v25.02.02

🚨 Breaking Changes

• Expose stream-ordering in scalar and avro APIs (#17766) @shrshi
• Add seed parameter to hash_character_ngrams (#17643) @davidwendt
• Performance improvements and simplifications for fixed size row-based rolling windows (#17623) @wence-
• Refactor distinct hash join to handle multiple probes with the same build table (#17609) @PointKernel
• Deprecate cudf::grouped_time_range_rolling_window (#17589) @wence-
• Remove "legacy" Dask DataFrame support from Dask cuDF (#17558) @rjzamora
• Return empty result for segmented_reduce if input and offsets are both empty (#17437) @davidwendt
• Rework minhash APIs for deprecation cycle (#17421) @davidwendt
• Change indices for dictionary column to signed integer type (#17390) @davidwendt

🐛 Bug Fixes

• Use protocol for dlpack instead of deprecated function (#18134) @vyasr
• Skip the failing connectorx polars tests (#18037) @Matt711
• Fix 'Unexpected short subpass' exception in parquet chunked reader. (#18019) @nvdbaranec
• Fix race check failures in shared memory groupby (#17985) @PointKernel
• Pin ibis version in the cudf.pandas integration tests <10.0.0 (#17975) @Matt711
• Fix the index type in the indexing operator of the span types (#17971) @vuule
• Add missing pin (#17915) @vyasr
• Fix third-party cudf.pandas tests (#17900) @galipremsagar
• Fix numpy data access by making attribute private (#17890) @galipremsagar
• Remove extra local var declaration from cudf.pandas 3rd-party integration shell script (#17886) @Matt711
• Move isinstance_cudf_pandas to fast_slow_proxy (#17875) @galipremsagar
• Make _Series_dtype method a property (#17854) @Matt711
• Fix the bug in determining the heuristics for shared memory groupby (#17851) @PointKernel
• Fix possible OOB mem access in Parquet decoder (#17841) @mhaseeb123
• Require batches to be non-empty in multi-batch JSON reader (#17837) @shrshi
• Fix rolling(min_periods=) with int and null data with mode.pandas_compat (#17822) @mroeschke
• Resolve race-condition in disable_module_accelerator (#17811) @galipremsagar
• Make Series(dtype=object) raise in mode.pandas_compat with non string data (#17804) @mroeschke
• Disable intended disabled ORC tests (#17790) @davidwendt
• Fix empty DataFrame construction not returning RangeIndex columns (#17784) @mroeschke
• Fix various .str methods for pandas compatability (#17782) @mroeschke
• Fix count API issue about ignoring nan values (#17779) @galipremsagar
• Add numba pinning to cudf repo (#17777) @galipremsagar
• Allow .sort_values(na_position=) to include NaNs in mode.pandas_compatible (#17776) @mroeschke
• allow deselecting nvcomp wheels (#17774) @jameslamb
• Use the aligned_resource_adaptor to allocate bloom filter device buffers (#17758) @mhaseeb123
• Avoid instantiating bloom filter query function for nested and bool types (#17753) @mhaseeb123
• Fix DataFrame.merge(Series, how="left"/"right") on column and index not resulting in a RangeIndex (#17739) @mroeschke
• [BUG] xfail Polars excel test (#17731) @Matt711
• Require to implement AutoCloseable for the classes derived from HostUDFWrapper (#17727) @ttnghia
• Remove jlowe as a java committer since he retired (#17725) @tgravescs
• Prevent use of invalid grid sizes in ORC reader and writer (#17709) @vuule
• Enforce schema for partial tables in multi-source multi-batch JSON reader (#17708) @shrshi
• Compute and use the initial string offset when building nested large string cols with chunked parquet reader (#17702) @mhaseeb123
• Fix writing of compressed ORC files with large stripe footers (#17700) @vuule
• Fix cudf.polars sum of empty not equalling zero (#17685) @mroeschke
• Fix formatting in logging (#17680) @vuule
• convert all nulls to nans in a specific scenario (#17677) @galipremsagar
