Rust Polars 0.41.0

💥 Breaking changes

• Make hive_partitioning parameter default to None, which is automatically enabled for single directory inputs, and disabled otherwise (#17106)
• Split replace functionality into two separate functions (#16921)
• Rename DataFrame.melt to unpivot and make parameters consistent with pivot (#17095)
• Default to writing binview data to IPC (#17084)
• Do not parse hive partitions from user provided directory/glob path (#17055)
• Add strict parameter to DataFrame/LazyFrame.drop and fix behavior to default to True (#17044)
• Remove supertype definition of List and non-List types (#16918)
• Native selector XOR set operation, guarantee consistent selector column-order (#16833)
• move offset_by implementation from polars-plan to polars-time, rename feature from DateOffset to OffsetBy (#16796)
• Rename str.concat to str.join and update default delimiter (#16790)
• Remove deprecated parameters in Series.cut/qcut and update struct field names (#16741)
• Expedited removal of certain deprecated functionality (#16754)
• Update some error types to more appropriate variants (#15030)
• Change default offset in group_by_dynamic from 'negative every' to 'zero' (#16658)
• Update clip to no longer propagate nulls in the given bounds (#14413)
• Change str.to_datetime to default to microsecond precision for format specifiers "%f" and "%.f" (#13597)
• Update resulting column names in pivot when pivoting by multiple values (#16439)
• Preserve nulls in ewm_mean, ewm_std, and ewm_var (#15503)
• Restrict casting for temporal data types (#14142)
• Rename struct fields of rle output to len/value and update data type of len field (#15249)
• Add check_names parameter to Series.equals and default to False (#16610)
• Deprecate str.explode in favor of str.split("").explode() (#16508)
• Deprecate how="outer" join type in favour of how="full" (left/right are *also* outer joins) (#16417)
• Change DataFrame.is_empty() to check height == 0 instead of width == 0 (#16351)

🚀 Performance improvements

• Default to writing binview data to IPC (#17084)
• Parallelize arrow conversion if binview -> large_bin (#17083)
• GC buffers in if_then_else view kernel (#16993)
• Desugar AND filter into multiple nodes (#16992)
• Optimize generic argsort of row-encoding (#16894)
• Improve rle_id iteration perf and set sorted flags (#16893)
• Optimize string/binary sort (#16871)
• Use split_at in split (#16865)
• Use split_at instead of double slice in chunk splits. (#16856)
• Don't rechunk in align_ if arrays are aligned (#16850)
• Don't create small chunks in parallel collect. (#16845)
• Add dedicated no-null branch in arg_sort (#16808)
• Speed up dt.offset_by 2x for constant durations (#16728)
• Toggle coalesce in join if non-coalesced key isn't projected (#16677)
• Make dt.truncate 1.5x faster when every is just a single duration (and not an expression) (#16666)
• Always prune unused columns in semi/anti join (#16665)
• make truncate 4x faster in simple cases (#16615)
• Cache arena's (and conversion) in SQL context (#16566)
• Partial schema cache. (#16549)
• improved numeric fill_(forward/backward) (#16475)
• only rechunk once per aggregate (#16469)
• Fix pathological small chunk parquet writing (#16433)

