Python Polars 1.33.0

💥 Breaking changes

• Remove, deprecate or change eager Exprs to be lazy compatible (#24027)

🚀 Performance improvements

• Native streaming int_range with len or count (#24280)
• Lower arg_unique natively to the streaming engine (#24279)
• Move unordering optimization to end (#24286)
• Do ordering simplification step after common sub-plan elimination (#24269)
• Always simplify order requirements in IR (#24192)
• Basic de-duplication of filter expressions (#24220)
• Cache the IR in pipe_with_schema (#24213)
• Lower arg_where natively to streaming engine (#24088)
• Lower Expr.shift to streaming engine (#24106)
• Lower order-preserving groupby to streaming engine (#24053)

✨ Enhancements

• Add CSE for custom io sources using pointer for hashing (#24297)
• Allow pl.Expr.log to take in an expression (#24226)
• Add caching to user credential providers (#23789)
• Expose mkdir parameter on write_parquet (#24239)
• Implement diff() in streaming engine (#24189)
• Enable Expr.diff(n) for negative n (#24200)
• Allow upcasting null-typed columns to nested column types in scans (#24185)
• Log pyarrow predicate conversion result in sensitive verbose logs (#24186)
• Drop PyArrow requirement for write_database with the ADBC engine (#24136)
• Add a deprecation warning for pl.Series.shift(Null) (#24114)
• Improve Debug formatting of DataType (#24056)
• Add LazyFrame.pipe_with_schema (#24075)
• Catch additional temporal attributes in BytecodeParser function analysis (#24076)
• Add cum_* as native streaming nodes (#23977)
• Add peak_{min,max} support for booleans (#24068)
• Add DataFrame.map_columns for eager evaluation (#23821)

🐞 Bug fixes

• Invalid conversion from non-bit numpy bools (#24312)
• Make dt.epoch('s') serializable (#24302)
• Make Expr.rechunk serializable (#24303)
• Schema mismatch for 'log' operation (#24300)
• Incorrect first/last aggregate in streaming engine (#24289)
• Fix group offsets in sliced groups (#24274)
• Panic in inexact date(time) conversion (#24268)
• Keep DSL cache after serialization and deserialization (#24265)
• Sanitize and warn about eval usage (#24262)
• Correct incorrect default in from_pandas overload for include_index (#24258)
• Unique with keep="none" in new optimization pass (#24261)
• Correct size limits for Decimal cast (#24252)
• Unordered unions in check order observing pass (#24253)
• Fix dtype for slice on Literal in agg context (#24137)
• Fix incorrect filter(lit(True)) when scanning hive (#24237)
• In-memory group_by on 128-bit integers (#24242)
• Fix panic in gather inside groupby with invalid indices (#24182)
• Release the GIL in map_groups (#24225)
• Remove extra explode in LazyGroupBy.{head,tail} (#24221)
• Fix panic in polars cloud CSV scan (#24197)
• Fix panic when loading categorical columns from IO plugin (#24205)
• Fix credential provider did not auto-init on partition sinks (#24188)
• Fix engine type for concat_list on AggScalar implode (#24160)
• Rolling_mean handle centered weights with len(values) < window_size (#24158)
• Reading is_in predicate for Parquet plain strings (#24184)
• Support native DuckDB connection in read_database (#24177)
• Make PyCategories pickleable (#24170)
• Remove unused unsound function to_mutable_slice (#24173)
• PyO3 extension types giving compat_level errors (#24166)
• Allow non-elementwise by in top_k (#24164)
• Fix sort_by for group_by_dynamic context (#24152)
• Input-independent length aggregations in streaming (#24153)
• Release GIL when iterating df in to_arrow (#24151)
• Respect non-elementwise join_where conditions (#24135)
• Fix mismatched pytest test collection error (#24133)
• Resolve schema mismatch for div on Boolean (#24111)
• Fix from_repr parsing of negative durations (#24115)
• Make group_by/partition_by iterator keys tuple[Any, ...] to enable tuple-unpacking (#24113)
• Keep name when doing empty group-aware aggregation (#24098)
• Implode instead of reshape_list (#24078)
• Rolling mean with weights incorrect when min_samples < window_size (#23485)
• Allow merge_sorted for all types (#24077)
• Include datatypes in row_encode expression (#24074)
• Include UDF materialized type in serialization (#24073)
• Correct .rolling() output type for non-aggregations (#24072)
• Correct planner output schema for join_asof (#24071)
• Correct output for fold and reduce (#24069)
• Expr.meta.output_name for struct fields (#24064)
• Ensure upcast operations on pl.Date default to microsecond precision (#23981)
• Add peak_{min,max} support for booleans (#24068)
• Planner output type for mean with strange input type (#24052)
• Remove, deprecate or change eager Exprs to be lazy compatible (#24027)

📖 Documentation

• Fix few typos (#24305)
• Add missing reference to LazyFrame.pipe_with_schema() on the website (#24285)
• Automatically register doctest.ELLIPSIS so we don't have to add the inline directive each time (#24146)
• Update categorical comparison documentation in user guide (#24249)
• Add missing references for Seriers.rolling_*_by methods (#24254)
• Fix formatting of Series.value_counts examples (#24245)
• Add hint to use DataFrame/Series constructors in from_arrow docstring (#22942)
• Update GPU un/supported features (#24195)
• Add DataFrame.map_columns to API (#24128)
• Update multiple pages in the Polars Cloud user guide (#23661)
• Fix str.find_many() docstring example (#24092)

📦 Build system

• Re-enable macos-x86-64 (#24266)
• Drop binary support for macos_x86-64 (#24257)

🛠️ Other improvements

• Remove PDS-H code (#24301)
• Get ready for even more cloud tests (#24292)
• Add tests for slices with caches (#24288)
• Readd ordering tests (#24284)
• Fix Makefile venv path (#24251)
• Remove unnecessary parentheses (#24244)
• Make non-nested shift{,_and_fill} ops generic (#24224)
• Remove unused Wrap (#24214)
• Allow upcasting null-typed columns to nested column types in scans (#24185)
• Automatically label a few more types of PR (#24147)
• Update toolchain (#24156)
• Add order_sensitive property for AExpr (#24116)
• Mark more tests as not possible on cloud (#24103)
• Turn AggExpr::Count from tuple to struct (#24096)
• Mark tests that may fail in cloud (#24067)
• Extend read database tests to capture more ADBC functionality (#24002)
• Make CI perf failures more lenient (#24066)
• Fix hive partition string encoding in CI by upgrading deltalake (#24018)
• Make tests with sinks run on cloud again (#24048)

Thank you to all our contributors for making this release possible!
@Kevin-Patyk, @MarcoGorelli, @NeejWeej, @agossard, @alexander-beedie, @aparna2198, @borchero, @coastalwhite, @deanm0000, @dsprenkels, @eitsupi, @etiennebacher, @gab23r, @henryharbeck, @jjurm, @kdn36, @math-hiyoko, @mcrumiller, @mroeschke, @nameexhaustion, @orlp, @r-brink, @ritchie46, @stijnherfst, @vdrn and @wence-

Python Polars 1.33.0-beta.1

💥 Breaking changes

• Remove, deprecate or change eager Exprs to be lazy compatible (#24027)

