v24.02.02

🚨 Breaking Changes

• Remove **kwargs from astype (#14765) @mroeschke
• Remove mimesis as a testing dependency (#14723) @mroeschke
• Update to Dask's shuffle_method kwarg (#14708) @pentschev
• Drop Pascal GPU support. (#14630) @bdice
• Update to CCCL 2.2.0. (#14576) @bdice
• Expunge as_frame conversions in Column algorithms (#14491) @wence-
• Deprecate cudf::make_strings_column accepting typed offsets (#14461) @davidwendt
• Remove deprecated nvtext::load_merge_pairs_file (#14460) @davidwendt
• Include writer code and writerVersion in ORC files (#14458) @vuule
• Remove null mask for zero nulls in json readers (#14451) @karthikeyann
• REF: Remove **kwargs from to_pandas, raise if nullable is not implemented (#14438) @mroeschke
• Consolidate 1D pandas object handling in as_column (#14394) @mroeschke
• Move chars column to parent data buffer in strings column (#14202) @karthikeyann
• Switch to scikit-build-core (#13531) @vyasr

🐛 Bug Fixes

• Bump to nvcomp 3.0.6. (#15128) @bdice
• [HOTFIX] Unpin numba<0.58 (#15031) @raydouglass
• Exclude tests from builds (#14981) @vyasr
• Fix the bounce buffer size in ORC writer (#14947) @vuule
• Revert sum/product aggregation to always produce int64_t type (#14907) @SurajAralihalli
• Fixed an issue with output chunking computation stemming from input chunking. (#14889) @nvdbaranec
• Fix total_byte_size in Parquet row group metadata (#14802) @etseidl
• Fix index difference to follow the pandas format (#14789) @amiralimi
• Fix shared-workflows repo name (#14784) @raydouglass
• Remove unparseable attributes from all nodes (#14780) @vyasr
• Refactor and add validation to IntervalIndex.init (#14778) @mroeschke
• Work around incompatibilities between V2 page header handling and zStandard compression in Parquet writer (#14772) @etseidl
• Fix calls to deprecated strings factory API (#14771) @davidwendt
• Fix ptx file discovery in editable installs (#14767) @vyasr
• Revise shuffle deprecation to align with dask/dask (#14762) @rjzamora
• Enable intermediate proxies to be picklable (#14752) @shwina
• Add CUDF_TEST_PROGRAM_MAIN macro to tests lacking it (#14751) @etseidl
• Fix CMake args (#14746) @vyasr
• Fix logic bug introduced in #14730 (#14742) @wence-
• [Java] Choose The Correct RoundingMode For Checking Decimal OutOfBounds (#14731) @razajafri
• Fix Groupby.get_group (#14728) @rjzamora
• Ensure that all CUDA kernels in cudf have hidden visibility. (#14726) @robertmaynard
• Split cuda versions for notebook testing (#14722) @raydouglass
• Fix to_numeric not preserving Series index and name (#14718) @mroeschke
• Update dask-cudf wheel name (#14713) @raydouglass
• Fix strings::contains matching end of string target (#14711) @davidwendt
• Update to Dask's shuffle_method kwarg (#14708) @pentschev
• Write file-level statistics when writing ORC files with zero rows (#14707) @vuule
• Potential fix for peformance regression in #14415 (#14706) @etseidl
• Ensure DataFrame column types are preserved during serialization (#14705) @mroeschke
• Skip numba test that fails on ARM (#14702) @brandon-b-miller
• Allow Z in datetime string parsing in non pandas compat mode (#14701) @mroeschke
• Fix nan_as_null not being respected when passing arrow object (#14688) @mroeschke
• Fix constructing Series/Index from arrow array and dtype (#14686) @mroeschke
• Fix Aggregation Type Promotion: Ensure Unsigned Input Types Result in Unsigned Output for Sum and Multiply (#14679) @SurajAralihalli
• Add BaseOffset as a final proxy type to pass instancechecks for offsets against BaseOffset (#14678) @shwina
• Add row conversion code from spark-rapids-jni (#14664) @ttnghia
• Unconditionally export the CCCL path (#14656) @vyasr
• Ensure libcudf searches for our patched version of CCCL first (#14655) @robertmaynard
• Constrain CUDA in notebook testing to prevent CUDA 12.1 usage until we have pynvjitlink (#14648) @vyasr
• Fix invalid memory access in Parquet reader (#14637) @etseidl
• Use column_empty over as_column([]) (#14632) @mroeschke
• Add (implicit) handling for torch tensors in is_scalar (#14623) @wence-
• Fix astype/fillna not maintaining column subclass and types (#14615) @mroeschke
• Remove non-empty nulls in cudf::get_json_object (#14609) @davidwendt
• Remove cuda::proclaim_return_type from nested lambda (#14607) @ttnghia
• Fix DataFrame.reindex when column reindexing to MultiIndex/RangeIndex (#14605) @mroeschke
• Address potential race conditions in Parquet reader (#14602) @etseidl
• Fix DataFrame.reindex removing column name (#14601) @mroeschke
• Remove unsanitized input test data from copy gtests (#14600) @davidwendt
• Fix race detected in Parquet writer (#14598) @etseidl
• Correct invalid or missing return types (#14587) @robertmaynard
• Fix unsanitized nulls from strings segmented-reduce (#14586) @davidwendt
• Upgrade to nvCOMP 3.0.5 (#14581) @davidwendt
• Fix unsanitized nulls produced by cudf::clamp APIs (#14580) @davidwendt
• Fix unsanitized nulls produced by libcudf dictionary decode (#14578) @davidwendt
• Fixes a symbol group lookup table issue (#14561) @elstehle
• Drop llvm16 from cuda118-conda devcontainer image (#14526) @charlesbluca
• REF: Make DataFrame.from_pandas process by column (#14483) @mroeschke
• Improve memory footprint of isin by using contains (#14478) @wence-
• Move creation of env.yaml outside the current directory (#14476) @davidwendt
• Enable pd.Timestamp objects to be picklable when cudf.pandas is active (#14474) @shwina
• Correct dtype of count aggregations on empty dataframes (#14473) @wence-
• Avoid DataFrame conversion in MultiIndex.from_pandas (#14470) @mroeschke
• JSON writer: avoid default stream use in string_scalar constructors (#14444) @vuule
• Fix default stream use in the CSV reader (#14443) @vuule
• Preserve DataFrame(columns=).columns dtype during empty-like construction (#14381) @mroeschke
• Defer PTX file load to runtime (#13690) @brandon-b-miller