• Define cudf repr methods on the Column (#17675) @mroeschke
• Fix groupby.len with null values in cudf.polars (#17671) @mroeschke
• Fix: DataFrameGroupBy.get_group was raising with length>1 tuples (#17653) @MarcoGorelli
• Fix possible int overflow in compute_mixed_join_output_size (#17633) @davidwendt
• Fix a minor potential i32 overflow in thrust::transform_exclusive_scan in PQ reader preprocessing (#17617) @mhaseeb123
• Fix failing xgboost test in the cudf.pandas third-party integration tests (#17616) @Matt711
• Fix dask_cudf.read_csv (#17612) @rjzamora
• Fix memcheck error in ReplaceTest.NormalizeNansAndZerosMutable gtest (#17610) @davidwendt
• Correctly accept a pandas.CategoricalDtype(pandas.IntervalDtype(...), ...) type (#17604) @mroeschke
• Add ability to modify and propagate names of columns object (#17597) @galipremsagar
• Ignore NaN correctly in .quantile (#17593) @mroeschke
• Fix groupby argmin/max gather of sorted-order indices (#17591) @davidwendt
• Fix ctest fail running libcudf tests in a Debug build (#17576) @davidwendt
• Specify a version for rapids_logger dependency (#17573) @jlowe
• Fix the ORC decoding bug for the timestamp data (#17570) @kingcrimsontianyu
• [JNI] remove rmm argument to set rw access for fabric handles (#17553) @abellina
• Document undefined behavior in div_rounding_up_safe (#17542) @davidwendt
• Fix nvcc-imposed UB in constexpr functions (#17534) @vuule
• Add anonymous namespace to libcudf test source (#17529) @davidwendt
• Propagate failures in pandas integration tests and Skip failing tests (#17521) @Matt711
• Fix libcudf compile error when logging is disabled (#17512) @davidwendt
• Fix Dask-cuDF clip APIs (#17509) @rjzamora
• Fix pylibcudf to_arrow with multiple nested data types (#17504) @mroeschke
• Fix groupby(as_index=False).size not reseting index (#17499) @mroeschke
• Revert "Temporarily skip tests due to dask/distributed#8953" (#17492) @Matt711
• Workaround for a misaligned access in read_csv on some CUDA versions (#17477) @vuule
• Fix some possible thread-id overflow calculations (#17473) @davidwendt
• Temporarily skip tests due to dask/distributed#8953 (#17472) @wence-
• Detect mismatches in begin and end tokens returned by JSON tokenizer FST (#17471) @shrshi
• Support dask>=2024.11.2 in Dask cuDF (#17439) @rjzamora
• Fix write_json failure for zero columns in table/struct (#17414) @karthikeyann
• Fix Debug-mode failing Arrow test (#17405) @zeroshade
• Fix all null list column with missing child column in JSON reader (#17348) @karthikeyann

📖 Documentation

• Fix forward merge 24.12->25.02 (#18002) @raydouglass
• Fix incorrect example in pylibcudf docs (#17912) @Matt711
• Explicitly call out that the GPU open beta runs on a single GPU (#17872) @taureandyernv
• Update cudf.pandas colab link in docs (#17846) @taureandyernv
• [DOC] Make pylibcudf docs more visible (#17803) @Matt711
• Cross-link cudf.pandas profiler documentation. (#17668) @bdice
• Document interpreter install command for cudf.pandas (#17358) @bdice
• add comment to Series.tolist method (#17350) @tequilayu

🚀 New Features

• Bump polars version to <1.22 (#17771) @Matt711
• Make more constexpr available on device for cuIO (#17746) @PointKernel
• Add public interop functions between pylibcudf and cudf classic (#17730) @Matt711
• Support dask_expr migration into dask.dataframe (#17704) @rjzamora
• Make tests build without relaxed constexpr (#17691) @PointKernel
• Set default logger level to warn (#17684) @vyasr
• Support multithreaded reading of compressed buffers in JSON reader (#17670) @shrshi
• Control pinned memory use with environment variables (#17657) @vuule
• Host compression (#17656) @vuule
• Enable text build without relying on relaxed constexpr (#17647) @PointKernel