🚀 Performance improvements

• Always simplify order requirements in IR (#24192)
• Basic de-duplication of filter expressions (#24220)
• Cache the IR in pipe_with_schema (#24213)
• Lower arg_where natively to streaming engine (#24088)
• Lower Expr.shift to streaming engine (#24106)
• Lower order-preserving groupby to streaming engine (#24053)

✨ Enhancements

• Allow pl.Expr.log to take in an expression (#24226)
• Add caching to user credential providers (#23789)
• Expose mkdir parameter on write_parquet (#24239)
• Implement diff() in streaming engine (#24189)
• Enable Expr.diff(n) for negative n (#24200)
• Allow upcasting null-typed columns to nested column types in scans (#24185)
• Log pyarrow predicate conversion result in sensitive verbose logs (#24186)
• Drop PyArrow requirement for write_database with the ADBC engine (#24136)
• Add a deprecation warning for pl.Series.shift(Null) (#24114)
• Improve Debug formatting of DataType (#24056)
• Add LazyFrame.pipe_with_schema (#24075)
• Catch additional temporal attributes in BytecodeParser function analysis (#24076)
• Add cum_* as native streaming nodes (#23977)
• Add peak_{min,max} support for booleans (#24068)
• Add DataFrame.map_columns for eager evaluation (#23821)

🐞 Bug fixes

• Correct size limits for Decimal cast (#24252)
• Unordered unions in check order observing pass (#24253)
• Fix dtype for slice on Literal in agg context (#24137)
• Fix incorrect filter(lit(True)) when scanning hive (#24237)
• In-memory group_by on 128-bit integers (#24242)
• Fix panic in gather inside groupby with invalid indices (#24182)
• Release the GIL in map_groups (#24225)
• Remove extra explode in LazyGroupBy.{head,tail} (#24221)
• Fix panic in polars cloud CSV scan (#24197)
• Fix panic when loading categorical columns from IO plugin (#24205)
• Fix credential provider did not auto-init on partition sinks (#24188)
• Fix engine type for concat_list on AggScalar implode (#24160)
• Rolling_mean handle centered weights with len(values) < window_size (#24158)
• Reading is_in predicate for Parquet plain strings (#24184)
• Support native DuckDB connection in read_database (#24177)
• Make PyCategories pickleable (#24170)
• Remove unused unsound function to_mutable_slice (#24173)
• PyO3 extension types giving compat_level errors (#24166)
• Allow non-elementwise by in top_k (#24164)
• Fix sort_by for group_by_dynamic context (#24152)
• Input-independent length aggregations in streaming (#24153)
• Release GIL when iterating df in to_arrow (#24151)
• Respect non-elementwise join_where conditions (#24135)
• Fix mismatched pytest test collection error (#24133)
• Resolve schema mismatch for div on Boolean (#24111)
• Fix from_repr parsing of negative durations (#24115)
• Make group_by/partition_by iterator keys tuple[Any, ...] to enable tuple-unpacking (#24113)
• Keep name when doing empty group-aware aggregation (#24098)
• Implode instead of reshape_list (#24078)
• Rolling mean with weights incorrect when min_samples < window_size (#23485)
• Allow merge_sorted for all types (#24077)
• Include datatypes in row_encode expression (#24074)
• Include UDF materialized type in serialization (#24073)
• Correct .rolling() output type for non-aggregations (#24072)
• Correct planner output schema for join_asof (#24071)
• Correct output for fold and reduce (#24069)
• Expr.meta.output_name for struct fields (#24064)
• Ensure upcast operations on pl.Date default to microsecond precision (#23981)
• Add peak_{min,max} support for booleans (#24068)
• Planner output type for mean with strange input type (#24052)
• Remove, deprecate or change eager Exprs to be lazy compatible (#24027)

📖 Documentation

• Fix formatting of Series.value_counts examples (#24245)
• Add hint to use DataFrame/Series constructors in from_arrow docstring (#22942)
• Update GPU un/supported features (#24195)
• Add DataFrame.map_columns to API (#24128)
• Update multiple pages in the Polars Cloud user guide (#23661)
• Fix str.find_many() docstring example (#24092)

📦 Build system

• Drop binary support for macos_x86-64 (#24257)

🛠️ Other improvements

• Remove unnecessary parentheses (#24244)
• Make non-nested shift{,_and_fill} ops generic (#24224)
• Remove unused Wrap (#24214)
• Allow upcasting null-typed columns to nested column types in scans (#24185)
• Automatically label a few more types of PR (#24147)
• Update toolchain (#24156)
• Add order_sensitive property for AExpr (#24116)
• Mark more tests as not possible on cloud (#24103)
• Turn AggExpr::Count from tuple to struct (#24096)
• Mark tests that may fail in cloud (#24067)
• Extend read database tests to capture more ADBC functionality (#24002)
• Make CI perf failures more lenient (#24066)
• Fix hive partition string encoding in CI by upgrading deltalake (#24018)
• Make tests with sinks run on cloud again (#24048)

Thank you to all our contributors for making this release possible!
@Kevin-Patyk, @agossard, @alexander-beedie, @aparna2198, @borchero, @coastalwhite, @deanm0000, @dsprenkels, @henryharbeck, @jjurm, @kdn36, @math-hiyoko, @mcrumiller, @mroeschke, @nameexhaustion, @orlp, @r-brink, @ritchie46, @stijnherfst, @vdrn and @wence-

Python Polars 1.32.3

🚀 Performance improvements

• Lower .sort(maintain_order=True).head() to streaming top_k (#24014)
• Lower top-k to streaming engine (#23979)
• Allow order pass through Filters and relax to row-seperable instead of elementwise (#23969)

✨ Enhancements

• Add native streaming for peaks_{min,max} (#24039)
• IR graph arrows, monospace font, box nodes (#24021)
• Add DataTypeExpr.default_value (#23973)
• Lower rle to a native streaming engine node (#23929)
• Add support for Int128 to pyo3-polars (#23959)

🐞 Bug fixes

• Scan of multiple sources with null datatype (#24065)
• Categorical in nested data in row encoding (#24051)
• Missing length update in builder for pl.Array repetition (#24055)
• Race condition in global categories init (#24045)
• Revert "fix: Don't encode entire CategoricalMapping when going to Arrow (#24036)" (#24044)
• Error when using named functions (#24041)
• Don't encode entire CategoricalMapping when going to Arrow (#24036)
• Fix cast on arithmetic with lit (#23941)
• Incorrect slice-slice pushdown (#24032)
• Dedup common cache subplan in IR graph (#24028)
• Allow join on Decimal in in-memory engine (#24026)
• Fix datatypes for eval.list in aggregation context (#23911)
• Allocator capsule fallback panic (#24022)
• Accept another zlib "magic header" file signature (#24013)
• Fix truediv dtypes so cast in list.eval is not dropped (#23936)
• Don't reuse cached return_dtype for expanded map expressions (#24010)
• Cache id is not a valid dot node id (#24005)
• Align map_elements with and without return_dtype (#24007)
• Fix column dtype lifetime for csv_write segfault on Categorical (#23986)
• Allow serializing LazyGroupBy.map_groups (#23964)
• Correct allocator name in PyCapsule (#23968)
• Mismatched types for write function for windows (#23915)
• Fix unpivot panic when index= column not found (#23958)

📖 Documentation

• Fix a typo in "lazy/execution" user-guide page (#23983)

