Duckdb

This is a bug fix release that we need to create because uploading artefacts failed for the last one. Contents are the same as https://github.com/duckdb/duckdb/releases/tag/v0.3.3 plus some minor bug fixes.

Note: We are going to switch to a new versioning scheme, so the next proper release will be 0.4.0

This preview release of DuckDB is named "Sansaniensis" after the (regrettably extinct) [Chenoanas sansaniensis](https://books.google.nl/books?hl=en&lr=&id=jcEyAQAAMAAJ&oi=fnd&pg=PA1&ots=Yo7uqiwWjB&sig=jeAB2gFbhShF5AWmK1obw8FfYZk&redir_esc=y#v=onepage&q&f=false).

Binary builds are listed below. Feedback is very welcome.

Note: Again, this release introduces a backwards-incompatible change to the on-disk storage format. We suggest you use the `EXPORT DATABASE` command with the old version followed by `IMPORT DATABASE` with the new version to migrate your data. See the [documentation](https://duckdb.org/docs/sql/statements/export) for details.

Below a list of changes in this release

Major Changes & Features

- 2926: DuckDB SQL backend for dbplyr
- 2970: JSON extension
- 2998: Support explicit aggregate state export, re-combination and finalisation
- 3069: S3 writes and 3225: S3 file glob
- 3092: Basic UDF support for Node API
- 3109: Storage dictionary compression
- 3111: Simple filtered aggregates
- 3165: Add support for a `DESCRIBE` statement that gives a summary of all tables
- 3178: 1-based indexing for lists and strings
- 3195: Add support for `FOREIGN KEY`
- 3253: Julia Client & C API Extensions
- 3292: Support NaN and Infinity in floating point columns


