What's Changed

• Smaller improvements of Scheduler
• active Execution Context is now set by scheduler
• sorting is parallel again
• Replace existing "catalog" logic with proper one
• Fix groupjoins and reenable them
• Restructure and improve github actions for pushes/pull requests
• Bump apache arrow to Apache 20
• MacOs support: build + python package
• Optimize loading/storing of members during sub-operator lowering
• UTF-8 aware string functions
• Remove EnforceCABIPass (now unnecessary)
• Add devcontainer
• Proper handling of char type
• Avoid generation of redundant operations and add folding rules
• Optimize hash lookups (including bloom filters from this paper
• Refactor dsa dialect into proper arrow dialect
• Generate documentation for MLIR dialects
• Fix bug under Linux ARM calling convention
• Improve LLVM Debugging Backend
• Fix memory leaks and improve ASAN build (by avoiding boost fibers for ASAN)
• Add string functions necessary for clickbench queries
• Fix problem with distinct aggregations and groupbys
• Improve sql cli tool (line-editing + history)

What's Changed

• Fix bug in DecomposeLambdas Pass
• Smaller Improvements for CT tool
• Use stock LLVM/MLIR 20, don't ship patched version as Python anymore
• Use Apache Arrow 19 don't use pyarrow anymore
• Refactor python package (also switch to nanobind) and build process
• Github actions: add dataset tests
• Refactor dataset generation scripts/SQL
• Add asan build to Makefile/CI and fix use-after-free error
• Replace tbb with custom fiber-based scheduler
• Don't use/require Python anymore for dependency management
• Github actions: build release binaries that should run on most Linux systems
• Fix/Cleanup Makefile

Regressions (expected to be fixed in next release):

• sorting became single-threaded when moving away from tbb
• moving away from tbb introduced smaller overheads when executing fast queries

Full Changelog: v0.0.2...v0.0.3

Changes:

• Improve runtime-header-tool
• Add Python Bindings for MLIR dialects to Python package
• Refactor TableBuilder -> ArrowColumnBuilder
• Smaller features (add a few new runtime functions for strings, floats)
• Sub-Operator: small fixes in lowering
• Bump used LLVM version (also: preparations for using LLVM binary packages without patches after LLVM 20 is released)
• Introduce explicit execution steps in subop dialect to simplifiy further lowerings
• Reimplement the Parallelization pass
• Implement general logic for handling settings
• New infrastructure for profiling/debugging
  • Refactor snapshotting instrumentation
  • Tool for analyzing snapshots and correlating them
  • New "translators" to create a plan in json format from an MLIR module
  • Remove existing web app for profiling (now in webinterface repository)
• Switch coverage tooling to LLVM
• Restructure structure of codebase (incl. namespace lingodb, mlir:: -> lingodb::compiler:: for LingoDB dialects)
• Remove dsa::ForOp and migrate existing use cases to scf.ForOp
• Add a simplified ssb benchmark (often used in GPU papers)
• util: add new BufferGetMemRef operation
• preparations for proper GPU support
• Fix relalg -> json tool
• SubOp: improve PrepareLowering pass to avoid redundant arguments/operations
• Improve CI (e.g., build python packages, docker containers for actions, ...)
• Update Readme

Full Changelog: v0.0.1...v0.0.2

Changes:

• Improve CI Pipelines
• Add/Improve tests
• Bugfixing
• Distribute dependencies as Python packages

First release not motivated by one of the first papers:
Changes:

• Build Basic Python Package for LingoDB
• Bug fixing and small improvements

Second release without explicit version for LingoDB paper (Declarative Sub-Operators for Universal Data Processing):
Main Changes:

• Implement Layer of Sub-Operators
• Parallel Execution based on Sub-Operators
• Modularization of Compilation/Execution Pipeline
• Add new Execution Backends for Debugging/Profiling/Faster compilation(Cranelift)
• New operators/features (e.g., full outer joins, window functions, index nested loop joins,...)
• Improve github actions (e.g., code coverage)
• Supports more benchmarks (e.g., TPC-DS)
• Bug fixes in several parts

Initial release without explicit version for LingoDB paper presented at VLDB 2022.