🛠️ Other improvements

• Update pyo3-polars versions (#24031)
• Remove insert_error_function (#24023)
• Remove cache hits, clean up in-mem prefill (#24019)
• Use .venv instead of venv in pyo3-polars examples (#24024)
• Fix test failing mypy (#24017)
• Remove outdated comment (#23998)
• Add a _plr.pyi to remove mypy issues (#23970)
• Don't define CountStar as dyn OptimizationRule (#23976)
• Rename atol and rtol to abs_tol and rel_tol (#23961)
• Introduce Row{Encode,Decode} as FunctionExpr (#23933)
• Dispatch through pl.map_batches and AnonymousColumnsUdf (#23867)

Thank you to all our contributors for making this release possible!
@JakubValtar, @Kevin-Patyk, @borchero, @cmdlineluser, @coastalwhite, @iishutov, @jarondl, @kdn36, @orlp, @rawhuul, @ritchie46 and @stijnherfst

Python Polars 1.32.2

🐞 Bug fixes

• Return correct python package version (#23951)

📖 Documentation

• Add arr.len() on the website (#23944)

Thank you to all our contributors for making this release possible!
@coastalwhite and @etiennebacher

Python Polars 1.32.1

🚀 Performance improvements

• Optimise BytecodeParser usage from warn_on_inefficient_map (#23809)
• Lower extend_constant to the streaming engine (#23824)
• Lower pl.repeat to streaming engine (#23804)
• Remove redundant clone (#23771)

✨ Enhancements

• Lower rle_id to a native streaming node (#23894)
• Pass endpoint_url loaded from CredentialProviderAWS to scan/write_delta (#23812)
• Dispatch scan_iceberg to native by default (#23912)
• Lower unique_counts and value_counts to streaming engine (#23890)
• Support initializing from __arrow_c_schema__ protocol in pl.Schema (#23879)
• Better handle broken local package environment in show_versions (#23885)
• Implement dt.days_in_month function (#23119)
• Making Expr.rolling_*_by methods available to pl.Series (#23742)
• Fix errors on native scan_iceberg (#23811)
• Reinterpret binary data to fixed size numerical array (#22840)
• Make rolling_map serializable (#23848)
• Ensure CachingCredentialProvider returns copied credentials dict (#23817)
• Change typing for .remote() from LazyFrameExt to LazyFrameRemote (#23825)
• Implement repeat_by for Array and Null (#23794)
• Add DeprecationWarning on passing physical ordering to Categorical (#23779)
• Pre-filtered decode and row group skipping with Iceberg / Delta / scans with cast options (#23792)
• Update BytecodeParser opcode awareness for upcoming Python 3.14 (#23782)

🐞 Bug fixes

• Categorical namespace functions fail on Enum columns (#23925)
• Properly set sumwise complete on filter for missing columns (#23877)
• Restore Arrow-FFI-based Python<->Rust conversion in pyo3-polars (#23881)
• Group By with filters (#23917)
• Fix read_csv ignoring Decimal schema for header-only data (#23886)
• Ensure collect() native Iceberg always scans latest when no snapshot_id is given (#23907)
• Writing List(Array) columns to JSON without panic (#23875)
• Fill Iceberg missing fields with partition values if present in metadata (#23900)
• Create file for streaming sink even if unspawned (#23672)
• Update cloud testing environment (#23908)
• Parquet filtering on multiple RGs with literal predicate (#23903)
• Incorrect datatype passed to libc::write (#23904)
• Properly feature gate TZ_AWARE_RE usage (#23888)
• Improve identification of "non group-key" aggregates in SQL GROUP BY queries (#23191)
• Spawning tokio task outside reactor (#23884)
• Correctly raise DuplicateError on asof_join with suffix="" (#23864)
• Fix errors on native scan_iceberg (#23811)
• Fix index out of bounds panic filtering parquet (#23850)
• Fix error on empty range requests (#23844)
• Fix handling of hive partitioning hive_start_idx parameter (#23843)
• Allow encoding of pl.Enum with smaller physicals (#23829)
• Filter sorted flag from physical in CategoricalChunked (#23827)
• Remove accidental todo! in repeat node (#23822)
• Make meta.pop operate on Expr only (#23808)
• Stack overflow in DslPlan serde (#23801)
• Clear credentials cached in Python when rebuilding object store (#23756)
• Datetime selectors with mixed timezone info (#23774)
• Support i128 in asof join (#23770)
• Remove sleep for credential refresh (#23768)

📖 Documentation

• Improve StackOverflow links in contributing guide (#23895)
• Fix pyo3 documentation page link (#23839)
• Document the pureness requirements of udfs (#23787)
• Correct the name.* methods on their removal of aliases (#23773)

📦 Build system

• Workaround for pyiceberg make requirements on Python 3.13 (#23810)
• Add pyiceberg to dev dependencies (#23791)

🛠️ Other improvements

• Ensure clippy and rustfmt run in CI when changing pyo3-polars (#23930)
• Fix pyo3-polars proc-macro re-exports (#23918)
• Rewrite evaluate_on_groups for .gather / .get (#23700)
• Move Python C API to python-polars (#23876)
• Improve/fix internal LRUCache implementation and move into "_utils" module (#23813)
• Relax constraint on maximum Python version for numba (#23838)
• Automatically tag PRs mentioning "SQL" with the appropriate label (#23816)
• Update typos package (#23818)
• Fix typos path (#23803)
• Remove deserialize_with_unknown_fields (#23802)
• Add pyiceberg to dev dependencies (#23791)
• Remove old schema file (#23798)
• Mark more tests as ready for cloud (#23743)
• Reduce required deps for pyo3-polars (#23761)

Thank you to all our contributors for making this release possible!
@JakubValtar, @Kevin-Patyk, @Liyixin95, @alexander-beedie, @cgevans, @cmdlineluser, @coastalwhite, @eitsupi, @gfvioli, @itamarst, @jimmmmmmmmmmmy, @kdn36, @math-hiyoko, @mcrumiller, @mpasa, @mrkn, @nameexhaustion, @orlp, @pka, @pomo-mondreganto, @ritchie46 and @stijnherfst

Rust Polars 0.50.0

🏆 Highlights

• Make Selector a concrete part of the DSL (#23351)
• Rework Categorical/Enum to use (Frozen)Categories (#23016)

🚀 Performance improvements

• Lower Expr.slice to streaming engine (#23683)
• Elide bound check (#23653)
• Preserve Column repr in ColumnTransform operations (#23648)
• Lower any() and all() to streaming engine (#23640)
• Lower row-separable functions in streaming engine (#23633)
• Lower int_range(len()) to with_row_index (#23576)
• Avoid double field resolution in with_columns (#23530)
• Rolling quantile lower time complexity (#23443)
• Use single-key optimization with Categorical (#23436)
• Improve null-preserving identification for boolean functions (#23317)
• Improve boolean bitwise aggregate performance (#23325)
• Enable Parquet expressions and dedup is_in values in Parquet predicates (#23293)
• Re-write join types during filter pushdown (#23275)
• Generate PQ ZSTD decompression context once (#23200)
• Trigger cache/cse optimizations when multiplexing (#23274)
• Cache FileInfo upon DSL -> IR conversion (#23263)
• Push more filters past joins (#23240)