Minor Changes & Bug Fixes

- 2954: JDBC enable usage of LocalDateTime via get/setObject
- 2985: Issue 2590: Add overwrite flag to CSV copy, overwrite only if query completes
- 3006: JDBC BigDecimal support
- 3007: Feature: Add support for Excel text format (Update 2996)
- 3010: R: Arrow timestamps in seconds, milliseconds and nanoseconds
- 3027: Allow literal option for regex
- 3030: Macros for floored division and modulo
- 3034: Substrait Installable Extension
- 3038: Issue 3037: DATE + TIME
- 3045: fix for STRING_AGG(...) failing (vector type mismatch) assertion when used with DISTINCT and ORDER BY inside STRING_AGG.
- 3050: R-pkg: cpp11-ify
- 3051: Enable printing of EXPLAIN query trees in the R client
- 3060: Issue 3056: Null list scatter
- 3065: Implementation of list_contains
- 3066: Report dlopen/dlsym errors in extension load
- 3067: ODBC: Many changes
- 3071: c-api: support decimal and prevent invalid type throws
- 3072: Add CMake setting to disable linking of extensions
- 3077: Issue 2498: Consistent NULL hashing
- 3084: Hmb1 select macro feature mr
- 3085: First batch of aggregation functions for python rapi
- 3087: Fix incorrect link for testing docs
- 3088: Add LIST_EXTRACT support for several missing types (including booleans)
- 3090: Fix for Arrow Timezone CI Fail
- 3093: Test/add parser error on missing newline
- 3097: Fix remaining winsock error
- 3098: Round-trippable Expression ToString()
- 3101: Issue 3100: MAKE_DATE struct variant
- 3102: Add flatten function for nested list
- 3107: Allow multiple statements in the R-client
- 3112: Issue 1423: PiecewiseMergeJoin multiple predicates
- 3115: ODBC: SQLGetDescRec and SQLSetDescRec
- 3116: NodeJS Windows CI Fix Attempt
- 3120: Fix 3119: correctly perform case insensitive comparisons in default function/view/schema creation
- 3121: Fix 3108: support ILIKE ESCAPE
- 3123: Issue 3020: Unify column names of DESCRIBE and SUMMARIZE
- 3124: Add Alias to projected columns
- 3125: Adds support for the REPLACE keyword in CREATE TABLE AS ... SELECT (CTAS) statements
- 3126: BLOB support for aggregate state combine
- 3139: More Python Relational API Functions
- 3142: Fix type hints for query_df
- 3147: Python: Relation dependency on Connection
- 3148: Alter Type of column from varchar to enum
- 3151: Support trailing commas in many lists in the parser
- 3154: JDBC - Bugfix - Negative timestamp fix
- 3156: JDBC - Add timestamp with timezone support
- 3158: Enable DISABLE_VPTR_SANITIZER by default on M1
- 3161: Implementation of unnest(null), list_position, rewrites
- 3162: Add support for list parameters in the Python API
- 3163: ODBC: Diagnostic Feature
- 3166: Move test_group_by_parallel out of the coverage tests
- 3171: Fix 3169: verify that WAL exists in PRAGMA database_size
- 3173: Adding chunk size parameter to functions that return arrow objects.
- 3174: Correctly zero-initialize pg_parser allocation heap to prevent potential corruption in clean up of parser state of SQL strings
- 3179: Change pipe so it doesnt print on terminal
- 3180: Fix for build error of R package on Windows
- 3190: Fix Python clean.sh: avoid deleting all locally installed libraries and instead only remove installed DuckDB libraries
- 3191: Returning feature for INSERT
- 3192: Tear down connections
- 3196: linking fixes for extensions
- 3197: Fix convert uuid to arrow str error
- 3199: Dbplyr-backend SQL translation of is.na() must use only IS NULL
- 3201: Enable universal builds for OSX
- 3208: numeric md5 functions
- 3209: JDBC - Fix missing metadata
- 3213: Making the deliminator more flexible and Q 02/17/20 of TPC-H for Substrait
- 3214: Extension linking fix
- 3218: Fixed empty list aggregates
- 3219: Issue 1423: IEJoin Implementation
- 3220: Fix for 3200
- 3221: FIX for parquet reader on diff files
- 3224: feat: expose from_substrait in python package
- 3226: ODBC: Windows setup