• Implement HOST_UDF aggregation for reduction and segmented reduction (#17645) @ttnghia
• Add JSON reader options structs to pylibcudf (#17614) @Matt711
• Refactor distinct hash join to handle multiple probes with the same build table (#17609) @PointKernel
• Add JSON Writer options classes to pylibcudf (#17606) @Matt711
• Add ORC reader options structs to pylibcudf (#17601) @Matt711
• Add Avro Reader options classes to pylibcudf (#17599) @Matt711
• Enable binaryop build without relying on relaxed constexpr (#17598) @PointKernel
• Measure the number of Parquet row groups filtered by predicate pushdown (#17594) @mhaseeb123
• Implement HOST_UDF aggregation for groupby (#17592) @ttnghia
• Plumb pylibcudf.io.parquet options classes through cudf python (#17506) @Matt711
• Add partition-wise Select support to cuDF-Polars (#17495) @rjzamora
• Add multi-partition Scan support to cuDF-Polars (#17494) @rjzamora
• Migrate cudf::io::merge_row_group_metadata to pylibcudf (#17491) @Matt711
• Add Parquet Reader options classes to pylibcudf (#17464) @Matt711
• Add multi-partition DataFrameScan support to cuDF-Polars (#17441) @rjzamora
• Return empty result for segmented_reduce if input and offsets are both empty (#17437) @davidwendt
• Abstract polars function expression nodes to ensure they are serializable (#17418) @pentschev
• Add CSV Reader options classes to pylibcudf (#17412) @Matt711
• Add support for pylibcudf.DataType serialization (#17352) @pentschev
• Enable rounding for Decimal32 and Decimal64 in cuDF (#17332) @a-hirota
• Remove upper bounds on cuda-python to allow 12.6.2 and 11.8.5 (#17326) @bdice
• Expose stream-ordering to groupby APIs (#17324) @shrshi
• Migrate ORC Writer to pylibcudf (#17310) @Matt711
• Support reading bloom filters from Parquet files and filter row groups using them (#17289) @mhaseeb123

🛠️ Improvements

• Update to nvcomp 4.2.0.11 (#18042) @bdice
• Remove pandas backend from cudf.pandas - ibis integration tests (#17945) @Matt711
• Revert CUDA 12.8 shared workflow branch changes (#17879) @vyasr
• Remove predicate param from DataFrameScan IR (#17852) @Matt711
• Remove cudf.Scalar from scatter APIs (#17847) @mroeschke
• Remove cudf.Scalar from interval_range (#17844) @mroeschke
• Add verify-codeowners hook (#17840) @KyleFromNVIDIA
• Build and test with CUDA 12.8.0 (#17834) @bdice
• Increase timeout for recently added test (#17829) @galipremsagar
• Apply ruff everywhere (notebooks and scripts) (#17820) @bdice
• Fix pre-commit.ci failures (#17819) @bdice
• Remove incorrect calls to set architectures (#17813) @vyasr
• Fix typo in exception raised when attempting to convert a string column to cupy (#17800) @dagardner-nv
• Add support for pyarrow-19 (#17794) @galipremsagar
• increase parallelism in nightly builds (#17792) @jameslamb
• Reduce libcudf memcheck tests output (#17791) @davidwendt
• Make cudf build with latest CCCL (#17788) @miscco
• Introduce some more rolling window benchmarks (#17787) @wence-
• Add shellcheck to pre-commit and fix warnings (#17778) @gforsyth
• Improve parquet reader very-long string performance (#17773) @pmattione-nvidia
• Update how to manage host UDF instance (#17770) @res-life
• Add getInts api for HostMemoryBuffer and UnsafeMemoryAccessor (#17767) @liurenjie1024
• Expose stream-ordering in scalar and avro APIs (#17766) @shrshi
• Standarize methods used from cudf.core._internals (#17765) @mroeschke
• Implement string join in cudf-polars (#17755) @wence-
• Deprecate dataframe protocol (#17736) @vyasr
• Add parquet reader long row test (#17735) @pmattione-nvidia
• Update kvikio call due to upstream changes (#17733) @kingcrimsontianyu
• Delay setting MultiIndex.level/codes until needed (#17728) @mroeschke
• Bounding pool size in multi-batch JSON reader (#17724) @shrshi