✨ Enhancements

• Expand on DataTypeExpr (#23249)
• Lower row-separable functions in streaming engine (#23633)
• Add scalar checks to range expressions (#23632)
• Expose POLARS_DOT_SVG_VIEWER to automatically dispatch to SVG viewer (#23592)
• Implement mean function in arr namespace (#23486)
• Implement vec_hash for List and Array (#23578)
• Add unstable pl.row_index() expression (#23556)
• Add Categories on the Python side (#23543)
• Implement partitioned sinks for the in-memory engine (#23522)
• Expose IRFunctionExpr::Rank in the python visitor (#23512)
• Raise and Warn on UDF's without return_dtype set (#23353)
• IR pruning (#23499)
• Expose IRFunctionExpr::FillNullWithStrategy in the python visitor (#23479)
• Support min/max reducer for null dtype in streaming engine (#23465)
• Implement streaming Categorical/Enum min/max (#23440)
• Allow cast to Categorical inside list.eval (#23432)
• Support pathlib.Path as source for read/scan_delta() (#23411)
• Enable default set of ScanCastOptions for native scan_iceberg() (#23416)
• Pass payload in ExprRegistry (#23412)
• Support reading nanosecond/Int96 timestamps and schema evolved datasets in scan_delta() (#23398)
• Support row group skipping with filters when cast_options is given (#23356)
• Execute bitwise reductions in streaming engine (#23321)
• Use scan_parquet().collect_schema() for read_parquet_schema (#23359)
• Add dtype to str.to_integer() (#22239)
• Add arr.slice, arr.head and arr.tail methods to arr namespace (#23150)
• Add is_close method (#23273)
• Drop superfluous casts from optimized plan (#23269)
• Added drop_nulls option to to_dummies (#23215)
• Support comma as decimal separator for CSV write (#23238)
• Don't format keys if they're empty in dot (#23247)
• Improve arity simplification (#23242)

🐞 Bug fixes

• Fix credential refresh logic (#23730)
• Fix to_datetime() fallible identification (#23735)
• Correct output datatype for dt.with_time_unit (#23734)
• Fix incorrect native Iceberg scan from tables with renamed/dropped columns/fields (#23713)
• Allow DataType expressions with selectors (#23720)
• Match output type to engine for interpolate on Decimal (#23706)
• Remaining bugs in with_exprs_and_input and pruning (#23710)
• Match output dtype to engine for cum_sum_horizontal (#23686)
• Field names for pl.struct in group-by (#23703)
• Fix output for str.extract_groups with empty string pattern (#23698)
• Match output type to engine for rolling_map (#23702)
• Fix incorrect join on single Int128 column for in-memory engine (#23694)
• Match output field name to lhs for BusinessDaycount (#23679)
• Correct the planner output datatype for strptime (#23676)
• Sort and Scan with_exprs_and_input (#23675)
• Revert to old behavior with name.keep (#23670)
• Fix panic loading from arrow Map containing timestamps (#23662)
• Selectors in self part of list.eval (#23668)
• Fix output field dtype for ToInteger (#23664)
• Allow decimal_comma with , separator in read_csv (#23657)
• Fix handling of UTF-8 in write_csv to IO[str] (#23647)
• Selectors in {Lazy,Data}Frame.filter (#23631)
• Stop splitfields iterator at eol in simd branch (#23652)
• Correct output datatype of dt.year and dt.mil (#23646)
• Logic of broadcast_rhs in binary functions to correct list.set_intersection for list[str] columns (#23584)
• Order-preserving equi-join didn't always flush final matches (#23639)
• Fix ColumnNotFound error when joining on col().cast() (#23622)
• Fix agg groups on when/then in group_by context (#23628)
• Output type for sign (#23572)
• Apply agg_fn on null values in pivot (#23586)
• Remove nonsensical duration variance (#23621)
• Don't panic when sinking nested categorical to Parquet (#23610)
• Correctly set value count output field name (#23611)
• Casting unused columns in to_torch (#23606)
• Allow inferring of hours-only timezone offset (#23605)
• Bug in Categorical <-> str compare with nulls (#23609)
• Honor n=0 in all cases of str.replace (#23598)
• Remove arbitrary 25 item limit from implicit Python list -> Series infer (#23603)
• Relabel duplicate sequence IDs in distributor (#23593)
• Round-trip Enum and Categorical metadata in plugins (#23588)
• Fix incorrect join_asof with by followed by head/slice (#23585)
• Allow writing nested Int128 data to Parquet (#23580)
• Enum serialization assert (#23574)
• Output type for peak_min / peak_max (#23573)
• Make Scalar Categorical, Enum and Struct values serializable (#23565)
• Preserve row order within partition when sinking parquet (#23462)
• Panic in create_multiple_physical_plans when branching from a single cache node (#23561)
• Prevent in-mem partition sink deadlock (#23562)
• Update AWS cloud documentation (#23563)
• Correctly handle null values when comparing structs (#23560)
• Make fold/reduce/cum_reduce/cum_fold serializable (#23524)
• Make Expr.append serializable (#23515)
• Float by float division dtype (#23529)
• Division on empty DataFrame generating null row (#23516)
• Partition sink copy_exprs and with_exprs_and_input (#23511)
• Unreachable with pl.self_dtype (#23507)
• Rolling median incorrect min_samples with nulls (#23481)
• Make Int128 roundtrippable via Parquet (#23494)
• Fix panic when common subplans contain IEJoins (#23487)
• Properly handle non-finite floats in rolling_sum/mean (#23482)
• Make read_csv_batched respect skip_rows and skip_lines (#23484)
• Always use cloudpickle for the python objects in cloud plans (#23474)
• Support string literals in index_of() on categoricals (#23458)
• Don't panic for finish_callback with nested datatypes (#23464)
• Support min/max aggregation for DataFrame/LazyFrame Categoricals (#23455)
• Fix var/moment dtypes (#23453)
• Fix agg_groups dtype (#23450)
• Clear cached_schema when apply changes dtype (#23439)
• Allow structured conversion to/from numpy with Array types, preserving shape (#23438)
• Null handling in full-null group_by_dynamic mean/sum (#23435)
• Enable default set of ScanCastOptions for native scan_iceberg() (#23416)
• Fix index calculation for nearest interpolation (#23418)
• Fix compilation failure with --no-default-features and --features lazy,strings (#23384)
• Parse parquet footer length into unsigned integer (#23357)
• Fix incorrect results with group_by aggregation on empty groups (#23358)
• Fix boolean min() in group_by aggregation (streaming) (#23344)
• Respect data-model in map_elements (#23340)
• Properly join URI paths in PlPath (#23350)
• Ignore null values in bitwise aggregation on bools (#23324)
• Fix panic filtering after left join (#23310)
• Out-of-bounds index in hot hash table (#23311)
• Fix scanning '?' from cloud with glob=False (#23304)
• Fix filters on inserted columns did not remove rows (#23303)
• Don't ignore return_dtype (#23309)
• Use safe parsing for get_normal_components (#23284)
• Fix output column names/order of streaming coalesced right-join (#23278)
• Restore concat_arr inputs expansion (#23271)

📖 Documentation

• Point the R Polars version on R-multiverse (#23660)
• Update GPU docs for RAPIDS CUDA 11 deprecation (#23620)
• Add page about billing to Polars Cloud user guide (#23564)
• Small user-guide improvement and fixes (#23549)
• Correct note in from_pandas about data being cloned (#23552)
• Fix a few typos in the "Streaming" section (#23536)
• Update streaming page (#23535)
• Update structure of Polars Cloud documentation (#23496)
• Update when_then in user guide (#23245)

