1.9.0rc10

@jpivarski removed std::cout from ArrayBuilder and print from several of the tests.

1.9.0rc8

@henryiii cleaned up dependabot: PR #1518.

@ianna set up Awkward's new header-only library: PRs #1523, #1524. Migrated old LayoutBuilder to v2: #1484.

@ManasviGoyal developed a new GrowableBuffer, using panels and in a header-only library: PR #1535.

@ioanaif fixed pre-commit issues: #1533. Initialized values behind mask in to_numpy: PR #1531. Fixed argument lists in deepcopy: PR #1532. Fixed is_none's use of axis: PR #1539.

@aryan26roy fixed the so-called "endianness bug on Mac" (actually the byte-swap was being optimized away): PR #1538.

@jpivarski fixed RecordArray.__repr__: PR #1537.

1.9.0rc7

@agoose77 fixed the MANIFEST file to include the new src/awkward/_v2/cpp-headers directory (which is the main reason for this pre-release): PR #1516.

@jpivarski updated the type-parser for v2: PR #1514.

1.9.0rc6

(Unless otherwise noted, from this release onward, all changes described in release notes are for v2.)

@ianna added a one-column implementation of ak.from_rdataframe: PR #1508.

@jpivarski added ak._v2.types.from_datashape, though the parser will need work to make it accept all v2 type strings: PR #1513.

1.9.0rc5

@jpivarski silenced a known-to-be-fixed NumPy warning: PR #1464.

Version 2.0 development:

@aryan26roy added ak.from_avro_file using AwkwardForth: PR #1491.

@swishdiff implemented autodiff for Awkward Arrays when they're backed by JAX: PRs #1447, #1490, #1496.

@agoose77 added missing methods to the TypeTracer.

@ioanaif fixed the slicing shape for arrays of booleans: PR #1497. Fixed __repr__ for overloaded behaviors (to remove "skip" from a lot of tests): PR #1487. Fixed typo that caused an error to be raised on non-erroneous cases: #1502.

@ianna changed links to awkward-1.0 URLs into awkward: PR #1475. Developed ak.to_rdataframe: PRs #1478. Started ak.from_rdataframe: PR # 1474.

@jpivarski flattened the directory structure inside src/awkward/_v2/operations: PR #1467. Renamed low-level methods to match high-level names: PRs #1468, #1470. Did a sweep to ensure that all __getitem__ methods (and the methods they call, recursively) set a slicer object in handle_error, to return more meaningful error messages: PR #1469. Fixed column-selection mechanism for Parquet: PR #1486. Fixed equality handling in "categorical": PR #1492. Fixed string uniqueness for 32-bit offsets: 82d2631. Typo in unmaskedarray.py: PR #1498. Slicing for UnmaskedArrays (such as those that come from Arrow): PR #1499.

1.9.0rc4

Changed nothing (relative to 1.9.0rc3) except for the name of the repo: scikit-hep/awkward-1.0 → scikit-hep/awkward. I just want to make sure that all of the test and deployment machinery is still working.

See https://github.com/orgs/scikit-hep/discussions/1.

1.9.0rc3

@jpivarski fixed to_layout with allow_records=False (v1 and v2): PR #1456.

Version 2.0 development:

@ioanaif added behaviors for "categorical" types: PR #1434. Added the mixins decorator: PR #1437. ak.firsts: PR #1443. ak.singletons: PR #1444. Fixed usage of is_unique: PR #1429. Enabled a lot of previously skipped tests: PR #1445.

@ianna implemented Awkward Arrays → RDataFrame: PR #1374, #1446.

@douglasdavis fixed a duplicate to_parquet/from_parquet in the io submodule of operations: PR #1435.

@jpivarski implemented to_parquet: #1440. Fixed tests of RDataFrame: PR #1457. Streamlined recursively_apply for small slices of big arrays (performance): PR #1458. Implemented a new to_json, started from_json: PR #1449.

1.9.0rc2

Version 2.0 development:

@martindurant factorized ak._v2.from_parquet in a way that is favorable to Dask: PR #1397.

@ioanaif enabled string and categorical behaviors: PR #1421.

@jpivarski fixed ak._v2.to_arrow for sliced ListOffsetArrays (i.e. offsets don't start at 0): PR #1425. Fixed reducers for 32-bit ListOffsetArrays: PR #1426. Implemented the nan* functions: PR #1428. Added a pretty-printer for types: PR #1430. Fixed bugs in optiontype-in-Parquet: PR #1431. Registered both v1 and v2 Numba extensions: PR #1432. Added __delitem__ to ak._v2.Array and ak._v2.Record: PR #1433.

1.9.0rc1

@agoose77 added ak.is_tuple (v1 and v2): PR #1351, #1356. Added a depth_limit option to ak.broadcast_arrays (v1 and v2): PR #1373.

@henryiii fixed builds for pypy 3.9: PR #1412.

@Ahmad-AlSubaie fixed the Numba extension for an upcoming deprecation: PR #1413.

@jpivarski fixed a case of VirtualArray materialization in ak.with_name: PR #1346. Made ak.cartesian true to its documentation in taking an iterable (potentially unknown len) as argument: PR #1364. Fixed a corner-case in reducers (v1 and v2): PR #1378. Fixed bug in sorting due to zero-length shortcuts (v1 and v2): PR #1387.

Version 2.0 development:

@ioanaif added ak.nan_to_num: PR #1352. Added ak.run_lengths: PR #1347. Added ak.unzip: PR #1354. Added ak.broadcast_arrays: PR #1368. Added ak.copy: PR #1367. Added ak.to_pandas: PR #1369. Added ak.unflatten: PR #1360. Made behavior-passing more consistent: PR #1415. Removed bytemask in favor of mask_as_bool: PR #1410.

@agoose77 cleaned up the reducer code, removing boilerplate: PR #1365. Added recursively_apply to layout Record to fix with_name: PR #1401. Fixed ak.from_numpy for string dtypes: PR #1404. Added an nplike for JAX: PR #1399.

@swishdiff built a framework for GPU backend in v2: PRs #1355, #1381. And added a large suite of kernels: PR #1390.

@jpivarski fixed missing features in the v2 typetracer: PRs #1329, #1340, #1380. Added high-level context to error messages, particularly important for asynchronous calls like the GPU backend (discussion #1321), but for all errors in v2: PRs #1327, #1393. Added foundations for the C++ JIT: PR #1359, #1372, #1376, #1383, #1398. Added a first version of ak.from_parquet: PR #1338. Fixed and improved pretty-printing and high-level presentation: PRs #1384, #1392, . Fixed a performance issue in the new to_list algorithm: PR #1418. Fixed a case of exposing layout in iteration: PR #1419.

1.8.0

Just like 1.8.0rc7, but this is not a release candidate. See the release candidates for all the changes with respect to 1.7.0.

The first non-release candidate to drop Python 2.7 and 3.5, and the first to support Python 3.10.