• Use GCC 13 in CUDA 12 conda builds. (#17721) @bdice
• Update minimal sphinx theme version so that we can use parallel doc builds (#17719) @vyasr
• Add more aggregation methods in pylibcudf (#17717) @mroeschke
• Make cudf._lib.string_udf work with pylibcudf Columns instead of cudf._lib Columns (#17715) @mroeschke
• Add special orc test data: timestamp interspersed with null values (#17713) @kingcrimsontianyu
• Add pylibcudf.null_mask.null_count (#17711) @mroeschke
• Ensure pyarrow.Scalar to pylibcudf.Scalar is cached (#17707) @mroeschke
• Adapt cudf numba config for numba 0.61 removal (#17705) @mroeschke
• Remove cudf._lib.scalar in favor of pylibcudf (#17701) @mroeschke
• Fix parquet reader list bug (#17699) @pmattione-nvidia
• Migrated Dynamic AST Expression Trees in Benchmarks and Tests to use AST Tree (#17697) @lamarrr
• Skip polars test that can generate timezones that chrono_tz doesn't know (#17694) @wence-
• Use 64-bit offsets only if the current strings column output chunk size exceeds threshold (#17693) @mhaseeb123
• Use latest ci-conda images (#17690) @bdice
• Add multi-source reading to JSON reader benchmarks (#17688) @shrshi
• Convert cudf.Scalar usage to pylibcudf and pyarrow usage (#17686) @mroeschke
• remove find_package(Python) in libcudf build (#17683) @jameslamb
• Fix build metrics report format with long placehold filenames (#17679) @davidwendt
• Use rapids-cmake for the logger (#17674) @vyasr
• Java Parquet reads via multiple host buffers (#17673) @jlowe
• Remove cudf._libs.types.pyx (#17665) @mroeschke
• Add support for Groupby.cumprod (#17661) @galipremsagar
• Implement .dt.total_seconds (#17659) @galipremsagar
• Avoid shallow copies in groupby methods (#17646) @mroeschke
• Avoid double MultiIndex factorization in groupby index result (#17644) @mroeschke
• Add seed parameter to hash_character_ngrams (#17643) @davidwendt
• Fix possible overflow in WriteCoalescingCallbackWrapper::TearDown (#17642) @davidwendt
• Remove pragma GCC diagnostic from source files (#17637) @davidwendt
• Move unnecessary utilities from cudf._lib.scalar (#17636) @mroeschke
• Support compression= in DataFrame.to_json (#17634) @mroeschke
• Bump Polars version to <1.18 (#17632) @Matt711
• Add public APIs to Access Underlying cudf and pandas Objects from cudf.pandas Proxy Objects (#17629) @galipremsagar
• Use Numba Config to turn on Pynvjitlink Features (#17628) @isVoid
• Use PyNVML 12 (#17627) @jakirkham
• Remove cudf._lib.utils in favor of python APIs (#17625) @mroeschke
• Performance improvements and simplifications for fixed size row-based rolling windows (#17623) @wence-
• Fix return types for MurmurHash3_x86_32 template specializations (#17622) @davidwendt
• Clean up namespaces and improve compression-related headers (#17621) @vuule
• Use more pylibcudf.types instead of cudf._lib.types (#17619) @mroeschke
• Remove patch that is only needed for clang-tidy to run on test files (#17618) @vyasr
• update telemetry actions to fluent-bit friendly style (#17615) @msarahan
• Introduce some simple benchmarks for rolling window aggregations (#17613) @wence-
• Bump the oldest pyarrow version to 14.0.2 in test matrix (#17611) @galipremsagar
• Use [[nodiscard]] attribute before __device__ (#17608) @vuule
• Use host_vector in flatten_single_pass_aggs (#17605) @vuule
• Stop memory_resource.hpp from including itself (#17603) @vyasr
• Replace the outdated cuco window concept with buckets (#17602) @PointKernel
• Check if nightlies have succeeded recently enough (#17596) @vyasr
• Deprecate cudf::grouped_time_range_rolling_window (#17589) @wence-
• A couple of fixes in rapids-logger usage (#17588) @vyasr
• Simplify expression transformer in Parquet predicate pushdown with ast::tree (#17587) @mhaseeb123