📦 Build system

• Update all rand code (#23387)
• Bump up rand & rand_distr (#22619)

🛠️ Other improvements

• Remove incorrect DeletionFilesList::slice (#23796)
• Remove old schema file (#23798)
• Remove Default for StreamingExecutionState (#23729)
• Explicit match to smaller dtypes before cast to Int32 in asof join (#23776)
• Expose PlPathRef via polars::prelude (#23754)
• Add hashes json (#23758)
• Add AExpr::is_expr_equal_to (#23740)
• Fix rank test to respect maintain order (#23723)
• IR inputs and exprs iterators (#23722)
• Store more granular schema hashes to reduce merge conflicts (#23709)
• Add assertions for unique ID (#23711)
• Use RelaxedCell in multiscan (#23712)
• Debug assert ColumnTransform cast is non-strict (#23717)
• Use UUID for UniqueID (#23704)
• Remove scan id (#23697)
• Propagate Iceberg physical ID schema to IR (#23671)
• Remove unused and confusing match arm (#23691)
• Remove unused ALLOW_GROUP_AWARE flag (#23690)
• Remove unused evaluate_inline (#23687)
• Remove unused field from AggregationContext (#23685)
• Remove `nod...

Python Polars 1.32.0

🏆 Highlights

• Make Selector a concrete part of the DSL (#23351)
• Rework Categorical/Enum to use (Frozen)Categories (#23016)

🚀 Performance improvements

• Lower Expr.slice to streaming engine (#23683)
• Elide bound check (#23653)
• Preserve Column repr in ColumnTransform operations (#23648)
• Lower any() and all() to streaming engine (#23640)
• Lower row-separable functions in streaming engine (#23633)
• Lower int_range(len()) to with_row_index (#23576)
• Avoid double field resolution in with_columns (#23530)
• Rolling quantile lower time complexity (#23443)
• Use single-key optimization with Categorical (#23436)
• Improve null-preserving identification for boolean functions (#23317)
• Improve boolean bitwise aggregate performance (#23325)
• Enable Parquet expressions and dedup is_in values in Parquet predicates (#23293)
• Re-write join types during filter pushdown (#23275)
• Generate PQ ZSTD decompression context once (#23200)
• Trigger cache/cse optimizations when multiplexing (#23274)
• Cache FileInfo upon DSL -> IR conversion (#23263)
• Push more filters past joins (#23240)
• Optimize Bitmap::make_mut (#23138)

✨ Enhancements

• Add Python-side caching for credentials and provider auto-initialization (#23736)
• Expand on DataTypeExpr (#23249)
• Lower row-separable functions in streaming engine (#23633)
• Add scalar checks to range expressions (#23632)
• Expose POLARS_DOT_SVG_VIEWER to automatically dispatch to SVG viewer (#23592)
• Implement mean function in arr namespace (#23486)
• Implement vec_hash for List and Array (#23578)
• Add unstable pl.row_index() expression (#23556)
• Add Categories on the Python side (#23543)
• Implement partitioned sinks for the in-memory engine (#23522)
• Raise and Warn on UDF's without return_dtype set (#23353)
• IR pruning (#23499)
• Support min/max reducer for null dtype in streaming engine (#23465)
• Implement streaming Categorical/Enum min/max (#23440)
• Allow cast to Categorical inside list.eval (#23432)
• Support pathlib.Path as source for read/scan_delta() (#23411)
• Enable default set of ScanCastOptions for native scan_iceberg() (#23416)
• Pass payload in ExprRegistry (#23412)
• Support reading nanosecond/Int96 timestamps and schema evolved datasets in scan_delta() (#23398)
• Support row group skipping with filters when cast_options is given (#23356)
• Execute bitwise reductions in streaming engine (#23321)
• Use scan_parquet().collect_schema() for read_parquet_schema (#23359)
• Add dtype to str.to_integer() (#22239)
• Add arr.slice, arr.head and arr.tail methods to arr namespace (#23150)
• Add is_close method (#23273)
• Drop superfluous casts from optimized plan (#23269)
• Added drop_nulls option to to_dummies (#23215)
• Support comma as decimal separator for CSV write (#23238)
• Don't format keys if they're empty in dot (#23247)
• Improve arity simplification (#23242)
• Allow expression input for length parameter in pad_start, pad_end, and zfill (#23182)

