ParquetSharp 2.3.0 Beta 1

• Upgrade to Arrow 1.0.1.
• Move back to Snappy as the default compression (upstream Arrow has temporarily disabled LZ4 due to format inconsistencies across libraries).
• Add long filename support on Windows (upstream fix, Issue #114)

ParquetSharp 2.2.0

• Promoted 2.2.0 Beta 4 to stable release.

ParquetSharp 2.2.0 Beta 4

• Upgrade to Arrow 0.17.1.
• Add support for UUID logical type via System.Guid (Issue #139).

ParquetSharp 2.2.0 Beta 3

• Fix to ensure the native DLL dependency is correctly passed to all using projects. Some combination of .NET Standard and .NET Framework projects have been reported not to always work (see PR #134).

ParquetSharp 2.2.0 Beta 2

• Fix reading incorrect string values due to recently introduced string reader cache.
• Add support for macOS, now part of the nuget package.
• Reduce the Linux binary size by adding a stripping build step.
• Completely refactor the continuous integration workflow on GitHub Actions.

ParquetSharp 2.2.0 Beta 1

Moving straight to 2.2.0 Beta without promoting 2.1.0 Beta due to a critical bug in Arrow 0.16.0 (Issue #116).

• Upgraded to Arrow 0.17.0.
• Fix the data corruption on buffered row group with plain encoding (upstream Issue #116).
• Add support for ByteStreamSplit (Issue #112).
• Add string cache when reading with dictionary encoding. This reduces memory usage and improves performance when reading string columns (Isse #118).

ParquetSharp 2.1.0 Beta 2

• Fix RowOriented.PaquetRowWriter generating an invalid Parquet file if not writing any rows (Issue #110).

ParquetSharp 2.1.0 Beta 1

• Upgraded to Arrow 0.16.
• Added bindings for the new AES encryption API. You can chose to encrypt all or some of your Parquet columns (Issue #82).
• Added explicit field-column mapping attribute for the row-oriented API (Issue #90).
• Match the C++ dispose / destructor implementation wrt exception safety. It looks like it's a good habit to explicitly call PaquetFileWriter.Close() instead of relying on the C++ destructor (or the C# Dispose()) as they will gobble and ignore any thrown exception (Issue #104).

ParquetSharp 2.0.2

• Promoted 2.0.2-beta5 to final release.

ParquetSharp 2.0.2 Beta 5

Performance improvements on LogicalColumnReader:

• Use DateTime ctor instead of FromBinary, as the later has more checks and overhead (fixes Issue #86).
• Do not use the conversion buffer if data can be read straight into the user destination buffer.