v6.0.0

This release contains the changes from the second Boost review.

What's Changed

• Change default rounding mode (along with consteval methods) by @mborland in #1023
• Reduce evaluation_type duplication and fix missing atan series by @mborland in #1024
• Remove deprecated types, and add note in the docs about this by @mborland in #1025
• Replace bulleted lists with tables by @mborland in #1032
• Implement check for sticky bit to fix rounding for all types and implement for decimal32_t by @mborland in #1030
• Fix lost precision in addition/subtraction and spurious overflows in construction by @mborland in #1037
• Implement proper constructor rounding for decimal64_t by @mborland in #1038
• Add note about not being correctly rounded by @mborland in #1042
• Fix rounding with decimal128_t by @mborland in #1039
• Decrease tested tolerances of sqrt by @mborland in #1043
• Fix rounding in normalize and fast types by @mborland in #1044
• Remove deprecated function by @mborland in #1046
• Allow setting a compile time rounding mode separate from runtime by @mborland in #1045
• Fix rounding correctness of additional 64-bit addition operations by @mborland in #1048
• Optimize normalization for the case where we know which branch to take by @mborland in #1050
• Remove generic digit counting by @mborland in #1051
• Consolidate Addition Impls, and fix impl for 128-bit types by @mborland in #1052
• Correct from_chars handling of infinities by @mborland in #1059
• Allow streaming of value with leading plus sign by @mborland in #1060
• Strip characters from strtod that are allowed, but not allowed in <charconv> by @mborland in #1061
• Fix parsing of payloads by @mborland in #1063
• Fix overconsumption of characters in istream by @mborland in #1064
• Increase precision of 128-bit division by @mborland in #1065
• Improve performance of new rounding facilities by @mborland in #1066
• Update coverage toolchain by @mborland in #1068
• Increase tolerances for decimal32_t pow by @mborland in #1067
• Improve Coverage and annotate unlikely lines by @mborland in #1069
• Remove deprecated macOS version by @mborland in #1073
• Fix double rounding in 32 and 64 bit FMAs by @mborland in #1072
• More coverage and associated bug fixes by @mborland in #1074
• Even More Coverage by @mborland in #1075
• Add testing of modules by @mborland in #1076
• Documentation Updates by @mborland in #1079
• Improve performance of rounding by @mborland in #1078
• Continued coverage improvements by @mborland in #1080
• More extensive testing of DPD conversions by @mborland in #1082
• Cover constructor rounding lines by forcing runtime eval by @mborland in #1084
• Fix build on POWER10 by @mborland in #1086
• Add notes to basics section on underflow, overflow, and non-finite values handling by @mborland in #1090
• Improve message from disallowed 3 arg constructor by @mborland in #1089
• Fix the name of a variable by @asensi in #1096
• Fix construction of overflow/underflow values by @mborland in #1098
• Fix unspecified format precision being 6 digits instead of shortest representation by @mborland in #1102
• Consolidate and expand section on design decisions and deviations by @mborland in #1104
• Improve <cfenv> examples and docs by @mborland in #1101
• Drone: llvm apt installation by @sdarwin in #1109
• Fix the name of a variable in 'basics.adoc' by @asensi in #1108
• Docs: fix example with literals by @akrzemi1 in #1118
• Fix typos in charconv docs by @akrzemi1 in #1117
• Fix handling of extreme size literals by @mborland in #1122
• All members of equivalence class should return the same hash value by @mborland in #1123
• Don't automatically include {fmt} by @mborland in #1124
• Avoid double counting digits in normalization process by @mborland in #1125
• Obey current rounding mode with rint/lrint/nearbyint by @mborland in #1128
• Add ARM64 windows runner and fix missing intrinsic on said platform by @mborland in #1129
• Fix detection of pure powers of 10 for large significands by @mborland in #1130
• Make construction from integral types explicit by default by @mborland in #1131
• Fix decimal_fast32_t escaping it's domain in mul and div by @mborland in #1132
• Up the Newton iterations in cbrt by @ckormanyos in #1134
• Directly manipulate value to calculate next by @mborland in #1133
• Add string constructor to all types by @mborland in #1135
• Fixes for proper handling of IEEE Section 7.2 by @mborland in #1136
• Fix handling of denorm next after by @mborland in #1146
• Rename parameter for clarity by @mborland in #1147
• Fix CI error by @mborland in #1148
• Improve handling of locales by @mborland in #1138
• Add cohort preserving to_chars with scientific format by @mborland in #1157
• Handle Git issue 1137 by @ckormanyos in #1145
• Refactor quantum preservation to be a chars_format option by @mborland in #1159
• Add cohort preserving from_chars by @mborland in #1160
• Add cohort preservation options to {fmt} and by @mborland in #1161
• Implement locale support of {fmt} and <format> by @mborland in #1163
• Add wider character support for {fmt} formatting by @mborland in #1166
• Add wchar_t formatting to std::format support by @mborland in #1170
• Fix locale placement in format string by @mborland in #1172
• Add dectest testing by @mborland in #1173
• Suppress warnings from find_package(fmt) by @mborland in #1176
• Fix construction of extremely large and extremely small numbers by @mborland in #1175
• Bump sha.js from 2.4.11 to 2.4.12 in /doc in the npm_and_yarn group across 1 directory by @dependabot[bot] in #1179
• Fix nan handling and add non-standard payload function for reading by @mborland in #1180
• Change encoding of fast non-finite values to allow for payloads by @mborland in #1182
• Implement total ordering function from IEEE 754 by @mborland in #1183
• Simplify numeric_limits and fix ODR violations by @mborland in #1220
• Improve ADL and bit conversions examples as well as re-locate the latter by @mborland in #1221
• Enable string constructor to parse payloads of NANs by @mborland in #1219
• Update format docs and examples by @mborland in #1222
• Remove old testing facilities ...