🐞 Bug fixes

• Load _expiry_time from botocore Credentials in CredentialProviderAWS (#23753)
• Fix credential refresh logic (#23730)
• Fix to_datetime() fallible identification (#23735)
• Correct output datatype for dt.with_time_unit (#23734)
• Fix incorrect native Iceberg scan from tables with renamed/dropped columns/fields (#23713)
• Allow DataType expressions with selectors (#23720)
• Match output type to engine for interpolate on Decimal (#23706)
• Remaining bugs in with_exprs_and_input and pruning (#23710)
• Match output dtype to engine for cum_sum_horizontal (#23686)
• Field names for pl.struct in group-by (#23703)
• Fix output for str.extract_groups with empty string pattern (#23698)
• Match output type to engine for rolling_map (#23702)
• Moved passing DeltaTable._storage_options (#23673)
• Fix incorrect join on single Int128 column for in-memory engine (#23694)
• Match output field name to lhs for BusinessDaycount (#23679)
• Correct the planner output datatype for strptime (#23676)
• Sort and Scan with_exprs_and_input (#23675)
• Revert to old behavior with name.keep (#23670)
• Fix panic loading from arrow Map containing timestamps (#23662)
• Selectors in self part of list.eval (#23668)
• Fix output field dtype for ToInteger (#23664)
• Allow decimal_comma with , separator in read_csv (#23657)
• Fix handling of UTF-8 in write_csv to IO[str] (#23647)
• Selectors in {Lazy,Data}Frame.filter (#23631)
• Stop splitfields iterator at eol in simd branch (#23652)
• Correct output datatype of dt.year and dt.mil (#23646)
• Logic of broadcast_rhs in binary functions to correct list.set_intersection for list[str] columns (#23584)
• Order-preserving equi-join didn't always flush final matches (#23639)
• Fix ColumnNotFound error when joining on col().cast() (#23622)
• Fix agg groups on when/then in group_by context (#23628)
• Output type for sign (#23572)
• Apply agg_fn on null values in pivot (#23586)
• Remove nonsensical duration variance (#23621)
• Don't panic when sinking nested categorical to Parquet (#23610)
• Correctly set value count output field name (#23611)
• Casting unused columns in to_torch (#23606)
• Allow inferring of hours-only timezone offset (#23605)
• Bug in Categorical <-> str compare with nulls (#23609)
• Honor n=0 in all cases of str.replace (#23598)
• Remove arbitrary 25 item limit from implicit Python list -> Series infer (#23603)
• Relabel duplicate sequence IDs in distributor (#23593)
• Round-trip Enum and Categorical metadata in plugins (#23588)
• Fix incorrect join_asof with by followed by head/slice (#23585)
• Change return typing of get_index_type() from DataType to PolarsIntegerType (#23558)
• Allow writing nested Int128 data to Parquet (#23580)
• Enum serialization assert (#23574)
• Output type for peak_min / peak_max (#23573)
• Make Scalar Categorical, Enum and Struct values serializable (#23565)
• Preserve row order within partition when sinking parquet (#23462)
• Prevent in-mem partition sink deadlock (#23562)
• Update AWS cloud documentation (#23563)
• Correctly handle null values when comparing structs (#23560)
• Make fold/reduce/cum_reduce/cum_fold serializable (#23524)
• Make Expr.append serializable (#23515)
• Float by float division dtype (#23529)
• Division on empty DataFrame generating null row (#23516)
• Partition sink copy_exprs and with_exprs_and_input (#23511)
• Unreachable with pl.self_dtype (#23507)
• Rolling median incorrect min_samples with nulls (#23481)
• Make Int128 roundtrippable via Parquet (#23494)
• Fix panic when common subplans contain IEJoins (#23487)
• Properly handle non-finite floats in rolling_sum/mean (#23482)
• Make read_csv_batched respect skip_rows and skip_lines (#23484)
• Always use cloudpickle for the python objects in cloud plans (#23474)
• Support string literals in index_of() on categoricals (#23458)
• Don't panic for finish_callback with nested datatypes (#23464)
• Pass DeltaTable._storage_options if no storage_options are provided (#23456)
• Support min/max aggregation for DataFrame/LazyFrame Categoricals (#23455)
• Fix var/moment dtypes (#23453)
• Fix agg_groups dtype (#23450)
• Fix incorrect _get_path_scheme (#23444)
• Fix missing overload defaults in read_ods and tree_format (#23442)
• Clear cached_schema when apply changes dtype (#23439)
• Allow structured conversion to/from numpy with Array types, preserving shape (#23438)
• Null handling in full-null group_by_dynamic mean/sum (#23435)
• Enable default set of ScanCastOptions for native scan_iceberg() (#23416)
• Fix index calculation for nearest interpolation (#23418)
• Overload for eager default in Schema.to_frame was False instead of True (#23413)
• Fix read_excel overloads so that passing list[str] to sheet_name does not raise (#23388)
• Removed special handling for bytes like objects in read_ndjson (#23361)
• Parse parquet footer length into unsigned integer (#23357)
• Fix incorrect results with group_by aggregation on empty groups (#23358)
• Fix boolean min() in group_by aggregation (streaming) (#23344)
• Respect data-model in map_elements (#23340)
• Properly join URI paths in PlPath (#23350)
• Ignore null values in bitwise aggregation on bools (#23324)
• Fix panic filtering after left join (#23310)
• Out-of-bounds index in hot hash table (#23311)
• Fix scanning '?' from cloud with glob=False (#23304)
• Fix filters on inserted columns did not remove rows (#23303)
• Don't ignore return_dtype (#23309)
• Raise error instead of return in Series class (#23301)
• Use safe parsing for get_normal_components (#23284)
• Fix output column names/order of streaming coalesced right-join (#23278)
• Restore concat_arr inputs expansion (#23271)
• Expose FieldsMapper (#23232)
• Fix time zone handling in dt.iso_year and dt.is_leap_year (#23125)

📖 Documentation

• Fix str.replace_many examples trigger deprecation warning (#23695)
• Point the R Polars version on R-multiverse (#23660)
• Update example for writing to cloud storage (#20265)
• Update GPU docs for RAPIDS CUDA 11 deprecation (#23620)
• Add docs of Expr.list.filter and Series.list.filter (#23589)
• Add page about billing to Polars Cloud user guide (#23564)
• Small user-guide improvement and fixes (#23549)
• Correct note in from_pandas about data being cloned (#23552)
• Fix a few typos in the "Streaming" section (#23536)
• Update streaming page (#23535)
• Update structure of Polars Cloud documentation (#23496)
• Update example code in pandas migration guide (#23403)
• Correct plugins user guide to reflect that teaching Expr.language is in a different section (#23377)
• Add example of using OR in join_where (#23375)
• Update when_then in user guide (#23245)

📦 Build system

• Update all rand code (#23387)

🛠️ Other improvements

• Remove unused functions from the rust side (#2...

Python Polars 1.32.0-beta.1

🏆 Highlights

• Make Selector a concrete part of the DSL (#23351)
• Rework Categorical/Enum to use (Frozen)Categories (#23016)

🚀 Performance improvements

• Lower Expr.slice to streaming engine (#23683)
• Elide bound check (#23653)
• Preserve Column repr in ColumnTransform operations (#23648)
• Lower any() and all() to streaming engine (#23640)
• Lower row-separable functions in streaming engine (#23633)
• Lower int_range(len()) to with_row_index (#23576)
• Avoid double field resolution in with_columns (#23530)
• Rolling quantile lower time complexity (#23443)
• Use single-key optimization with Categorical (#23436)
• Improve null-preserving identification for boolean functions (#23317)
• Improve boolean bitwise aggregate performance (#23325)
• Enable Parquet expressions and dedup is_in values in Parquet predicates (#23293)
• Re-write join types during filter pushdown (#23275)
• Generate PQ ZSTD decompression context once (#23200)
• Trigger cache/cse optimizations when multiplexing (#23274)
• Cache FileInfo upon DSL -> IR conversion (#23263)
• Push more filters past joins (#23240)
• Optimize Bitmap::make_mut (#23138)

✨ Enhancements

• Add Python-side caching for credentials and provider auto-initialization (#23736)
• Expand on DataTypeExpr (#23249)
• Lower row-separable functions in streaming engine (#23633)
• Add scalar checks to range expressions (#23632)
• Expose POLARS_DOT_SVG_VIEWER to automatically dispatch to SVG viewer (#23592)
• Implement mean function in arr namespace (#23486)
• Implement vec_hash for List and Array (#23578)
• Add unstable pl.row_index() expression (#23556)
• Add Categories on the Python side (#23543)
• Implement partitioned sinks for the in-memory engine (#23522)
• Raise and Warn on UDF's without return_dtype set (#23353)
• IR pruning (#23499)
• Support min/max reducer for null dtype in streaming engine (#23465)
• Implement streaming Categorical/Enum min/max (#23440)
• Allow cast to Categorical inside list.eval (#23432)
• Support pathlib.Path as source for read/scan_delta() (#23411)
• Enable default set of ScanCastOptions for native scan_iceberg() (#23416)
• Pass payload in ExprRegistry (#23412)
• Support reading nanosecond/Int96 timestamps and schema evolved datasets in scan_delta() (#23398)
• Support row group skipping with filters when cast_options is given (#23356)
• Execute bitwise reductions in streaming engine (#23321)
• Use scan_parquet().collect_schema() for read_parquet_schema (#23359)
• Add dtype to str.to_integer() (#22239)
• Add arr.slice, arr.head and arr.tail methods to arr namespace (#23150)
• Add is_close method (#23273)
• Drop superfluous casts from optimized plan (#23269)
• Added drop_nulls option to to_dummies (#23215)
• Support comma as decimal separator for CSV write (#23238)
• Don't format keys if they're empty in dot (#23247)
• Improve arity simplification (#23242)
• Allow expression input for length parameter in pad_start, pad_end, and zfill (#23182)