📖 Documentation

• Disable parallel build (#14796) @vyasr
• Add pylibcudf to the docs (#14791) @vyasr
• Describe unpickling expectations when cudf.pandas is enabled (#14693) @shwina
• Update CONTRIBUTING for pyproject-only builds (#14653) @vyasr
• More doxygen fixes (#14639) @vyasr
• Enable doxygen XML generation and fix issues (#14477) @vyasr
• Some doxygen improvements (#14469) @vyasr
• Remove warning in dask-cudf docs (#14454) @wence-
• Update README links with redirects. (#14378) @bdice
• Add pip install instructions to README (#13677) @shwina

🚀 New Features

• Add ci check for external kernels (#14768) @robertmaynard
• JSON single quote normalization API (#14729) @shrshi
• Write cuDF version in Parquet "created_by" metadata field (#14721) @etseidl
• Implement remaining copying APIs in pylibcudf along with required helper functions (#14640) @vyasr
• Don't constrain numba<0.58 (#14616) @brandon-b-miller
• Add DELTA_LENGTH_BYTE_ARRAY encoder and decoder for Parquet (#14590) @etseidl
• JSON - Parse mixed types as string in JSON reader (#14572) @karthikeyann
• JSON quote normalization (#14545) @shrshi
• Make DefaultHostMemoryAllocator settable (#14523) @gerashegalov
• Implement more copying APIs in pylibcudf (#14508) @vyasr
• Include writer code and writerVersion in ORC files (#14458) @vuule
• Parquet sub-rowgroup reading. (#14360) @nvdbaranec
• Move chars column to parent data buffer in strings column (#14202) @karthikeyann
• PARQUET-2261 Size Statistics (#14000) @etseidl
• Improve GroupBy JIT error handling (#13854) @brandon-b-miller
• Generate unified Python/C++ docs (#13846) @vyasr
• Expand JIT groupby test suite (#13813) @brandon-b-miller