- 3230: Fix for 3215
- 3231: CodeCov apparently does not like target anymore, + validate codecov configuration in CI
- 3232: Fix 3223: Prune trees with LIMIT 0
- 3233: Fix 3205: implement skip for struct and list column readers
- 3234: Fix 3222: avoid int overflow in RleBpDecoder
- 3235: Removing std::binary_function because its deprecated in cpp11
- 3238: Parquet globbing: Better support for schema evolution
- 3239: Properly send mapped function on RAPI
- 3244: Returning feature for update and delete statements
- 3245: test: prevent master ci from being canceled
- 3251: Improving CI performance
- 3255: Adding check_same_thread option to py connector
- 3256: ci: restructure to run format/tidy/codecov for python
- 3263: JDBC - Enable useage of ENUM type
- 3264: Allow the use of native replacement scan of parquet/csv files with the dbplyr-backend
- 3266: Fix 3260: if a schema name is provided, pass along the schema name and table name to the replacement scan
- 3268: Add python module constants from PEP249
- 3270: python extension loading tests
- 3274: Implementation of the list aggregate function
- 3276: Properly resizing DuckDB batches that reference arrow batches
- 3281: ODBC: LibreOffice on Linux
- 3287: Several Julia API enhancements
- 3289: ODBC: windows installer has to copy and pass program parameters
- 3293: Issue 3262: Nested comparison fixes
- 3295: Allow strftime arguments to be specified in both orders
- 3299: [Arrow] Read Record Batch Reader and Scanners
- 3301: Use FieldWriter/FieldReader in Checkpoint Manager
- 3303: Add benchmark_runner based regression test, and remove random engine from cycle counter
- 3305: Fix 3304: correctly trigger old result-set materialization prior in duckdb_value_is_null
- 3307: Fix 3271: correctly check for negative limit and offset
- 3308: Cleanup some unused include
- 3311: Fix 3272: avoid incorrectly moving LIMIT to before the ORDER clause in certain cases
- 3312: Issue 3290: Histogram Temporal Bindings
- 3313: Issue 3275: Global Operator State
- 3314: Issue 3262: Dead Code Removal
- 3318: R bug-fixes: Pipes in tests and silence R NOTE
- 3323: Fix 3291: delete root entries after a catalog entry is deleted and no more transactions can reference it
- 3328: Fix 3294: initialize validity mask to correct size in placeholder code
- 3329: Fix use after free bug in local storage
- 3331: Add automatic issue labeler
- 3335: 3324 Throw error instead of crashing when misusing prepared statements
- 3338: Issue 3298: Nested Unflat MinMax
- 3339: Add a create_value for strings to the Julia API
- 3342: Issue 1423: IEJoin Performance Improvements
- 3343: Issue 3118: Non-string Formats
- 3355: Fix 3349: correctly handle empty table case in grouping set
- 3356: Fix 3350: correctly update statistics after a SINGLE join, since SINGLE joins may introduce NULL values
- 3357: Fix 3351: avoid double-binding functions in subqueries after an initial binding error
- 3358: Fix 3352: fix several bugs with index selection for verifying foreign key constraints
- 3359: Fix 3353: check if any results are returned in sqlite api wrapper before looking at the chunk
- 3360: Fix 3354: add missing recursive CTE node traversal in correlated subquery flattening
- 3362: ODBC: unixodbc_setup.sh
- 3363: Fix 3361: correctly handle overflows in sequences with cycles
- 3369: Fix 3367: correctly detect overflow in statistics propagation for ORDER BY
- 3370: Fix 3366: check for invalid double values in progress bar
- 3372: Fix 3365: correctly handle overflows in left-shifts
- 3376: Fix Python Windows CI
- 3378: Correctly handle rowid scans in Julia DataFrame scan (for COUNT(*))
- 3380: ArgMinMax: correctly manage memory of strings in both values and arguments
- 3381: Python extension loading
- 3383: ODBC: unixodbc_setup.sh usage message
- 3399: removed -k param for gzip
- 3406: Install AWS CLI in CentOS Extension