🐞 Bug fixes

• Load _expiry_time from botocore Credentials in CredentialProviderAWS (#23753)
• Fix credential refresh logic (#23730)
• Fix to_datetime() fallible identification (#23735)
• Correct output datatype for dt.with_time_unit (#23734)
• Fix incorrect native Iceberg scan from tables with renamed/dropped columns/fields (#23713)
• Allow DataType expressions with selectors (#23720)
• Match output type to engine for interpolate on Decimal (#23706)
• Remaining bugs in with_exprs_and_input and pruning (#23710)
• Match output dtype to engine for cum_sum_horizontal (#23686)
• Field names for pl.struct in group-by (#23703)
• Fix output for str.extract_groups with empty string pattern (#23698)
• Match output type to engine for rolling_map (#23702)
• Moved passing DeltaTable._storage_options (#23673)
• Fix incorrect join on single Int128 column for in-memory engine (#23694)
• Match output field name to lhs for BusinessDaycount (#23679)
• Correct the planner output datatype for strptime (#23676)
• Sort and Scan with_exprs_and_input (#23675)
• Revert to old behavior with name.keep (#23670)
• Fix panic loading from arrow Map containing timestamps (#23662)
• Selectors in self part of list.eval (#23668)
• Fix output field dtype for ToInteger (#23664)
• Allow decimal_comma with , separator in read_csv (#23657)
• Fix handling of UTF-8 in write_csv to IO[str] (#23647)
• Selectors in {Lazy,Data}Frame.filter (#23631)
• Stop splitfields iterator at eol in simd branch (#23652)
• Correct output datatype of dt.year and dt.mil (#23646)
• Logic of broadcast_rhs in binary functions to correct list.set_intersection for list[str] columns (#23584)
• Order-preserving equi-join didn't always flush final matches (#23639)
• Fix ColumnNotFound error when joining on col().cast() (#23622)
• Fix agg groups on when/then in group_by context (#23628)
• Output type for sign (#23572)
• Apply agg_fn on null values in pivot (#23586)
• Remove nonsensical duration variance (#23621)
• Don't panic when sinking nested categorical to Parquet (#23610)
• Correctly set value count output field name (#23611)
• Casting unused columns in to_torch (#23606)
• Allow inferring of hours-only timezone offset (#23605)
• Bug in Categorical <-> str compare with nulls (#23609)
• Honor n=0 in all cases of str.replace (#23598)
• Remove arbitrary 25 item limit from implicit Python list -> Series infer (#23603)
• Relabel duplicate sequence IDs in distributor (#23593)
• Round-trip Enum and Categorical metadata in plugins (#23588)
• Fix incorrect join_asof with by followed by head/slice (#23585)
• Change return typing of get_index_type() from DataType to PolarsIntegerType (#23558)
• Allow writing nested Int128 data to Parquet (#23580)
• Enum serialization assert (#23574)
• Output type for peak_min / peak_max (#23573)
• Make Scalar Categorical, Enum and Struct values serializable (#23565)
• Preserve row order within partition when sinking parquet (#23462)
• Prevent in-mem partition sink deadlock (#23562)
• Update AWS cloud documentation (#23563)
• Correctly handle null values when comparing structs (#23560)
• Make fold/reduce/cum_reduce/cum_fold serializable (#23524)
• Make Expr.append serializable (#23515)
• Float by float division dtype (#23529)
• Division on empty DataFrame generating null row (#23516)
• Partition sink copy_exprs and with_exprs_and_input (#23511)
• Unreachable with pl.self_dtype (#23507)
• Rolling median incorrect min_samples with nulls (#23481)
• Make Int128 roundtrippable via Parquet (#23494)
• Fix panic when common subplans contain IEJoins (#23487)
• Properly handle non-finite floats in rolling_sum/mean (#23482)
• Make read_csv_batched respect skip_rows and skip_lines (#23484)
• Always use cloudpickle for the python objects in cloud plans (#23474)
• Support string literals in index_of() on categoricals (#23458)
• Don't panic for finish_callback with nested datatypes (#23464)
• Pass DeltaTable._storage_options if no storage_options are provided (#23456)
• Support min/max aggregation for DataFrame/LazyFrame Categoricals (#23455)
• Fix var/moment dtypes (#23453)
• Fix agg_groups dtype (#23450)
• Fix incorrect _get_path_scheme (#23444)
• Fix missing overload defaults in read_ods and tree_format (#23442)
• Clear cached_schema when apply changes dtype (#23439)
• Allow structured conversion to/from numpy with Array types, preserving shape (#23438)
• Null handling in full-null group_by_dynamic mean/sum (#23435)
• Enable default set of ScanCastOptions for native scan_iceberg() (#23416)
• Fix index calculation for nearest interpolation (#23418)
• Overload for eager default in Schema.to_frame was False instead of True (#23413)
• Fix read_excel overloads so that passing list[str] to sheet_name does not raise (#23388)
• Removed special handling for bytes like objects in read_ndjson (#23361)
• Parse parquet footer length into unsigned integer (#23357)
• Fix incorrect results with group_by aggregation on empty groups (#23358)
• Fix boolean min() in group_by aggregation (streaming) (#23344)
• Respect data-model in map_elements (#23340)
• Properly join URI paths in PlPath (#23350)
• Ignore null values in bitwise aggregation on bools (#23324)
• Fix panic filtering after left join (#23310)
• Out-of-bounds index in hot hash table (#23311)
• Fix scanning '?' from cloud with glob=False (#23304)
• Fix filters on inserted columns did not remove rows (#23303)
• Don't ignore return_dtype (#23309)
• Raise error instead of return in Series class (#23301)
• Use safe parsing for get_normal_components (#23284)
• Fix output column names/order of streaming coalesced right-join (#23278)
• Restore concat_arr inputs expansion (#23271)
• Expose FieldsMapper (#23232)
• Fix time zone handling in dt.iso_year and dt.is_leap_year (#23125)

📖 Documentation

• Fix str.replace_many examples trigger deprecation warning (#23695)
• Point the R Polars version on R-multiverse (#23660)
• Update example for writing to cloud storage (#20265)
• Update GPU docs for RAPIDS CUDA 11 deprecation (#23620)
• Add docs of Expr.list.filter and Series.list.filter (#23589)
• Add page about billing to Polars Cloud user guide (#23564)
• Small user-guide improvement and fixes (#23549)
• Correct note in from_pandas about data being cloned (#23552)
• Fix a few typos in the "Streaming" section (#23536)
• Update streaming page (#23535)
• Update structure of Polars Cloud documentation (#23496)
• Update example code in pandas migration guide (#23403)
• Correct plugins user guide to reflect that teaching Expr.language is in a different section (#23377)
• Add example of using OR in join_where (#23375)
• Update when_then in user guide (#23245)

📦 Build system

• Update all rand code (#23387)

🛠️ Other improvements

• Add hashes json (#23758)
• Add `AExpr::is_expr...