✨ Enhancements

• Make hive_partitioning parameter default to None, which is automatically enabled for single directory inputs, and disabled otherwise (#17106)
• Split replace functionality into two separate functions (#16921)
• Improve schema inference for hive partitions (#17079)
• Rename DataFrame.melt to unpivot and make parameters consistent with pivot (#17095)
• print row index in explain + dot (#17074)
• Support top-level pl.col autocompletion for iPython (#17080)
• predicate + projection pushdown in NDJson (#17068)
• Allow (non-)coalescing in join_asof (#17066)
• Turn of coalescing and fix mutation of join on expressions (#17061)
• Expand NDJson glob into one SCAN (#17063)
• Do not parse hive partitions from user provided directory/glob path (#17055)
• Support directory paths in scans for Parquet, IPC and CSV (#17017)
• Implement general array equality checks (#17043)
• Add strict parameter to DataFrame/LazyFrame.drop and fix behavior to default to True (#17044)
• allow experimental metadata use on release (#17005)
• first working prototype of new streaming engine (#16970)
• Desugar AND filter into multiple nodes (#16992)
• use min/max metadata on debug builds with POLARS_METADATA_FLAGS=extensive (#16963)
• Add SQL support for INTERSECT and EXCEPT ops (#16960)
• Allow setting file cache TTL on a per-file basis (#16891)
• Implement multiply and division for lhs duration (#16948)
• Raise on invalid temporal arithmetic (#16934)
• Always end with a in-memory sink on collect (#16928)
• Normalize value_counts (#16917)
• add eq/ne for more FixedSizeLists (#16902)
• setup skeleton (#16900)
• add fundamentals for new async-based streaming execution engine (#16884)
• Cache downloaded cloud IPC files (#16892)
• Improve read_csv SQL table reading function defaults (better handle dates) (#16866)
• Support SQL VALUES clause and inline renaming of columns in CTE & derived table definitions (#16851)
• convert to give time zone in .str.to_datetime when values are offset-aware (#16742)
• Support SQL "SELECT" with no tables, optimise registration of globals (#16836)
• Native selector XOR set operation, guarantee consistent selector column-order (#16833)
• Extend recognised EXTRACT and DATE_PART SQL part abbreviations (#16767)
• Improve error message when raising integers to negative integers, improve docs (#16827)
• Return datetime for mean/median of Date colum (#16795)
• Expose overflowing cast (#16805)
• Expose a few more expression nodes in the expression IR (#16781)
• Support array arithmetic for equally sized shapes (#16791)
• Support cloud storage in scan_csv (#16674)
• Streamline SQL INTERVAL handling and improve related error messages, update sqlparser-rs lib (#16744)
• Support use of ordinal values in SQL ORDER BY clause (#16745)
• Support executing polars SQL against pandas and pyarrow objects (#16746)
• add env locked metadata functions (#16719)
• Remove deprecated parameters in Series.cut/qcut and update struct field names (#16741)
• Expedited removal of certain deprecated functionality (#16754)
• Update date_range to no longer produce datetime ranges (#16734)
• Remove deprecated top_k parameters nulls_last, maintain_order, and multithreaded (#16599)
• Support order-by in window functions (#16743)
• Add SQL support for NULLS FIRST/LAST ordering (#16711)
• Update some error types to more appropriate variants (#15030)
• Initial SQL support for INTERVAL strings (#16732)
• Enforce deprecation of offset arg in truncate and round (#16655)
• eliminate ProjectionExprs and handle CSE by stacking extra columns (#16682)
• Change default offset in group_by_dynamic from 'negative every' to 'zero' (#16658)
• Update clip to no longer propagate nulls in the given bounds (#14413)
• Change str.to_datetime to default to microsecond precision for format specifiers "%f" and "%.f" (#13597)
• Update resulting column names in pivot when pivoting by multiple values (#16439)
• Preserve nulls in ewm_mean, ewm_std, and ewm_var (#15503)
• Restrict casting for temporal data types (#14142)
• Add many more auto-inferable datetime formats for str.to_datetime (#16634)
• Rename struct fields of rle output to len/value and update data type of len field (#15249)
• Add check_names parameter to Series.equals and default to False (#16610)
• Dedicated SQLInterface and SQLSyntax errors (#16635)
• Add DIV function support to the SQL interface (#16678)
• add additional control to write_parquet::statistics parameter (#16575)
• Support non-coalescing streaming left join (#16672)
• Allow wildcard and exclude before struct expansions (#16671)
• Support per-column nulls_last on sort operations (#16639)
• Add split_at method to arrow Array (#16620)
• Initial support for SQL ARRAY literals and the UNNEST table function (#16330)
• Don't allow struct.with_fields in grouping (#16629)
• Add SQL support for TRY_CAST function (#16589)
• add fuzzer for expressions (#16581)
• handle CSE dtypes in NodeTraverser.get_dtype (#16552)
• check if by column is sorted, rather than just checking sorted flag, in group_by_dynamic, upsample, and rolling (#16494)
• Add general metadata structure to ChunkedArray (#16399)
• Add is_column_selection() to expression meta, enhance expand_selector (#16479)
• NDarray/Tensor support (#16466)
• Allow designation of a custom name for the value_counts "count" column (#16434)
• Default rechunk=False for read_parquet (#16427)
• Add field expression as selector with an struct scope (#16402)
• Field expansion renaming (#16397)
• add cluster_with_columns plan optimization (#16274)
• Change DataFrame.is_empty() to check height == 0 instead of width == 0 (#16351)
• add Expr.interpolate_by (#16313)