This preview release of DuckDB is named "Gibberifrons" after the [Sunda teal](https://en.wikipedia.org/wiki/Sunda_teal)

Binary builds are listed below. Feedback is very welcome.

Note: Again, this release introduces a backwards-incompatible change to the on-disk storage format. We suggest you use the `EXPORT DATABASE` command with the old version followed by `IMPORT DATABASE` with the new version to migrate your data. See the [documentation](https://duckdb.org/docs/sql/statements/export) for details.

Below a list of changes in this release

Features
- 2797: Add interface support for incremental/deferred query execution in main thread
- 2682: Issue 561: Time Zone Support, 2703: 561: Time Zone Foundations
- 2578: Enable support for multi-dot column references
- 2613: Add support for EXPLAIN ANALYZE
- 2569: Support installation of extensions from central repository
- 2843: Parquet Writer: Support writing all types (test_all_types), and write column statistics, 2838: Add support for reading and writing the MAP type to the Parquet reader/writer, 2832: Parquet Writer Rework: Support complex types, 2597: Extend Parquet writer support for UNSIGNED types, various timestamp types, and correctly set converted types in more cases

Minor Changes & Bug Fixes

- 3042: Issue 3029: Prevent CURRENT_QUERY folding
- 3041: Fix 3017: detect recursive view construction and throw an appropriate error
- 3040: Fix 3015: fix edge case issue with many null values in physical Top N
- 3039: Update Java DuckDBAppender to allow appending null values - fixes issue 3035
- 3028: Install covr from main development branch
- 3026: Issue 2965: CTE parameter binding
- 3024: Issue 3023: Simple filtered aggregates
- 3021: Move install_extension test to test_slow
- 3016: Issue 2923: Quote ENUM types
- 3014: Issue 2908: UTF8 BOM Headers
- 3013: Fix 3005: correctly normalify vector in array_slice
- 3012: Fix 2999: Support ANY_JOIN in flattening correlated subqueries
- 3011: Fix 2986: correctly copy old data over when calling repalloc in parser
- 3004: Fix 2997: allow maximum CSV line size to be configurable
- 3001: Document installation of dependencies in Python package README
- 3000: Lookup parent frames for replacement scans
- 2983: Fix 2975: fix off-by-one in transform of string constant to decimal in case of large negative numbers
- 2982: Python API - allow duckdb.query to run arbitrary queries
- 2980: Fix dead link in readme
- 2979: Fixing Python CI
- 2978: Ensure that users cant run concurrent queries using cursor()
- 2974: Use cpp11 for rpkg
- 2973: ODBC: SQLGetTypeInfo, SQLTables, and SQLColumns
- 2971: Arrow use_async argument deprecation
- 2969: FieldWriter/FieldReader for Backwards/Forwards Compatible Storage
- 2963: Fixing parquet scan with or filters (issue 2955)
- 2957: Fixing wrongful release of arrow arrays
- 2956: fix: use selection vector to ensure ids are ordered
- 2953: Fix cancellation configuration
- 2952: Check coverage for R
- 2951: Adding Cross Product Relation
- 2949: Update copyright info to name DuckDB Foundation
- 2946: Style R code
- 2943: Disable arrow integration tests on Windows
- 2941: Prefix inlined third party libraries with duckdb_
- 2940: CMakeLists.txt: Add 'delayimp.lib' to delay-loaded DLL on MSVC
- 2939: src/include/duckdb/common/assert.hpp: Added winapi
- 2937: Install Python 3.7 from source on Ubuntu 16
- 2935: adding the v to the amalgamation version string
- 2930: list_slice test fix
- 2925: Enable building the HTTPFS extension for both Windows and OSX
- 2924: Add Node 17 to build factory
- 2921: Fix for array_slice when end index is out of bounds
- 2920: ODBC: cursor functions