• Remove unused functionality in cudf._lib.utils.pyx (#17586) @mroeschke
• Use cuda-python cuda.bindings import names. (#17585) @bdice
• Use no-sync copy for fixed-width types in cudf::concatenate (#17584) @davidwendt
• Remove cudf._lib.groupby in favor of inlining pylibcudf (#17582) @mroeschke
• Remove unused code of json schema in JSON reader (#17581) @karthikeyann
• Expose Scalar's constructor and Scalar#getScalarHandle() to public (#17580) @ttnghia
• Allow large strings in nvtext benchmarks (#17579) @davidwendt
• Remove cudf._lib.reduce in favor of inlining pylibcudf (#17574) @mroeschke
• Use batched memcpy when writing ORC statistics (#17572) @vuule
• Allow large strings in nvbench strings benchmarks (#17571) @davidwendt
• Update version references in workflow (#17568) @AyodeAwe
• Enable all json reader options in pylibcudf read_json (#17563) @karthikeyann
• Remove cudf._lib.parquet in favor of inlining pylibcudf (#17562) @mroeschke
• Fix CMake format in cudf/_lib/CMakeLists.txt (#17559) @mroeschke
• Remove "legacy" Dask DataFrame support from Dask cuDF (#17558) @rjzamora
• Replace direct cudaMemcpyAsync calls with utility functions (within /include) (#17557) @vuule
• Remove cudf._lib.interop in favor of inlining pylibcudf (#17555) @mroeschke
• gate telemetry dispatch calls on TELEMETRY_ENABLED env var (#17551) @msarahan
• Replace direct cudaMemcpyAsync calls with utility functions (within /src) (#17550) @vuule
• Remove unused BufferArrayFromVector (#17549) @Matt711
• Move cudf._lib.copying to cudf.core._internals (#17548) @mroeschke
• Update cuda-python lower bounds to 12.6.2 / 11.8.5 (#17547) @bdice
• Fix typos, rename types, and add null_probability benchmark axis for distinct (#17546) @PointKernel
• Mark more constexpr functions as device-available (#17545) @vyasr
• Use cooperative-groups instead of cub warp-reduce for strings contains (#17540) @davidwendt
• Remove cudf._lib.nvtext in favor of inlining pylibcudf (#17535) @mroeschke
• Add XXHash_32 hasher (#17533) @PointKernel
• Remove unused masked keyword in column_empty (#17530) @mroeschke
• Remove Thrust patch in favor of CMake definition for Thrust 32-bit offset types. (#17527) @bdice
• [JNI] Enables fabric handles for CUDA async memory pools (#17526) @abellina
• Force Thrust to use 32-bit offset type. (#17523) @bdice
• Replace cudf::detail::copy_if logic with thrust::copy_if and gather (#17520) @davidwendt
• Replaces uses of cudf._lib.Column.from_unique_ptr with pylibcudf.Column.from_libcudf (#17517) @Matt711
• Move cudf._lib.aggregation to cudf.core._internals (#17516) @mroeschke
• Migrate copy_column and Column.from_scalar to pylibcudf (#17513) @Matt711
• Remove cudf._lib.transform in favor of inlining pylibcudf (#17505) @mroeschke
• Remove cudf._lib.string.convert/split in favor of inlining pylibcudf (#17496) @mroeschke
• Move cudf._lib.sort to cudf.core._internals (#17488) @mroeschke
• Remove cudf._lib.csv in favor in inlining pylibcudf (#17485) @mroeschke
• Update PyTorch to >=2.4.0 to get fix for CUDA array interface bug, and drop CUDA 11 PyTorch tests. (#17475) @bdice
• Remove cudf._lib.binops in favor of inlining pylibcudf (#17468) @mroeschke
• Remove cudf._lib.orc in favor of inlining pylibcudf (#17466) @mroeschke
• skip most CI on devcontainer-only changes (#17465) @jameslamb
• Set build type for all examples (#17463) @vyasr
• Update the hook versions in pre-commit (#17462) @wence-
• Remove cudf._lib.string_casting in favor of inlining pylibcudf (#17460) @mroeschke
• Remove cudf._lib.filling in favor of inlining pylibcudf (#17459) @mroeschke
• Update MurmurHash3_x64_128 to use the cuco equivalent implementation (#17457) @PointKernel
• Move cudf._lib.stream_compaction to cudf.core._internals (#17456) @mroeschke