🛠️ Improvements

• Pin pytest<8 (#14920) @galipremsagar
• Move cudf::char_utf8 definition from detail to public header (#14779) @davidwendt
• Clean up TimedeltaIndex.__init__ constructor (#14775) @mroeschke
• Clean up DatetimeIndex.__init__ constructor (#14774) @mroeschke
• Some frame.py typing, move seldom used methods in frame.py (#14766) @mroeschke
• Remove **kwargs from astype (#14765) @mroeschke
• fix benchmarks compatibility with newer pytest-cases (#14764) @jameslamb
• Add pynvjitlink as a dependency (#14763) @brandon-b-miller
• Resolve degenerate performance in create_structs_data (#14761) @SurajAralihalli
• Simplify ColumnAccessor methods; avoid unnecessary validations (#14758) @mroeschke
• Pin pytest-cases<3.8.2 (#14756) @mroeschke
• Use _from_data instead of _from_columns for initialzing Frame (#14755) @mroeschke
• Consolidate cudf object handling in as_column (#14754) @mroeschke
• Reduce execution time of Parquet C++ tests (#14750) @vuule
• Implement to_datetime(..., utc=True) (#14749) @mroeschke
• Remove usages of rapids-env-update (#14748) @KyleFromNVIDIA
• Provide explicit pool size and avoid RMM detail APIs (#14741) @harrism
• Implement cudf.MultiIndex.from_arrays (#14740) @mroeschke
• Remove unused/single use methods (#14739) @mroeschke
• refactor CUDA versions in dependencies.yaml (#14733) @jameslamb
• Remove unneeded methods in Column (#14730) @mroeschke
• Clean up base column methods (#14725) @mroeschke
• Ensure column.fillna signatures are consistent (#14724) @mroeschke
• Remove mimesis as a testing dependency (#14723) @mroeschke
• Replace as_numerical with as_numerical_column/codes (#14719) @mroeschke
• Use offsetalator in gather_chars (#14700) @davidwendt
• Use make_strings_children for fill() specialization logic (#14697) @davidwendt
• Change io::detail::orc namespace into io::orc::detail (#14696) @ttnghia
• Fix call to deprecated factory function (#14695) @davidwendt
• Use as_column instead of arange for range like inputs (#14689) @mroeschke
• Reorganize ORC reader into multiple files and perform some small fixes to cuIO code (#14665) @ttnghia
• Split parquet test into multiple files (#14663) @etseidl
• Custom error messages for IO with nonexistent files (#14662) @vuule
• Explicitly pass .dtype into is_foo_dtype functions (#14657) @mroeschke
• Basic validation in reader benchmarks (#14647) @vuule
• Update dependencies.yaml to support CUDA 12.*. (#14644) @bdice
• Consolidate memoryview handling in as_column (#14643) @mroeschke
• Convert FieldType to scoped enum (#14642) @vuule
• Use instance over is_foo_dtype (#14641) @mroeschke
• Use isinstance over is_foo_dtype internally (#14638) @mroeschke
• Remove unnecessary **kwargs in function signatures (#14635) @mroeschke
• Drop nvbench patch for nvml. (#14631) @bdice
• Drop Pascal GPU support. (#14630) @bdice
• Add cpp/doxygen/xml to .gitignore (#14613) @davidwendt
• Create strings-specific make_offsets_child_column for multiple offset types (#14612) @davidwendt
• Use the offsetalator in cudf::concatenate for strings (#14611) @davidwendt
• Make Parquet ColumnIndex null_counts optional (#14596) @etseidl
• Support freq in DatetimeIndex (#14593) @shwina
• Remove legacy benchmarks for cuDF-python (#14591) @osidekyle
• Remove WORKSPACE env var from cudf_test temp_directory class (#14588) @davidwendt
• Use exceptions instead of return values to handle errors in CompactProtocolReader (#14582) @vuule
• Use cuda::proclaim_return_type on device lambdas. (#14577) @bdice
• Update to CCCL 2.2.0. (#14576) @bdice