- 2919: Preserve Identifier Case by default, rather than lower-casing all unquoted identifiers
- 2918: Add left join type to python relational API
- 2917: list_slice alias for array_slice
- 2912: Fixing Arrow Stream Output Error
- 2906: not export deletes to allow crosscompile
- 2905: Use difftime_str instead of hms class
- 2904: Use correct hms value
- 2903: Robust cancellation via concurrency
- 2901: Minor tweaks in R code
- 2899: Support for GROUP BY ALL/ORDER BY ALL
- 2895: Add duckdb_keywords function
- 2894: OSS Fuzz Fixes
- 2890: ODBC driver assets
- 2888: Add JDBC's setTimestamp method
- 2886: Use double to store scale factor in TPC-H answers
- 2885: Better support for Parquet 2 Page Layout
- 2880: fix python profile html generator to support updated JSON format.
- 2871: Issue 1423: Refactor PiecewiseMergeJoin
- 2870: bug fix for order by header
- 2869: Value Rework: Make all members private and use accessors
- 2867: test: fixed typo for decimal type in test_all_types
- 2865: Issue 2782: Non-Gregorian Calendars
- 2864: C API: Deprecate direct access into duckdb_result and duckdb_column structs
- 2863: ODBC: conformance core level of SQLAllocHandle, SQLFreeStmt, SQLGetDiagRec, SQLSetStmtAttr, and descriptors
- 2856: Issue 2842: Implement make_date/time/timestamp
- 2855: right case for winsock2.h
- 2854: Issue 2831: Disable WINDOW FILTER
- 2852: Issue 2827: Postgres datepart compatibility
- 2851: When using Replace, keep the alias of the original column
- 2850: Avoid reading past the end of the string when parsing a time as part of an interval
- 2848: Fix 2847: Write parquet files to test dir
- 2846: Issue 2780: Remove DATETZ type
- 2845: Turn FileSync into a nop in the PipeFileSystem, and add clear error message when ParquetReader is used to read from a FIFO stream (e.g. /dev/stdin)
- 2844: Support read_csv_auto from /dev/stdin
- 2837: Fix potential use after move detected by linter, and add DISABLE_DUCKDB_REMOTE_INSTALL define
- 2835: Issue 2834: ICU BindAdapterData::Equals
- 2833: CI Fixes, TSAN Fix
- 2830: Fix 2828: setup.py BUILD_HTTPFS does not work
- 2825: Fix 2823: Correctly alter cardinality estimation in LIMIT/SAMPLE clauses
- 2822: Issue 2779: DATE_PART structs
- 2819: add `INTERVAL` support to node.js
- 2818: Adding Parquet extension to node module
- 2817: R: Make method definition more similar to S3
- 2808: Revert stack-overflow detection back to having a maximum (expression) depth in the transformer
- 2807: Add discord badge
- 2806: Refactor SQLLogicTest
- 2802: R-Dataset: Add flag to enable async scanner
- 2801: Add support for list equivalent of range/generate_series with timestamp+interval
- 2800: Fix 2749: avoid copying subquery in BETWEEN expressions
- 2799: Fix 2791: allocate data and mask locations to correctly handle counts > vector_size in the row_gather
- 2798: Fix issue in perfect hash join when (max - min) does not fit into an INT64
- 2796: Addition of QUALIFY clause
- 2794: Add Extension Install CI tests, and fix various issues found through this process
- 2793: Enum Functions
- 2792: PhysicalStreamingWindow operator
- 2788: Issue 2778: Missing date parts
- 2784: adds `BOOLEAN` support to node.js bindings
- 2773: Fix 2761: correctly check that stderr is pointing to a terminal
- 2770: Fix a bug in the sqlite3_api_wrapper that resulted in not correctly reporting why a database could not be opened
- 2767: Issue 561: Implement ICU AGE
- 2765: Add Connection::GetTableNames method to C++ API that allows you to extract the required table names from a query
- 2760: Issue 561: Implement ICU DATEDIFF
- 2759: Fix string constructor usage
- 2757: Extension dashes to underscores
- 2756: Issue 561: Implement ICU DATESUB
- 2753: Python - Test All Types