🐞 Bug fixes

• Expand i128 primitive type match (#17076)
• Fix decompress_impl for csv with n_rows set (#17118)
• adds "polars-ops/timezones" dependency for "timezones" feature (#17115)
• Fix incorrect window std for chunked series (#17110)
• make GetOutput::get_field fallible (#17114)
• bubble error when no available bitrepr (#17116)
• Fix melt panic (#17088)
• Exclude index from expansion in rolling/group_by_dynamic (#17086)
• fix #17043 binary compare (#17052)
• Fix oob of join with literals and empty table (#17047)
• Don't silently accept multi-table FROM clauses (implicit JOIN syntax) (#17028)
• fix get categories on multiple row groups (#17041)
• Don't split up ANDed filters that are group-aware (#17031)
• Harden "async" check for users with out-of-date sqlalchemy libraries (#17029)
• error when sort_by of unequal length (#17026)
• properly catch not found explode cols (#17020)
• Correctly convert data frames to NumPy for C index order (#17000)
• Raise on invalid arithmetic shapes (#16986)
• Don't pushdown predicates in cross join if the refer to both tables (#16983)
• Fix projection pushdown with literal joins (#16981)
• Handle strictness for Decimal Series construction (#15309)
• properly set FAST_EXPLODE_LIST metadata (#16951)
• Raise informative error when writing object to file (#16954)
• Remove supertype definition of List and non-List types (#16918)
• Reject non-integral offset and length in AExpr::Slice (#16874)
• properly read/write fixed-sized lists from/to parquet files (#16747)
• Remove unwrap in extend() (#16890)
• Fix should_rechunk check (#16852)
• Standardised additional SQL interface errors (#16829)
• Ensure that splitted ChunkedArray also flattens chunks (#16837)
• Reduce needless panics in comparisons (#16831)
• Reset if next caller clones inner series (#16812)
• Raise on non-positive json schema inference (#16770)
• describe / explain streaming plan (#16771)
• Rewrite implementation of top_k/bottom_k and fix a variety of bugs (#16804)
• Fix comparison of UInt64 with zero (#16799)
• properly set boolean distinct count (#16782)
• Fix incorrect parquet statistics written for UInt64 values > Int64::MAX (#16766)
• Fix boolean distinct (#16765)
• DATE_PART SQL syntax/parsing, improve some error messages (#16761)
• Column selection wasn't applied when reading CSV with no rows (#16739)
• Only flush if operator can flush in streaming outer join (#16723)
• Raise unsupported cat array (#16717)
• Restrict casting for temporal data types (#14142)
• Improve read_database check for SQLAlchemy async Session objects (#16680)
• Full null on dyn int (#16679)
• Fix filter shape on empty null (#16670)
• Potentially deal with empty range (#16650)
• Use of SQL ORDER BY should not cause reordering of SELECT cols (#16579)
• ensure df in empty parquet (#16621)
• get_dtype handles input node schema and CSE (#16582)
• Crash using empty Series in LazyFrame.select() (#16592)
• small safety issue in CWC filtermap (#16591)
• bail on bool floordiv (#16578)
• Resolve multiple SQL JOIN issues (#16507)
• solve panic in cluster_with_columns, found with small fuzzer (#16562)
• Project last column if count query (#16569)
• Properly split struct columns (#16563)
• Ensure strict chunking in chunked partitioned group by (#16561)
• Error selecting columns after non-coalesced join (multiple join keys) (#16559)
• Don't panic on hashing nested list types (#16555)
• deal with realiases in cluster_with_columns (#16548)
• Ensure deduced join key names are unique (#16551)
• Crash selecting columns after non-coalesced join (#16541)
• Fix group gather of single literal (#16539)
• throw an invalid operation exception on performing a sum over a list of strs (#16521)
• properly fill null into unknown struct fields (#16518)
• Fix df.chunked for struct (#16504)
• Mix of column and field expansion (#16502)
• Fix split_chunks for nested dtypes (#16493)
• Handle struct.fields as special case of alias (#16484)
• Correct schema for list.sum (#16483)
• properly display other length in simple project and display more columns up to a maximum length (#16464)
• allow search_sorted directly on multiple chunks, and fix behavior around nulls (#16447)
• Fix use of COUNT(*) in SQL GROUP BY operations (#16465)
• properly set schemas in cluster_with_columns when reassigning to columns (#16463)
• Fix panic when computing min() of Duration series. (#16455)
• keep track of removed indices in CWC (#16443)
• Fix struct 'with_fields' schema for update dtypes (#16428)
• Fix error reading lists of CSV files that contain comments (#16426)
• Fix struct arithmetic schema (#16396)
• Fix don't panic on chunked to_numpy conversion (#16393)
• Don't check nulls before conversion (#16392)
• fix a printing issue in IR::Filter (#16378)
• correct AExpr.to_field for bitwise and logical and/or (#16360)

