[/
            Copyright Hans Dembinski 2018 - 2019.
   Distributed under the Boost Software License, Version 1.0.
      (See accompanying file LICENSE_1_0.txt or copy at
            https://www.boost.org/LICENSE_1_0.txt)
]

[section:changelog Changelog]

[heading Boost 1.82]

* Replace `detail::span` and `detail::make_span` with implementations in `boost::core`
* Documentation improvements
* Protect usage of `std::min` and `std::max` in some cases, contributed by Han Jiang (min,max macros are illegially set by popular Windows headers so we need to work around)
* Added test to catch usage of unprotected min,max tokens in the library in the future 
* Fixes to support latest clang-14 and deduction guides in gcc-11+

[heading Boost 1.81]

* Added new `accumulators::fraction` to compute fractions, their variance, and confidence intervals
* Added interval computers for fractions: `utility::clopper_pearson`, `utility::wilson_interval`, `utility::jeffreys_interval`, `utility::wald_interval` which can compute intervals with arbitrary confidence level
* Added `utility::confidence_level` and `utility::deviation` types to pass confidence levels as probabilities or in multiples of standard deviation for all interval computers, respectively
* Fixed internal `sub_array` and `span` in C++20

[heading Boost 1.80]

* Fixed segfault in `indexed` when trying to iterate over histogram with axes of zero size (physical or logical) under certain conditions
* Removed previously deprecated API:
  * class `accumulators::thread_safe`: use `accumulators::count<T, true>`
  * Methods `accumulators::sum::large` and `accumulators::sum::small`: use `accumulators::sum::large_part` and `accumulators::sum::small_part`
  * Type alias `algorithm::reduce_option`: use `algorithm::reduce_command`
  * Template function `axis::traits::static_options`: use `axis::traits::get_options`
  * Template function `axis::traits::static_is_inclusive`: use `axis::traits::is_inclusive`
  * Type alias `indexed::range_iterator`: use `indexed::iterator`
  * Type alias `indexed::accessor::reference`: use `indexed::accessor::const_reference`

[heading Boost 1.79]

* Division support for `accumulators::weighted_sum`
* `axis::regular` and `axis::variable` now represent closed intervals if overflow bin is absent, this is needed for example to correctly bin a probability in the closed interval [0, 1]
* cpp17 support: set axis compile-time options via deduction guide

[heading Boost 1.78]

* Deprecate `accumulators::sum::small` and `::large` as workaround for bug in windows.h
* Fixed broken division operator in `accumulators::count`
* Fix bug in 1D histogram::fill when axis is growing and argument is single value
* Improvements to documentation

[heading Boost 1.77]

* `indexed` now allows iteration over user-defined rectangular subrange
* Deprecate `accumulators::thread_safe`, users should replace with optionally thread-safe `accumulators::count`
* Fixed broken scaling operator in `accumulators::count`
* Fixed bug in `axis::integer::inclusive`
* Improvements to ostreaming: support for more accumulators, drawing unicode bar charts is now the default

[heading Boost 1.76]

* Fixes
  * `algorithm::reduce` with `crop` now does not remove the counts in flow bins anymore if the selected range actually overlaps with the flow bins, making the treatment of flow bins consistent with inner bins
  * `accumulators::mean` and `accumulators::weighted_mean` now compute the correct variance when `operator+=` was called, for example, when histograms with accumulators are added; this was wrong before leading to too small variances
  * detection of weight support in user-defined accumulators was broken at compile-time if accumulator used `operator+=` instead of `operator()`; tests only accidentally passed for builtin `weighted_sum`

[heading Boost 1.75]

* Fixes
  * `axis::variable{{0, 1, inf}}.value(1)` now returns 1 instead of NaN
  * `axis::variable` ctor now throws `std::invalid_argument` if one values is NaN
  * Implemented a workaround for a gcc bug (will be fixed upstream in gcc-10.x) that occurred when `<boost/type_traits.hpp>` was included before `<boost/histogram/axis/traits.hpp>`
  * Fixed several new warnings from the latest compiler versions
  * Fixed warnings that appear when compiling against the C++20 standard

* Documentation
  * Documented undefined behaviour when value and variance are requested for `accumulator::mean` and `accumulator::weighted_mean` but the accumulator does not have enough entries to compute them
  * Added explanation to rationale why axis metadata is checked for equality

[heading Boost 1.74]

* New features
  * New axis `boolean`, a special kind of integer axis with only two values and no state (except metadata)
  * Histograms with growing category axes are now addable even if the categories differ; the category axis of the output histogram is the union of the category axes of the input histograms
  * Added axis::traits::continuous to match axis::traits::is_continuous

