Disk Format
* Store array schemas under `__schema` directory [2258](https://github.com/TileDB-Inc/TileDB/pull/2258)
New features
* Perform early audit for acceptable aws sdk windows path length [2260](https://github.com/TileDB-Inc/TileDB/pull/2260)
* Support setting via config s3 BucketCannedACL and ObjectCannedACL via SetACL() methods [2383](https://github.com/TileDB-Inc/TileDB/pull/2383)
* Update spdlog dependency to 1.9.0 fixing c++17 compatibility and general improvements [1973](https://github.com/TileDB-Inc/TileDB/pull/1973)
* Added Azure SAS token config support and new config option [2420](https://github.com/TileDB-Inc/TileDB/pull/2420)
* Load all array schemas in storage manager and pass the appropriate schema pointer to each fragment [2415](https://github.com/TileDB-Inc/TileDB/pull/2415)
* First revision of the Interval class [2417](https://github.com/TileDB-Inc/TileDB/pull/2417)
* Add `tiledb_schema_evolution_t` and new apis for schema evolution [2426](https://github.com/TileDB-Inc/TileDB/pull/2426)
* Add `ArraySchemaEvolution` to cpp_api and its unit tests are also added. [2462](https://github.com/TileDB-Inc/TileDB/pull/2462)
* Add c and cpp api functions for getting the array schema of a fragment [2468](https://github.com/TileDB-Inc/TileDB/pull/2468)
* Add capnp serialization and rest support for array schema evolution objects [2467](https://github.com/TileDB-Inc/TileDB/pull/2467)
Improvements
* `encryption_key` and `encryption_type` parameters have been added to the config; internal APIs now use these parameters to set the key. [2245](https://github.com/TileDB-Inc/TileDB/pull/2245)
* Initial read refactor [2374](https://github.com/TileDB-Inc/TileDB/pull/2374)
* Create class ByteVecValue from typedef [2368](https://github.com/TileDB-Inc/TileDB/pull/2368)
* Encapsulate spdlog.h [2396](https://github.com/TileDB-Inc/TileDB/pull/2396)
* Update OSX target to 10.14 for release artifacts [2401](https://github.com/TileDB-Inc/TileDB/pull/2401)
* Add nullable (and unordered, nullable) support to the smoke test. [2405](https://github.com/TileDB-Inc/TileDB/pull/2405)
* Initial sparse global order reader [2395](https://github.com/TileDB-Inc/TileDB/pull/2395)
* Remove sm.sub_partitioner_memory_budget [2402](https://github.com/TileDB-Inc/TileDB/pull/2402)
* Update the markdown documents for our new version of array schemas [2416](https://github.com/TileDB-Inc/TileDB/pull/2416)
* Sparse global order reader: no more result cell slab copy. [2411](https://github.com/TileDB-Inc/TileDB/pull/2411)
* Sparse global order reader: initial memory budget improvements. [2413](https://github.com/TileDB-Inc/TileDB/pull/2413)
* Optimization of result cell slabs generation for sparse global order reader. [2414](https://github.com/TileDB-Inc/TileDB/pull/2414)
* Remove selective unfiltering. [2410](https://github.com/TileDB-Inc/TileDB/pull/2410)
* Updated Azure Storage Lite SDK to 0.3.0 [2419](https://github.com/TileDB-Inc/TileDB/pull/2419)
* Respect memory budget for sparse global order reader. [2425](https://github.com/TileDB-Inc/TileDB/pull/2425)
* Use newer Azure patch for all platforms to solve missing header error [2433](https://github.com/TileDB-Inc/TileDB/pull/2433)
* increased diag output for differences reported by tiledb_unit (some of which may be reasonable) [2437](https://github.com/TileDB-Inc/TileDB/pull/2437)
* Adjustments to schema evolution new attribute reads [2484](https://github.com/TileDB-Inc/TileDB/pull/2484)
* Change `Quickstart` link in readthedocs/doxygen `index.rst` [2448](https://github.com/TileDB-Inc/TileDB/pull/2448)
* Initial sparse unordered with duplicates reader. [2441](https://github.com/TileDB-Inc/TileDB/pull/2441)
* Add calls to `malloc_trim` on context and query destruction linux to potentially reduce idle memory usage [2443](https://github.com/TileDB-Inc/TileDB/pull/2443)