📖 Documentation

• Fix some warnings during doc build (#17077)
• Lots of additions to the SQL reference docs (#16990)
• Update the Overview section of the contributing guide (#15674)
• Update outdated performance section (#16409)

📦 Build system

• Bump Rust toolchain version to nightly-2024-06-03 (#16669)
• bump brotli from 5.0.0 to 6.0.0 (#16643)

🛠️ Other improvements

• Split replace functionality into two separate functions (#16921)
• Rename DataFrame.melt to unpivot and make parameters consistent with pivot (#17095)
• Add pivot test #17081 (#17090)
• Make Series and ChunkedArray ops fallible (#16965)
• Support directory paths in scans for Parquet, IPC and CSV (#17017)
• expose all metadata to polars-expr (#17018)
• Move in-memory engine to its own crate (#17039)
• Remove file cache test (#17038)
• Point polars-stream to crates/ again (#17024)
• Fix failing file cache test in CI (#17014)
• Use proper join type in test (#16994)
• Fix file cache verbose logging leakage during pytest (#16984)
• Remove redundant projection attribute in IR::DataFrameScan (#16952)
• Follow-up changes to polars-parquet (#16949)
• Factor out some apply calls in duration namespace (#16941)
• extend new streaming engine with some initial nodes (#16940)
• setup skeleton (#16900)
• Refactor parts of IR. (#16899)
• Remove inner Arc from FileCacheEntry (#16870)
• Update links to API references (#16843)
• Prepare update of API reference URLs (#16816)
• Rename allow_overflow to wrap_numerical (#16807)
• Remove unneeded code (#16838)
• Don't enter streaming engine for groupby-> agg mean/median … (#16810)
• move offset_by implementation from polars-plan to polars-time, rename feature from DateOffset to OffsetBy (#16796)
• Improve safety of amortized_iter (#16820)
• start further use of polars-compute in polars-parquet (#16788)
• Remove deprecated MutableBitmap.null_count method (#16797)
• Rename str.concat to str.join and update default delimiter (#16790)
• add distinct_non_null_count method (#16789)
• Remove needless inner type clone (#16718)
• Simplify NodeTraverser.get_dtype (#16712)
• Fix incorrect debug assertion in ChunkedArray::from_chunks_and_dtype (#16697)
• separate Aggregation evaluation PhysicalExpr from conversion (#16688)
• Update version resolver for 1.0.0 release (#16705)
• Branch earlier in binary type resolving (#16685)
• Avoid AWS pinning to outdated crc32c version (#16681)
• use binary try_{add, mul, ...} ops in borrowed dispatch (#16580)
• Split dates_times into separate date and time modules (#16667)
• Add test for 16642 (#16646)
• Remove duplicate tag in CODEOWNERS (#16625)
• CWC prealloc pushable and potential_pushable (#16626)
• Update dprint hook versions and enable JSON linting (#16611)
• defensively invalidate metadata and start on copying of min_value, max_value and distinct_count (#16593)
• Deprecate str.explode in favor of str.split("").explode() (#16508)
• make Parquet Statistics into enum instead of trait (#16485)
• small change on CWC where we can reuse the old pushable_set_bits and reserve space for input_exprs (#16468)
• Deprecate how="outer" join type in favour of how="full" (left/right are *also* outer joins) (#16417)
• Include license file in polars-expr crate (#16421)

Thank you to all our contributors for making this release possible!
@BGR360, @JulianCologne, @KDruzhkin, @Kylea650, @MarcoGorelli, @Mottl, @Object905, @alexander-beedie, @ankane, @bertiewooster, @borchero, @c-peters, @cmdlineluser, @coastalwhite, @dangotbanned, @datenzauberai, @dependabot, @dependabot[bot], @hattajr, @henryharbeck, @itamarst, @machow, @marenwestermann, @mcrumiller, @mdavis-xyz, @messense, @montanarograziano, @nameexhaustion, @orlp, @p3i0t, @r-brink, @ritchie46, @siddharth-gulia, @stinodego, @tkellogg, @twoertwein, @universalmind303 and @wence-