* Fixes
  * `metadata()` method of builtin axis types now return a mutable reference even if the axis is const, as described in the documentation

* Documentation
  * Sorted headers in reference and sorted items in namespaces
  * Removed some internal typenames from templates
  * Added reference for `sample_type`

* Other
  * Simplified internal metaprogramming
  * Replaced Boost Assert with plain cassert

[heading Boost 1.73]

* New features
  * accumulators
    * New `count` accumulator, a thin wrapper around a primitive arithmetic type; serves as a base class for users who want to add arbitrary metadata to each bin
    * `sum` gained a `value()` method for consistency with other accumulators
  * algorithm
    * `reduce`
      * Now supports positional commands
      * `category` axis now supports `slice` command
      * Added `crop` command, like `shrink`, but discards content of removed bins
      * Renamed `reduce_option` to `reduce_command` (old name available but deprecated)
    * `sum` gained a new argument, allowing users to select whether to sum over all bins or only the inner bins (excluding underflow and overflow)
  * axis::traits
    * Added traits `is_ordered` and `ordered` to distinguish between different kinds of DiscreteAxis types; used internally to handle `category` axis and `integer` axis differently
    * Renamed `static_options` to `get_options` and `static_is_inclusive` to `is_inclusive` (old names available but deprecated)

* Fixes
  * Fixed failing ODR test in bundled Boost due to wrong paths
  * Fixed a bug in histogram::operator*= that could occur in user-defined accumulators
  * Fixed a bug in `indexed` which could cause arithmetic past the end position
  * Fixed a bug where the variance in `accumulators::weighted_sum` was not correctly computed if a normal histogram was added to the histogram that used `accumulators::weighted_sum`
  * Fixed a bug where compiling with `-ffast-math` would break `accumulators::sum`
  * Fixed `algorithm::reduce` to work with axes without *flow bins, which did not compile before

* Other
  * Added an example and documentation on how to use Boost.Histogram as a backend
  * Improved the docs on how to use custom accumulators and Boost.Accumulators
  * Many small documentation improvements
  * Improved docs for `algorithm::reduce`
  * Improved experimental cmake build support
  * Use `index_type` consistently everywhere instead of raw `int`
  * Benchmark update in the docs

[heading Boost 1.72]

* New features
  * Better deduction guides for axis types and histogram type on C++17 compilers
  * performance improvements in the indexing code
  * new histogram::fill method for accelerated filling from chunks of values
  * ASCII bar charts for 1D histograms when boost/histogram/ostream.hpp is included (contributed by Przemyslaw Bartosik)
  * Passing invalid axis options causes user-friendly compile-time errors
  * Wrong usage of weight() and sample() causes user-friendly compile-time errors
  * algorithm::empty returns true if all histogram values are equal to the default value (contributed by Henry Schreiner)
  * algorithm::reduce with shrink is now well defined when values are not bin edges
  * axis::traits::rank returns the number of values that this axis accepts
  * axis::traits::is_continuous returns whether an axis is continuous or discrete
  * axis::traits::is_inclusive returns whether an axis has a bin for any possible input
  * limited weight support for mean accumulator
  * accumulators::weighted_mean::sum_of_weights_squared method added

* Fixes
  * Support of -std=c++2a and -stdlib=libc++ on clang, compatibility with gcc-9
  * Fixed: weight could be passed instead of a sample during filling without triggering an error
  * Fixed: segfault when too many indices were passed to algorithm::project
  * Fixed: indexed range generator did not work with storage based on std::array
  * Fixed: weighted_mean() + weighted_mean() != weighted_mean() (discovered and reported by Henry Schreiner)
  * Fixed: axis::option::test(...) returned true if any bits in the test mask were set (now returns true if all bits in the test mask are set)
  * Protecting calls to min, max with against macro expansion
  * Replaced all crlf with lf in concepts.qbk
  * Missing throw of std::invalid_argument when user passes too many axes to histogram
  * Corrected documentation of serialization support in concepts section
  * Resolved warnings for various compiler versions

* Other
  * Added Boost.Histogram logo
  * Added missing copyright notices
  * axis::category::value returns copy for scalar types and const reference otherwise
  * std::ostringstream not used anymore to generate exception messages to reduces code bloat
  * Documentation improvements
    * Restructured overview section in documentation
    * Updated user guide to demonstrate new features
    * Updated accumulator examples
    * Concepts explain how accumulators can optionally accept weights
  * Updated benchmark code
  * New test checks consistency of b2 and cmake build systems
  * New test checks One-Definition-Rule (fails if non-templated function is not declared inline)
  * Serialization code decoupled from Boost.Serialization
  * Removed dependency on Boost.CallableTraits