v5.2.0

What's Changed

• Update backend by @mborland in #995
• Add Cmake pkg-config support by @mborland in #996
• Fuzz {fmt} and <format> by @mborland in #998
• Fix modules support by @mborland in #1000
• Fix to_chars for 0 in general format by @mborland in #1003
• Merge performance changes from cppalliance/int128 by @mborland in #1005
• Test and fix consteval to_chars by @mborland in #1006
• Implement fast path of no precision scientific formatting by @mborland in #1012
• Fast path for unspecified precision fixed format to_chars by @mborland in #1013
• Add fast path for shortest length hex formatting by @mborland in #1016
• New <charconv> path bug fixes by @mborland in #1019
• Deprecate name of fenv header. Change to cfenv by @mborland in #1020

Full Changelog: v5.1.2...v5.2.0

v5.1.2

What's Changed

• Fixed a bug with printing/streaming/formatting of zeros in general format resulted in "0.0e+00"

v5.1.1

What's Changed

• Test {fmt} using header only so it actually runs in CI by @mborland in #983
• Fix sign processing bug in {fmt} by @mborland in #987
• Fix to_chars rounding for 0.999... -> 1(.0000) by @mborland in #989
• Fix sign parsing with <format> by @mborland in #990
• Fix formatting support so that we can format into the middle of a string by @mborland in #992

Full Changelog: v5.1.0...v5.1.1

v5.1.0

What's Changed

• Add and document user facing normalization function by @mborland in #956
• Fix variable shadowing by @mborland in #957
• Update style of docs by @mborland in #958
• Fix escaping of C++ and punctuation by @mborland in #960
• Improve documentation on relevant headers by @mborland in #963
• Division Improvements for 64 and 128 bit types by @mborland in #964
• Expand the fast type explanation section by @mborland in #966
• Add formatting and printing examples by @mborland in #969
• Coverage Improvements by @mborland in #970
• Add user and automatic configuration macro to disable exceptions by @mborland in #972
• Improve coverage and fix ceil bug by @mborland in #974
• Fix significand being used for long double conversion for 128-bit types by @mborland in #976
• Fix duplicate symbols on some PPC64 platforms by @mborland in #978
• Small improvements and bug fix to equality implementation by @mborland in #979
• Fix <format> support for clang by @mborland in #982

Full Changelog: v5.0.0...v5.1.0

v5.0.0

This release address the review feedback from the first Boost review in January 2025, and many issues along the way.