- 2751: Fix 2750: check enable_external_access flag in more locations
- 2748: Cleanup pointer swizzling code
- 2746: Fix 2745: correctly detect when a RECURSIVE CTE does not contain a reference to itself (i.e. is not a recursive CTE at all)
- 2744: C++17 warning fixes, plus CI tests for compiling (parquet) amalgamation with C++17
- 2742: Common: typo and remove unused files
- 2739: Add test_all_types function
- 2738: GH Actions: avoid upload on master, and try to fix codecov failures on master
- 2737: Typo: Rename Alises to Aliases
- 2732: Tests for Python + 1732
- 2731: Pushing down OR filters
- 2730: Issue 561: ICU Date addition
- 2729: Add duckdb_functions table function
- 2727: RAII for SortedBlock
- 2725: Windows: Remove OVERLAPPED IO flag, it is not required and seems to cause concurrency problems
- 2723: Fix 2471: correctly handle offset passed by ::UpdateSegment, and handle it earlier to clean up code
- 2722: Infer COPY TO format from file extension
- 2721: Update of PERCENT keyword in the LIMIT clause(2671)
- 2719: Fix 2713: correctly bind multi-part column references in correlated subqueries
- 2718: Add "position" as an alias to the "instr" function
- 2712: Use and_kleene for Arrow filters
- 2711: Fix CSE optimizer: keep around expressions as they might be referenced in the expression_map
- 2710: Modify LEAST/GREATEST to ignore NULL values
- 2708: Fix 2701: Handle VALUES lists in correlated subqueries
- 2706: Ignore cancel request on master
- 2702: Rename snappy namespace to duckdb_snappy, and enable Parquet extension for the exported symbol checker
- 2697: Support reading of ZSTD files, and add support for writing GZIP and ZSTD files
- 2696: FetchDF for nested types (Lists, Maps and Structs)
- 2693: Fix 2663: correctly implement EXTRACT(EPOCH from TIME) as seconds since midnight
- 2692: Fix 2678: Fix undefined behavior for sequences close to INT64 min/INT64 max
- 2691: Correctly handle synchronous I/O being returned from Windows API
- 2690: Use templated memcpy/memcmp more
- 2689: Reduce memory footprint of Python/R compilation by default by disabling unity builds unless DUCKDB_BUILD_UNITY flag is enabled
- 2686: Fix 2685: prevent CSE optimizer from causing short-circuiting issues, and add disable_optimizers setting
- 2684: Python 3.10 builds
- 2681: Benchmark/test format cleanup
- 2679: Bitpacking storage compression
- 2673: remove extra semicolon.
- 2665: Merge arrow_register into register function
- 2662: Fix 2656: correctly transform DISTINCT with ORDER BY into DISTINCT ON
- 2660: Issue 2614: Wide windows
- 2659: Fix 2652: Make struct fields case insensitive
- 2658: Fix various Windows unicode issues
- 2657: Initializing matchers to avoid valgrind complaints
- 2655: Add templated memcpy/memcmp
- 2650: Fix bug in parquet reader causing list columns to be parsed incorrectly (2557)
- 2648: Fix 686: remove hard-coded memory limit in parser and fix error message propagation from exceptions thrown in parser
- 2644: Fix 2641: correctly handle tab delimiters in COPY TO/FROM
- 2642: Issue 2552: SUBSTR BIGINT
- 2639: Fix 2586: correct semantics for extract(second, ...) from intervals
- 2636: Hashing enum values (Enum comparison bug fix)
- 2635: Fix for R factor scans
- 2631: ODBC: Refactoring Parameter Descriptor
- 2630: Try to Cast Enums to other types
- 2626: Escape column names in dbWriteTable() Fixes 2622
- 2623: Fix 2612: Correctly check that default is set on copy of SetDefaultInfo
- 2611: Multithreaded Python
- 2609: Configuration Rework & Cleanup
- 2602: Adding pragma option for parquet binary as string
- 2583: Issue 2549: Support IGNORE NULLS
- 2576: Add generate_subscripts macro
- 2526: Sequence ownership