[heading Boost 1.71]

* New features
  * Thread-safe accumulators boost::histogram::accumulators::thread_safe based on std::atomics
  * Support for thread-safe storages
  * Support for compiling without exceptions/RTTI (increases performance by 10-20 %) [with contributions from Glen Fernandez]
  * Performance improvements for 1D and 2D histograms
  * boost::histogram::indexed now returns forward iterator range instead of input iterator range
  * boost::histogram::indexed_range::accessor is now non-copyable and acts like reference to cell value, making most algorithms from the stdlib work
  * boost::histogram::algorithm::reduce
    * New slice option
    * Fuse shrink, slice, and rebin options passed for the same axis
    * Support histograms with some axis types with reduction support
  * boost::histogram::algorithm::project accepts runtime indices for static histograms

* Fixes
  * boost::histogram::algorithm::reduce also works on histograms that have some axis types without reduction support
  * boost::histogram::axis::traits::update now works correctly for boost::histogram::axis::variant

* Other
  * 100 % test coverage
  * Drastically reduced internal Boost dependencies
  * Improved documentation and examples
  * Internally replaced boost::variant with boost::variant2
  * boost::histogram::axis::traits::is_reducible detects reducible axis types
  * Cleanup and refactoring of internals
  * Guarantee no-throw moves for all builtin axis types
  * Improved internal benchmarks
  * Compile cleanly at higher warning levels

[heading Boost 1.70]

First Boost release, version 4.0 in former internal counting.

* Removed Python bindings, will be developed in separate repository
* All axes can be made optionally circular, except category axis
* All axes now support generic attached metadata
* All axes have now independently configurable underflow/overflow extra bins
* Removed circular axis (which is just a circular regular axis)
* Added indexed adaptor generator for convenient and fast iteration over histograms
* Support for axes that can grow in range
* Support for axes which accept multiple values (example: hexagonal binning)
* Support for profiles and more generally, arbitrary accumulators in each cell
* Support for serializing to xml archives
* Added compatibility with Boost.Range, Boost.Units, and Boost.Accumulators
* Added deduction guides for axis types and histogram
* Performance improvements
* Renamed `adaptive_storage` to `unlimited_storage`
* Replaced `boost::multiprecision::cpp_int` with custom type to decouple libraries
* Internal simplification of `axis::variant`
* Make all storages behave more like containers, simplifying the design
* Histograms supports add, sub, mul, div
* replaced `operator[]` in axis with explicit method `bin`
* replaced `operator()` in axis with explicit method `index`
* replaced internal use of `boost::containers` with stdlib containers
* Much improved docs, reference documentation, user guide, more examples

[heading 3.2 (not in Boost)]

* Allocator support everywhere
* Internal refactoring

[heading 3.1 (not in Boost)]

* Renamed `bincount` method to `size`
* Support for axes with only overflow and no underflow bin
* category axis now by default has bin for "other" input that does not fall
  into the predefined categories, making it consistent with other axes
* NaN is now consistently put into overflow bin for all axes
* Eliminated warnings about safe internal conversions on MSVC
* Established a cpp house style with corresponding .clang-format file
* Better detection of Python library on all systems
* Improved code coverage by testing more input errors
* Raise ValueError instead of generic RuntimeError in Python on input errors

[heading 3.0 (not in Boost)]

* Support for efficient adding of multiple histograms and scaling
* Re-design of category axis as a general mapping between unique values and bins
* Re-design of the bin description an axis returns upon element access
* Interface cleanup and simplification, more consistency between Python and C++ interface
* Iterators for histogram and axes for STL compatibility
* Regular axis in C++: Allowing transforms with state
* Regular axis in Python: Support for all C++ transforms
* Polymorphic axis::any type
* Internal refactoring from boost::mpl and boost::fusion to std::tuple and boost::mp11
* Windows support

[heading 2.0 (not in Boost)]

* Added static_histogram (v1.0 only had dynamic_histogram).
* Merged wfill(...) and fill(...) interface.
* Support custom allocators in storage classes.
* Replaced static_storage with array_storage.
* Replaced dynamic_storage with unlimited_storage, which adds the capability to grow the bin counter into a cpp_int, thus avoiding integer overflow completely.
* Serialization uses binary_archive instead of text_archive. The latter is portable, but the performance is terrible.
* Python interface changed: histograms are now iterable, returning axis classes
* Support reduction (removing a subset of axes and returning the equivalent histogram; in other words, those axes are integrated over)

[heading 1.0 (not in Boost)]

* First stable version.

[endsect]