Version 1.3.0 focused on performance, stability, documentation and API improvements/enhancements.
New features
* New guided tutorial series added to documentation.
* Query times improved dramatically with internal parallelization using TBB (multiple PRs)
* Optional deduplication pass on writes can be enabled [636](https://github.com/TileDB-Inc/TileDB/pull/636)
* New internal statistics reporting system to aid in performance optimization [736](https://github.com/TileDB-Inc/TileDB/pull/736)
* Added string type support: ASCII, UTF-8, UTF-16, UTF-32, UCS-2, UCS-4 [415](https://github.com/TileDB-Inc/TileDB/pull/415)
* Added `TILEDB_ANY` datatype [446](https://github.com/TileDB-Inc/TileDB/pull/446)
* Added parallelized VFS read operations, enabled by default [499](https://github.com/TileDB-Inc/TileDB/pull/499)
* SIGINT signals will cancel in-progress queries [578](https://github.com/TileDB-Inc/TileDB/pull/578)
Improvements
* Arrays must now be open and closed before issuing queries, which clarifies the TileDB consistency model.
* Improved handling of incomplete queries and variable-length attribute data.
* Added parallel S3, POSIX, and Win32 reads and writes, enabled by default.
* Query performance improvements with parallelism (using TBB as a dependency).
* Got rid of special S3 "directory objects."
* Refactored sparse reads, making them simpler and more amenable to parallelization.
* Refactored dense reads, making them simpler and more amenable to parallelization.
* Refactored dense ordered writes, making them simpler and more amenable to parallelization.
* Refactored unordered writes, making them simpler and more amenable to parallelization.
* Refactored global writes, making them simpler and more amenable to parallelization.
* Added ability to cancel pending background/async tasks. SIGINT signals now cancel pending tasks.
* Async queries now use a configurable number of background threads (default number of threads is 1).
* Added checks for duplicate coordinates and option for coordinate deduplication.
* Map usage via the C++ API `operator[]` is faster, similar to the `MapItem` path.
Bug Fixes
* Fixed bugs with reads/writes of variable-sized attributes.
* Fixed file locking issue with simultaneous queries.
* Fixed S3 issues with simultaneous queries within the same context.
API additions
C API
* Added `tiledb_array_alloc`
* Added `tiledb_array_{open, close, free}`
* Added `tiledb_array_reopen`
* Added `tiledb_array_is_open`
* Added `tiledb_array_get_query_type`
* Added `tiledb_array_get_schema`
* Added `tiledb_array_max_buffer_size` and `tiledb_array_max_buffer_size_var`
* Added `tiledb_query_finalize` function.
* Added `tiledb_ctx_cancel_tasks` function.
* Added `tiledb_query_set_buffer` and `tiledb_query_set_buffer_var` which sets a single attribute buffer
* Added `tiledb_query_get_type`
* Added `tiledb_query_has_results`
* Added `tiledb_vfs_get_config` function.
* Added `tiledb_stats_{enable,disable,reset,dump}` functions.
* Added `tiledb_kv_alloc`
* Added `tiledb_kv_reopen`
* Added `tiledb_kv_has_key` to check if a key exists in the key-value store.
* Added `tiledb_kv_free`.
* Added `tiledb_kv_iter_alloc` which takes as input a kv object
* Added `tiledb_kv_schema_{set,get}_capacity`.
* Added `tiledb_kv_is_dirty`
* Added `tiledb_kv_iter_reset`
* Added `sm.num_async_threads`, `sm.num_tbb_threads`, and `sm.enable_signal_handlers` config parameters.
* Added `sm.check_dedup_coords` and `sm.dedup_coords` config parameters.
* Added `vfs.num_threads` and `vfs.min_parallel_size` config parameters.
* Added `vfs.{s3,file}.max_parallel_ops` config parameters.
* Added `vfs.s3.multipart_part_size` config parameter.
* Added `vfs.s3.proxy_{scheme,host,port,username,password}` config parameters.
C++ API
* Added `Array::{open, close}`
* Added `Array::reopen`
* Added `Array::is_open`
* Added `Array::query_type`
* Added `Context::cancel_tasks()` function.
* Added `Query::finalize()` function.
* Added `Query::query_type`
* Added `Query::has_results`
* Changed the return type of the `Query` setters to return the object reference.
* Added an extra `Query` constructor that omits the query type (this is inherited from the input array).