This preview release of DuckDB is named "Spectabilis" after the [King Eider](https://en.wikipedia.org/wiki/King_eider)

Binary builds are listed below. Feedback is very welcome.

Note: Again, this release introduces a backwards-incompatible change to the on-disk storage format. We suggest you use the EXPORT DATABASE command with the old version followed by IMPORT DATABASE with the new version to migrate your data. See the [documentation](https://duckdb.org/docs/sql/statements/export) for details.

Below a list of changes in this release

**Features**
- 2393: Switch to Push-Based Execution Model
- 2417: Add support for GROUPING SETS, ROLLUP, CUBE and GROUPING/GROUPING_ID
- 2347: Support for ENUM Types & 2404: Native mapping between R factors and DuckDB ENUMs
- 2419: Allow WHERE clause referring aliases defined in SELECT clause
- 2473: Adding Compression Option for Column Definitions to SQL Parser
- 2489: Implement MAD (Moving Absolute Deviation) Aggregate
- 2520: Add LIST_CONCAT and LIST_APPEND functions
- 2522: MSD (Most Significant Digit) Radix Sort
- 2529: Implement REGEXP_EXTRACT
- 2482: Add support for EVEN
- 2555: Adding BINARY_AS_STRING parameter to the parquet scan

**Minor Changes & Bug Fixes**
- 2377: Fix current_schema() and current_schemas()
- 2380: Add tests for "did you mean" error message
- 2381: Let FTS index creation respect the current schema.
- 2382: Dropping support to Python 2
- 2385: BLOB support for JDBC
- 2389: Replace __getattr__ on PyBind11 classes with individual properties.
- 2390: Moving big categorical tests to tests_slow folder
- 2394: Add SET s3_endpoint support for in-house Ceph
- 2397: Add Python .pyi stubs
- 2402: GH Actions Upload: retry asset upload with timeout
- 2403: Not altering original DF when renaming columns in the binder
- 2405: Fix bug with enum::varchar cast on null values
- 2408: Rewrite Arrow table register for R using replacement scans
- 2409: Adding dependency on Enum Types -> Tables
- 2412: Updated src readme to state we use push-based execution.
- 2413: Fix for 2411
- 2421: Fix 2407: use correct template parameters for DATE in arg_min/arg_max
- 2423: Fix 2416: fix binding issues related to binding parameters, null values, etc in list_extract and array_length
- 2424: Issue 2388: QUANTILE_DISC for VARCHAR
- 2430: Fix for 2426
- 2431: More fixes for 2416
- 2434: Issue 2388: Moving VARCHAR QUANTILE_DISC
- 2437: implement GEN_RANDOM_UUID
- 2438: Issue 2432: PERCENTILE_XXXX ignores DESC
- 2439: fix: pass absolute path to `System.load()` in Java
- 2444: Fix 2440: correctly report run-time errors in Python client
- 2445: Several OSS Fuzz Fixes
- 2448: Move to codecov v2, and use add_library for vector operations for low RAM machines
- 2449: Enum to Enum Comparisons
- 2451: Hooold the loooock for Python Strings under Dataframe Object Columns
- 2453: Fix when getting single values from ENUMs
- 2455: Doc Improve: Trying to Update doc in `QueryResult`
- 2456: Add --test-dir parameter to unittest so we can test out-of-tree extensions with it
- 2462: Fix 2452: Implement Coalesce instead of rewriting to CASE chain
- 2463: More OSS Fuzz fixes
- 2474: Allow Fetch of chunks containing a multiple of vector sizes for the Arrow Record Batch Reader
- 2476: Really holding the lock this time
- 2477: Sorted aggregate: only re-order when ordering count > 0
- 2485: Benchmarks: Handle comparisons for values that do not have a VARCHAR ->TYPE cast (e.g. complex/list types)
- 2487: Conditionally define UNLIKELY in Thrift
- 2488: ODBC: fetching the first chunk in SQLExecute
- 2490: Clean up RadixSort code
- 2491: These tests are now passing with arrow 6
- 2494: Emit full vectors from VALUES lists instead of emitting individual tuples
- 2497: Upgrade Catch to v2.13.7
- 2500: Fix nested string order
- 2501: Add CIFuzz action
- 2503: Fix for py string conversion on large strings
- 2504: More precision for `SUM` and `AVG`
- 2517: Move Kahan sum to separate method (fsum, sum_kahan)
- 2521: Issue 2515: Windowed quantile list
- 2527: testing: Add oss-fuzz fuzzer
- 2536: Fix 2518: in read_csv_auto don't override names if names have been provided
- 2537: Fix 2531: in recursive CTE avoid waiting for events to finish if an event has thrown an error
- 2539: Restructuring CI Workflow
- 2542: Issue 2530: Reset windowed lists
- 2550: ODBC: Running PSQLODBC tests on Win64
- 2556: Fixed directory separator bug
- 2558: Fixing positional reference binding in ORDER BY clause
- 2559: Issue 2552: General ordered aggregates
- 2561: Fix master CI: Python workflow needs auth tokens for deployment
- 2563: Move ExtensionHelper into main DuckDB Class
- 2564: R Client: Moving UTF encoding to R to avoid multithreading issues
- 2567: Fix 2538: crash in CSV auto-detect when reading ZSTD data
- 2573: Move HTTPFS builds to separate test to avoid deploying them by default on Linux
- 2574: Naming optimizer-created aggregates so plans are interpretable
- 2579: Support to Arrow 6
- 2580: Check magic bytes before checksum when opening a DuckDB database file
- 2585: Fix 2584: correctly handle edge cases for bigger than 1 increments in range table function
- 2587: Replacement Scans for Arrow Objects
- 2592: Fix 2577: Rework case statement to avoid rewrite into nested binary cases
- 2593: Fix 2591: avoid using ungrouped aggregate for non-combineable aggregates
- 2594: Fix 2588: timestamp -> date cast is not invertible
- 2595: Fix 2543: case insensitive replacement scans
- 2598: OSS Fuzz Fixes
- 2603: The asset upload script for releases was broken somehow
- 2605: Clean up reupload by splitting it into two functions
- 2604: Fix 2599: maintain correct dependencies between UNION ALL nodes so that output is deterministic/in-line with what a sequential execution would produce

This preview release of DuckDB is named "Gracilis" after the [Grey Teal](https://en.wikipedia.org/wiki/Grey_teal)

Binary builds are listed below. Feedback is very welcome.

Note: Again, this release introduces a backwards-incompatible change to the on-disk storage format. We suggest you use the EXPORT DATABASE command with the old version followed by IMPORT DATABASE with the new version to migrate your data. See the [documentation](https://duckdb.org/docs/sql/statements/export) for details.

This release contains a novel join method, the 1959: Perfect Hash Join. Otherwise there are many improvements and bug fixes however, a list is below:

- 2377: Fix current_schema() and current_schemas()
- 2371: Installing pandoc so R pkg can be fully checked
- 2369: More CI fixes
- 2361: ODBC: First version Winsetup
- 2360: Issue 2348: Support bankers rounding as default function
- 2359: Implement UUID data type
- 2358: Fixes minor flag comment
- 2357: Issue 2351: Ordered aggregate transformation
- 2356: Issue 2286: Numeric SUMMARIZE stats
- 2353: Get rid of a couple of Clang warnings.
- 2346: Avoid re-using RE2 regex between threads, since RE2 regex objects have locks internally
- 2328: CI Fixes
- 2321: Issue 1998: Exact Timestamp Subtraction
- 2320: Change SET's default scope from GLOBAL to (PG-compatible) SESSION.
- 2319: Fix 2315: correctly detect that index joins cannot be used for multi-column indexes, and clean up TPC-DS extension
- 2318: Remove extra semicolon.
- 2317: Support SET SCHEMA and SET SEARCH_PATH
- 2316: Fix for 2304
- 2314: Fix 2313: No Out of Range Error in Index Benchmark
- 2311: Issue 2310: Create Index Benchmark
- 2306: Windows (64 bits) and -DDISABLE_UNITY=1
- 2302: Fix 2301: ART Leaf Node shrink
- 2300: Fix 2293: Correctly escape all special characters (quotes, newlines, tabs, etc) in JSON output of query profiler
- 2299: Fix 2296: Avoid requesting O_RDWR permissions when we only need O_WRONLY so we can write to FIFO streams
- 2298: Fix 2294: In CSV reader correctly generate column names with many columns
- 2297: Issue 2241: Transacted Index Reinsert
- 2290: Fix 2289: align default continue prompt
- 2287: ODBC: list catalog, schema and test_blob
- 2284: Make S3 credentials session scoped
- 2282: Fix for bug 2281
- 2280: Initial support to read lists from R client
- 2279: Fix 2277: add support for lists of structs to LIST_EXTRACT
- 2276: Add support for EXCLUDE and REPLACE clauses
- 2275: Run only CRAN tests for valgrind
- 2274: Fix 2267: For structs, get the required amount of rows from a non-list child if there is one
- 2272: Rename force_parallelism to verify_parallelism
- 2271: Make regression run parallel with 2 Threads
- 2266: A refactoring around FileSystem
- 2265: Allow optional extensions when building the R package
- 2263: Regression Test: TPC-DS/H20AI and other adjustments
- 2262: Fix 2261: add support for filters pushed down into decimal columns in Parquet files
- 2259: Fix a TSAN error for test/sql/window/test_partition_flushing.test
- 2255: Make sorting even faster
- 2254: More descriptive out-of-memory error when db is launched in-memory
- 2253: Regression Tests
- 2251: Changes from CRAN 0.2.9 release
- 2234: Add support for S3 session token (STS)
- 2228: Date functions on Windows

This preview release of DuckDB is named "Platyrhynchos" after the very well-known [Mallard](https://en.wikipedia.org/wiki/Mallard), πλατυρυγχος meaning "broad-billed".

Binary builds are listed below. Feedback is very welcome.

Note: Again, this release introduces a backwards-incompatible change to the on-disk storage format. We suggest you use the EXPORT DATABASE command with the old version followed by IMPORT DATABASE with the new version to migrate your data. See the [documentation](https://duckdb.org/docs/sql/statements/export) for details.

Major changes (not listing minor bug fixes and small changes):

Storage
- 2099: **Persistent Storage Compression Framework** & RLE compression for numeric columns
- 2157: Multi-Occupancy Blocks in the Storage


SQL
- 2247: Support session and global `SET` statement variable scopes.
- 2230: Full Case Insensitivity in the Binder & Catalog
- 2208: Allow duplicate column names in subqueries/`CREATE TABLE AS`, similar to SQLite
- 2206: Allow unnamed subqueries
- 2189: Add support for `SUMMARIZE` of SQL statement/table
- 2158: `ORDER BY` aggregates
- 2171: Implement `LAST`
- 2146: Support for `RANGE` and `ARRAY_LENGTH` scalar functions
- 2133: Implement `DATESUB` & 2090: Implement `DATEDIFF`
- 2128: Fall back to VARCHAR when the type of a prepared statement parameter is ambiguous

Performance Improvements
- 2226: Issue 1657: Reduce Window copying
- 2221: Prefetch Parquet Meta Data
- 2172: Top-N Rework using new sort code
- 2167: Use new sort code in window functions
- 2098: Reduce unnecessary sorting overhead
- 2077: Fetch Arrow - Streaming

C Client:
- 2173: Don't strdup error msg
- 2115: Improve performance of duckdb_value functions, add support for HUGEINT type and add extra result helper functions
- 2107: CAPI Cleanup & DATE/TIME/TIMESTAMP rework

R Client
- 2250: Support column subsets in dbAppendTable()
- 2134: Implement dbAppendTable()
- 2136: Implement dbBind() according to specification
- 2119: Enable more DBItest tests

ODBC Client
- 2184: Enabling pyodbc & 2124: Enable nanodbc & 2159: Making the R odbc package work with our ODBC driver
- 2169: Supporting SQLPutData, SQLParamData, SQLMoreResults & - 2096: SQLFetchScroll and SQLGetPrivateProfileString
- 2236: Windows x64 Support