Rust Polars 0.49.1

🚀 Performance improvements

• Optimize Bitmap::make_mut (#23138)

✨ Enhancements

• Allow expression input for length parameter in pad_start, pad_end, and zfill (#23182)

🐞 Bug fixes

• Expose FieldsMapper (#23232)
• Fix time zone handling in dt.iso_year and dt.is_leap_year (#23125)

📖 Documentation

• Update when_then in user guide (#23245)

🛠️ Other improvements

• Connect Python assert_dataframe_equal() to Rust back-end (#23207)
• Fix time zone handling in dt.iso_year and dt.is_leap_year (#23125)
• Update Rust Polars versions (#23229)

Thank you to all our contributors for making this release possible!
@Kevin-Patyk, @mcrumiller, @mrkn, @stijnherfst and @zyctree

Rust Polars 0.49.0

💥 Breaking changes

• Remove old streaming engine (#23103)

🚀 Performance improvements

• Improve streaming groupby CSE (#23092)
• Move row index materialization in post-apply to occur after slicing (#22995)
• Add first_(true|false)_idx to BooleanChunked and use in bool arg_(min|max) (#22907)
• Don't go through row encoding for most types on index_of (#22903)
• Optimise low-level null scans and arg_max for bools (when chunked) (#22897)
• Optimize multiscan performance (#22886)

✨ Enhancements

• Native implementation for Iceberg positional deletes (#23091)
• Remove old streaming engine (#23103)
• Make match_chunks public (#23101)
• Implement StructFunction expressions in into_py (#23022)
• Basic implementation of DataTypeExpr in Rust DSL (#23049)
• Add required: bool to ParquetFieldOverwrites (#23013)
• Support serializing name.map_fields (#22997)
• Support serializing Expr::RenameAlias (#22988)
• Remove duplicate verbose logging from FetchedCredentialsCache (#22973)
• Add keys column in finish_callback (#22968)
• Add extra_columns parameter to scan_parquet (#22699)
• Add CORR function to polars SQL (#22690)
• Add per partition sort and finish callback to sinks (#22789)
• Add and test DataFrame equality functionality (#22865)
• Support descendingly-sorted values in search_sorted() (#22825)
• Derive DSL schema (#22866)

🐞 Bug fixes

• Restrict custom aggregate_function in pivot to pl.element() (#23155)
• Don't leak SourceToken in in-memory sink linearize (#23201)
• Fix panic reading empty parquet with multiple boolean columns (#23159)
• Raise ComputeError instead of panicking in truncate when mixing month/week/day/sub-daily units (#23176)
• Materialize list.eval with unknown type (#23186)
• Only set sorting flag for 1st column with PQ SortingColumns (#23184)
• Typo in AExprBuilder (#23171)
• Null return from var/std on scalar column (#23158)
• Support Datetime broadcast in list.concat (#23137)
• Ensure projection pushdown maintains right table schema (#22603)
• Don't create i128 scalars if dtype-128 is not set (#23118)
• Add Null dtype support to arg_sort_by (#23107)
• Raise error by default on invalid CSV quotes (#22876)
• Fix group_by mean and median returning all nulls for Decimal dtype (#23093)
• Fix hive partition pruning not filtering out __HIVE_DEFAULT_PARTITION__ (#23074)
• Fix AssertionError when using scan_delta() on AWS with storage_options (#23076)
• Fix deadlock on collect(background=True) / collect_concurrently() (#23075)
• Incorrect null count in rolling_min/max (#23073)
• Preserve file:// in LazyFrame node traverser (#23072)
• Respect column order in register_io_source schema (#23057)
• Incorrect output when using sort with group_by and cum_sum (#23001)
• Implement owned arithmetic for Int128 (#23055)
• Do not schema-match structs with different field counts (#23018)
• Fix confusing error message on duplicate row_index (#23043)
• Add include_nulls to Agg::Count CSE check (#23032)
• View buffer exceeding 2^32 - 1 bytes in concatenate_view (#23017)
• Fix incorrect size_hint() for FlatIter (#23010)
• Fix incorrect result selecting pl.len() from scan_csv with skip_lines (#22949)
• Allow for IO plugins with reordered columns in streaming (#22987)
• Method str.zfill was inconsistent with Python and pandas when string contained leading '+' (#22985)
• Integer underflow in propagate_nulls (#22986)
• Fix cum_min and cum_max does not preserve inf or -inf values at series start (#22896)
• Setting compat_level=0 for sink_ipc (#22960)
• Support arrow Decimal32 and Decimal64 types (#22954)
• Update arrow format (#22941)
• Fix filter pushdown to IO plugins (#22910)
• Improve numeric stability rolling_mean<f32> (#22944)
• Allow subclasses in type equality checking (#22915)
• Return early in pl.Expr.__array_ufunc__ when only single input (#22913)
• Add inline implodes in type coercion (#22885)
• Correct int_ranges to raise error on invalid inputs (#22894)
• Set the sorted flag on Array after it is sorted (#22822)
• Don't silently overflow for temporal casts (#22901)
• Fix error using write_csv with storage_options (#22881)
• Schema resolution .over(mapping_strategy="join") with non-aggregations (#22875)
• Ensure rename behaves the same as select (#22852)

📖 Documentation

• Update when_then in user guide (#23245)
• Minor improvement to cum_count docstring example (#23099)
• Add missing entry for LazyFrame __getitem__ (#22924)

📦 Build system

• Actually disable ir_serde by default (#23046)
• Add a feature flag for serde_ignored (#22957)
• Fix warnings, update DSL version and schema hash (#22953)

🛠️ Other improvements

• Update Rust Polars versions (#23229)
• Change flake to use venv (#23219)
• Add default_alloc feature to py-polars (#23202)
• Added more descriptive error message by replacing FixedSizeList with Array (#23168)
• Connect Python assert_series_equal() to Rust back-end (#23141)
• Refactor skip_batches to use AExprBuilder (#23147)
• Use ir_serde instead of serde for IRFunctionExpr (#23148)
• Separate FunctionExpr and IRFunctionExpr (#23140)
• Improve Series equality functionality and prepare for Python integration (#23136)
• Add PolarsPhysicalType and use it to dispatch into_series (#23080)
• Remove AExpr::Alias (#23070)
• Add components for Iceberg deletion file support (#23059)
• Feature gate StructFunction::JsonEncode (#23060)
• Propagate iceberg position delete information to IR (#23045)
• Add environment variable to get Parquet decoding metrics (#23052)
• Turn pl.cumulative_eval into its own AExpr (#22994)
• Add make test-streaming (#23044)
• Move scan parameter parsing for parquet to reusable function (#23019)
• Use a ref-counted UniqueId instead of usize for cache_id (#22984)
• Implement Hash and use SpecialEq for RenameAliasFn (#22989)
• Turn list.eval into an AExpr (#22911)
• Only check for unknown DSL fields if minor is higher (#22970)
• Don't enable ir_serde together with serde (#22969)
• Make dtype field on Logical non-optional (#22966)
• Add new (Frozen)Categories and CategoricalMapping (#22956)
• Add a CI check for DSL schema changes (#22898)
• Add schema parameters to expr.meta (#22906)
• Update rust toolchain in nix flake (#22905)
• Update toolchain (#22859)

Thank you to all our contributors for making this release possible!
@Athsus, @DahaoALG, @FabianWolff, @JakubValtar, @Kevin-Patyk, @MarcoGorelli, @SanjitBasker, @alexander-beedie, @borchero, @bschoenmaeckers, @cmdlineluser, @coastalwhite, @deanm0000, @dsprenkels, @eitsupi, @florian-klein, @i1oveMyse1f, @ion-elgreco, @itamarst, @kdn36, @kutal10, @math-hiyoko, @mcrumiller, @mrkn, @mroeschke, @nameexhaustion, @nikaltipar, @orlp, @paskhaver, @ritchie46, @stijnherfst, @thomasfrederikhoeck and @zyctree