• Clean up xxhash_64 implementations (#17455) @PointKernel
• Update Hadoop dependency in Java pom (#17454) @jlowe
• Adapt to rmm logger changes (#17451) @vyasr
• Require approval to run CI on draft PRs (#17450) @bdice
• Expose stream-ordering in nvtext API (#17446) @shrshi
• Use exec_policy_nosync in write_json (#17445) @karthikeyann
• Remove cudf._lib.json in favor of inlining pylibcudf (#17443) @mroeschke
• Remove cudf._lib.null_mask in favor of inlining pylibcudf (#17440) @mroeschke
• Expose stream-ordering in replace API (#17436) @shrshi
• Expose stream-ordering in copying APIs (#17435) @shrshi
• Expose stream-ordering in column view APIs (#17434) @shrshi
• Apply clang-tidy autofixes from new rules (#17431) @vyasr
• Remove cudf._lib.round in favor of inlining pylibcudf (#17430) @mroeschke
• Update MurmurHash3_x86_32 to use the cuco equivalent implementation (#17429) @PointKernel
• Remove cudf._lib.replace in favor of inlining pylibcudf (#17428) @mroeschke
• Remove nvtx/ranges.hpp include from cuda.cuh (#17427) @davidwendt
• Remove the unused detail int_fastdiv.h header (#17426) @PointKernel
• Remove cudf._lib.lists in favor of inlining pylibcudf (#17425) @mroeschke
• Remove cudf._lib.quantile (#17424) @mroeschke
• Remove cudf._lib.rolling in favor of inlining pylibcudf (#17423) @mroeschke
• Avoid converting Decimal32/Decimal64 in to_arrow and from_arrow APIs (#17422) @zeroshade
• Rework minhash APIs for deprecation cycle (#17421) @davidwendt
• Use thread_index_type in binary-ops jit kernel.cu (#17420) @davidwendt
• Change binops for-each kernel to thrust::for_each_n (#17419) @davidwendt
• Move cudf._lib.search to cudf.core._internals (#17411) @mroeschke
• Use grid_1d utilities in copy_range.cuh (#17409) @davidwendt
• Remove cudf._lib.text in favor of inlining pylibcudf (#17408) @mroeschke
• Run clang-tidy checks in PR CI (#17407) @bdice
• Update strings/text source to use grid_1d for thread/block/stride calculations (#17404) @davidwendt
• Expose stream-ordering to strings attribute APIs (#17398) @shrshi
• Expose stream-ordering to interop APIs (#17397) @shrshi
• Remove unused type aliases (#17396) @PointKernel
• Remove some cudf._lib.strings files in favor of inlining pylibcudf (#17394) @mroeschke
• Update xxhash_64 to utilize the cuco equivalent implementation (#17393) @PointKernel
• Change indices for dictionary column to signed integer type (#17390) @davidwendt
• Return categorical values in to_numpy/to_cupy (#17388) @mroeschke
• Forward-merge branch-24.12 to branch-25.02 (#17379) @bdice
• Remove unused IO utilities from cudf python (#17374) @Matt711
• Remove cudf._lib.datetime in favor of inlining pylibcudf (#17372) @mroeschke
• Remove cudf._lib.join in favor of inlining pylibcudf (#17371) @mroeschke
• Remove cudf._lib.merge in favor of inlining pylibcudf (#17370) @mroeschke
• Remove cudf._lib.partitioning in favor of inlining pylibcudf (#17369) @mroeschke
• Remove cudf._lib.reshape in favor of inlining pylibcudf (#17368) @mroeschke
• Remove cudf._lib.timezone in favor of inlining pylibcudf (#17366) @mroeschke
• Remove cudf._lib.transpose in favor of inlining pylibcudf (#17365) @mroeschke
• Move make_strings_column benchmark to nvbench (#17340) @davidwendt
• Improve strings contains/find performance for smaller strings (#17330) @davidwendt
• Use rapids-logger to generate the cudf logger (#17307) @vyasr
• Mukernels strings (#17286) @pmattione-nvidia
• Add write_parquet to pylibcudf (#17263) @mroeschke
• Single-partition Dask executor for cuDF-Polars (#17262) @rjzamora
• Add breaking change workflow trigger (#17248) @AyodeAwe
• Precompute AST arity (#17234) @bdice
• Update to CCCL 2.7.0-rc2. (#17233) @bdice
• Make column_empty mask buffer creation consistent with libcudf (#16715) @mroeschke