* Added `Map::{open, close}`
* Added `Map::reopen`
* Added `Map::is_dirty`
* Added `Map::has_key` to check for key presence.
* A `tiledb::Map` defined with only one attribute will allow implicit usage, e.x. `map[key] = val` instead of `map[key][attr] = val`.
* Added optional attributes argument in `Map::Map` and `Map::open`
* `MapIter` can be used to create iterators for a map.
* Added `MapIter::reset`
* Added `MapSchema::set_capacity` and `MapSchema::capacity`
* Support for trivially copyable objects, such as a custom data struct, was added. They will be backed by an `sizeof(T)` sized `char` attribute.
* `Attribute::create<T>` can now be used with compound `T`, such as `std::string` and `std::vector<T>`, and other
objects such as a simple data struct.
* Added a `Dimension::create` factory function that does not take tile extent,
which sets the tile extent to `NULL`.
* `tiledb::Attribute` can now be constructed with an enumerated type (e.x. `TILEDB_CHAR`).
* Added `Stats` class (wraps C API `tiledb_stats_*` functions)
* Added `Config::save_to_file`
Breaking changes
C API
* `tiledb_query_finalize` must **always** be called before `tiledb_query_free` after global-order writes.
* Removed `tiledb_vfs_move` and added `tiledb_vfs_move_file` and `tiledb_vfs_move_dir` instead.
* Removed `force` argument from `tiledb_vfs_move_*` and `tiledb_object_move`.
* Removed `vfs.s3.file_buffer_size` config parameter.
* Removed `tiledb_query_get_attribute_status`.
* All `tiledb_*_free` functions now return `void` and do not take `ctx` as input (except for `tiledb_ctx_free`).
* Changed signature of `tiledb_kv_close` to take a `tiledb_kv_t*` argument instead of `tiledb_kv_t**`.
* Renamed `tiledb_domain_get_rank` to `tiledb_domain_get_ndim` to avoid confusion with matrix def of rank.
* Changed signature of `tiledb_array_get_non_empty_domain`.
* Removed `tiledb_array_compute_max_read_buffer_sizes`.
* Changed signature of `tiledb_{array,kv}_open`.
* Removed `tiledb_kv_iter_create`
* Renamed all C API functions that create TileDB objects from `tiledb_*_create` to `tiledb_*_alloc`.
* Removed `tiledb_query_set_buffers`
* Removed `tiledb_query_reset_buffers`
* Added query type argument to `tiledb_array_open`
* Changed argument order in `tiledb_config_iter_alloc`, `tiledb_ctx_alloc`, `tiledb_attribute_alloc`, `tiledb_dimension_alloc`, `tiledb_array_schema_alloc`, `tiledb_kv_schema_load`, `tiledb_kv_get_item`, `tiledb_vfs_alloc`
C++ API
* Fixes with `Array::max_buffer_elements` and `Query::result_buffer_elements` to comply with the API docs. `pair.first` is the number of elements of the offsets buffer. If `pair.first` is 0, it is a fixed-sized attribute or coordinates.
* `std::array<T, N>` is backed by a `char` tiledb attribute since the size is not guaranteed.
* Headers have the `tiledb_cpp_api_` prefix removed. For example, the include is now `include <tiledb/attribute.h>`
* Removed `VFS::move` and added `VFS::move_file` and `VFS::move_dir` instead.
* Removed `force` argument from `VFS::move_*` and `Object::move`.
* Removed `vfs.s3.file_buffer_size` config parameter.
* `Query::finalize` must **always** be called before going out of scope after global-order writes.
* Removed `Query::attribute_status`.
* The API was made header only to improve cross-platform compatibility. `config_iter.h`, `filebuf.h`, `map_item.h`, `map_iter.h`, and `map_proxy.h` are no longer available, but grouped into the headers of the objects they support.
* Previously a `tiledb::Map` could be created from a `std::map`, an anonymous attribute name was defined. This must now be explicitly defined: `tiledb::Map::create(tiledb::Context, std::string uri, std::map, std::string attr_name)`
* Removed `tiledb::Query::reset_buffers`. Any previous usages can safely be removed.
* `Map::begin` refers to the same iterator object. For multiple concurrent iterators, a `MapIter` should be manually constructed instead of using `Map::begin()` more than once.
* Renamed `Domain::rank` to `Domain::ndim` to avoid confusion with matrix def of rank.
* Added query type argument to `Array` constructor
* Removed iterator functionality from `Map`.
* Removed `Array::parition_subarray`.