macos-15-intel for x86 wheel building and testing by @ariostas in #3720
- docs: add new versions to version selector by @ariostas in #3707
- docs: 3.14 is supported by @henryiii in #3717
- docs: add T90REAL as a contributor for code by @allcontributors[bot] in #3742
- docs: Add dependency and version change guidelines by @ianna in #3745
- docs: add shwina as a contributor for code by @allcontributors[bot] in #3759
- docs: Add an example summation reducer using cccl by @maxymnaumchyk in #3762
New Contributors
Full Changelog: v2.8.10...v2.8.11