* Add logger internals for `std::string` and `std::stringstream` for developer convenience [2454](https://github.com/TileDB-Inc/TileDB/pull/2454)
* Allow empty attribute writes. [2461](https://github.com/TileDB-Inc/TileDB/pull/2461)
* Refactored readers: serialization. [2458](https://github.com/TileDB-Inc/TileDB/pull/2458)
* Allow null data pointers for writes. [2481](https://github.com/TileDB-Inc/TileDB/pull/2481)
* Update backwards compatibility arrays for 2.3.0 [2487](https://github.com/TileDB-Inc/TileDB/pull/2487)
Deprecations
* Deprecate all `*_with_key` APIs. [2245](https://github.com/TileDB-Inc/TileDB/pull/2245) [#2308](https://github.com/TileDB-Inc/TileDB/pull/2308) [#2412](https://github.com/TileDB-Inc/TileDB/pull/2412)
Bug fixes
* Fix to correctly apply capnproto create_symlink avoidance patch [2264](https://github.com/TileDB-Inc/TileDB/pull/2264)
* The bug for calculating max_size_validity for var_size attribute caused incomplete query [2266](https://github.com/TileDB-Inc/TileDB/pull/2266)
* Always run ASAN with matching compiler versions [2277](https://github.com/TileDB-Inc/TileDB/pull/2277)
* Fix some loop bounds that reference non-existent elements [2282](https://github.com/TileDB-Inc/TileDB/pull/2282)
* Treating `std::vector` like an array; accessing an element that is not present to get its address. [2276](https://github.com/TileDB-Inc/TileDB/pull/2276)
* Fix buffer arguments in unit-curl.cc [2287](https://github.com/TileDB-Inc/TileDB/pull/2287)
* Stop loop iterations within limits of vector being initialized. [2320](https://github.com/TileDB-Inc/TileDB/pull/2320)
* Modify FindCurl_EP.cmake to work for WIN32 -EnableDebug builds [2319](https://github.com/TileDB-Inc/TileDB/pull/2319)
* Fixing test failure because of an uninitialized buffer. [2386](https://github.com/TileDB-Inc/TileDB/pull/2386)
* Change a condition that assumed MSVC was the only compiler for WIN32 [2388](https://github.com/TileDB-Inc/TileDB/pull/2388)
* Fix defects in buffer classes: read, set_offset, advance_offset [2342](https://github.com/TileDB-Inc/TileDB/pull/2342)
* Use CHECK_SAFE() to avoid multi-threaded conflict [2394](https://github.com/TileDB-Inc/TileDB/pull/2394)
* Use tiledb _SAFE() items when overlapping threads may invoke code [2418](https://github.com/TileDB-Inc/TileDB/pull/2418)
* Changes to address issues with default string dimension ranges in query [2436](https://github.com/TileDB-Inc/TileDB/pull/2436)
* Only set cmake policy CMP0076 if cmake version in use knows about it [2463](https://github.com/TileDB-Inc/TileDB/pull/2463)
* Fix handling curl REST request having all data in single call back [2485](https://github.com/TileDB-Inc/TileDB/pull/2485)
* Write queries should post start/end timestamps for REST arrays [2492](https://github.com/TileDB-Inc/TileDB/pull/2492)
API additions
* Introduce new `tiledb_experimental.h` c-api header for new feature that don't have a stabilized api yet [2453](https://github.com/TileDB-Inc/TileDB/pull/2453)
* Introduce new `tiledb_experimental` cpp-api header for new feature that don't have a stabilized api yet [2453](https://github.com/TileDB-Inc/TileDB/pull/2462)
C API
* Refactoring [get/set]_buffer APIs [2315](https://github.com/TileDB-Inc/TileDB/pull/2315)
* Add `tiledb_fragment_info_get_array_schema` functions for getting the array schema of a fragment [2468](https://github.com/TileDB-Inc/TileDB/pull/2468)
* Add `tiledb_schema_evolution_t` and new apis for schema evolution [2426](https://github.com/TileDB-Inc/TileDB/pull/2426)
C++ API
* Refactoring [get/set]_buffer APIs [2399](https://github.com/TileDB-Inc/TileDB/pull/2399)
* Add `FragmentInfo::array_schema` functions for getting the array schema of a fragment [2468](https://github.com/TileDB-Inc/TileDB/pull/2468)
* Add `ArraySchemaEvolution` to cpp_api and its unit tests are also added. [2462](https://github.com/TileDB-Inc/TileDB/pull/2462)