• Update dependencies.yaml to new pip index (#14575) @vyasr
• Simplify Python CMake (#14565) @vyasr
• Java expose parquet pass_read_limit (#14564) @revans2
• Add column sanitization checks in CUDF_TEST_EXPECT_COLUMN_* macros (#14559) @SurajAralihalli
• Use cudf_test temp_directory class for nvtext::subword_tokenize gbenchmark (#14558) @davidwendt
• Fix return type of prefix increment overloads (#14544) @vuule
• Make bpe_merge_pairs_impl member private (#14543) @davidwendt
• Small clean up in io::statistics (#14542) @vuule
• Change json gtest environment variable to compile-time definition (#14541) @davidwendt
• Remove extra total chars size calculation from cudf::concatenate (#14540) @davidwendt
• Refactor IndexedFrame.hash_values to use cudf::hashing functions, add xxhash64 to cudf Python. (#14538) @bdice
• Move non-templated inline function definitions from table_view.hpp to table_view.cpp (#14535) @davidwendt
• Add JNI for strings::code_points (#14533) @thirtiseven
• Add a test for issue 12773 (#14529) @vyasr
• Split libarrow build dependencies. (#14506) @bdice
• Implement IndexedFrame.duplicated with distinct_indices + scatter (#14493) @wence-
• Expunge as_frame conversions in Column algorithms (#14491) @wence-
• Remove unsanitized null from input strings column in rank_tests.cpp (#14475) @davidwendt
• Refactor Parquet kernel_error (#14464) @etseidl
• Deprecate cudf::make_strings_column accepting typed offsets (#14461) @davidwendt
• Remove deprecated nvtext::load_merge_pairs_file (#14460) @davidwendt
• Introduce Comprehensive Pathological Unit Tests for Issue #14409 (#14459) @aocsa
• Expose stream parameter in public nvtext APIs (#14456) @davidwendt
• Include encode type in the error message when unsupported Parquet encoding is detected (#14453) @ZelboK
• Remove null mask for zero nulls in json readers (#14451) @karthikeyann
• Refactor cudf.Series.init (#14450) @mroeschke
• Remove the use of volatile in Parquet (#14448) @vuule
• REF: Remove **kwargs from to_pandas, raise if nullable is not implemented (#14438) @mroeschke
• Testing stream pool implementation (#14437) @shrshi
• Match pandas join ordering obligations in pandas-compatible mode (#14428) @wence-
• Forward-merge branch-23.12 to branch-24.02 (#14426) @bdice
• Use isinstance(..., cudf.IntervalDtype) instead of is_interval_dtype (#14424) @mroeschke
• Use isinstance(..., cudf.CategoricalDtype) instead of is_categorical_dtype (#14423) @mroeschke
• Forward-merge branch-23.12 to branch-24.02 (#14422) @bdice
• REF: Remove instances of pd.core (#14421) @mroeschke
• Expose streams in public filling APIs for label_bins (#14401) @ZelboK
• Consolidate 1D pandas object handling in as_column (#14394) @mroeschke
• Limit DELTA_BINARY_PACKED encoder to the same number of bits as the physical type being encoded (#14392) @etseidl
• Add SHA-1 and SHA-2 hash functions. (#14391) @bdice
• Expose streams in Parquet reader and writer APIs (#14359) @shrshi
• Update to fmt 10.1.1 and spdlog 1.12.0. (#14355) @bdice
• Replace default stream for scalars and column factories usages (because of defaulted arguments) (#14354) @karthikeyann
• Expose streams in ORC reader and writer APIs (#14350) @shrshi
• Convert compression and io to string axis type in IO benchmarks (#14347) @SurajAralihalli
• Add cuDF devcontainers (#14015) @trxcllnt
• Refactoring of Buffers (last step towards unifying COW and Spilling) (#13801) @madsbk
• Switch to scikit-build-core (#13531) @vyasr
• Simplify null count checking in column equality comparator (#13312) @vyasr