BREAKING CHANGES include:

• Fixed internal encoding of IEEE 754 types based on various feedback and library comparisons
• Fast types now use fixed width integers internally instead of std::uint_fastXX_t reducing size on many platforms
• All type names have been changed to better align with established practice:
  • decimalXX is now decimalXX_t
  • decimalXX_fast is now decimal_fastXX_t
• Header paths have changed similarly: <boost/decimal/decimal32.hpp> becomes <boost/decimal/decimal32_t.hpp>
• A new 128-bit integer backend is now used in computations, and in the backend of decimal128_t and decimal_fast128_t. This integer has different alignment (16 bytes) than the previous backend (8 bytes).

What's Changed

• Fix 32 and 64 bit encodings by @mborland in #914
• Fix decimal128 encoding and test against libbson (when available) by @mborland in #918
• Refactor int128 location by @mborland in #920
• Remove conversions to 8 and 16 bit integer types by @mborland in #921
• Add additional constraints to the constructors by @mborland in #922
• Move to Antora Based Documentation by @mborland in #924
• Address review documentation improvements by @mborland in #925
• 917 Part 1 (decimal32_fast) by @mborland in #926
• Fix boundary check in to_integer by @mborland in #928
• Fix to_chars for decimal128 being disproportionally slower by @mborland in #929
• Rename types and add deprecation warnings to the old ones by @mborland in #930
• Remove duplicated test for infinity by @mborland in #931
• Update macOS performance documentation and benchmarks bug fixes by @mborland in #935
• Misc testing and doc improvements by @mborland in #936
• Handle unclear code from local metal benching by @ckormanyos in #937
• Misc doc improvements by @mborland in #939
• Change BID and DPD conversions default return types by @mborland in #941
• Update Linux Benchmarks and Add Intel Benchmarks by @mborland in #944
• Update x64 Windows Benchmarks in Docs by @mborland in #945
• Improve layout of benchmark documentation by @mborland in #946
• Replace internal fast types with fixed types by @mborland in #948
• Improve const correctness in library functions by @mborland in #949
• Merge branch 'develop' into local_metal_benching by @ckormanyos in #938
• Improve const correctness, comments, and large parameter passing by @mborland in #950
• Bugfixes for Intel Benchmarks and update docs with complete results by @mborland in #952
• Merge to Master for v5.0.0 by @mborland in #953

Full Changelog: v4.1.0...v5.0.0

v4.1.0

What's Changed

• Implement new U256 backend by @mborland in #899
• U256 integration by @mborland in #905
• Improve and fix digit counting by @mborland in #907
• Add fmtlib support by @mborland in #908
• Improve benchmarks and minor fixes by @mborland in #910
• Improve decimal128_fast multiplication performance by @mborland in #912

Full Changelog: v4.0.0...v4.1.0

v4.0.0

This release contains many of the issues that were reported during the review period in Jan of this year.

What's Changed

• Fix limits and printing of subnormal_min by @mborland in #796
• Move literals into namespace literals by @mborland in #809
• Fix narrowing in equality of mixed types by @mborland in #810
• Fix calculation of resultant digits to keep us in the range of uint128 by @mborland in #812
• Remove sprintf and fix conceptual requirements for by @mborland in #813
• Make -0 = +0 by @mborland in #815
• Allow implicit widening between decimal types and keep explicit narrowing by @mborland in #816
• Allow separate header usage and test all includes by @mborland in #818
• Change values of fast nans and infs to reduce comparisons by @mborland in #820
• Fix decimal128_fast scalblnd implementation by @mborland in #821
• Make uint128 public since we return in a few places to the user by @mborland in #822
• Move concepts into detail namespace by @mborland in #847
• Remove decimal32 bool constructor by @mborland in #849
• Fix nextafter and nexttowards by @mborland in #850
• Use delegating constructor for Integers by @mborland in #852
• Improve and document explicit conversion saturation by @mborland in #851
• Make powers of 10 calculations constexpr by @mborland in #853
• Increase ellint tolerances by @mborland in #855
• Test fixed zero with precision argument by @mborland in #860
• Replace usages of deprecated OS by @mborland in #861
• Document typedefs by @mborland in #862
• Reduce code duplication in 32 bit types by @mborland in #858
• Faster 32-bit mul by @mborland in #865
• Add one-shot decoding for 64-bit types and improve mul by @mborland in #866
• Use traits to reduce duplication / improve logic by @mborland in #863
• Remove comparison operator code duplications by @mborland in #867
• Fix library alias by @mborland in #871
• 64 bit add and sub by @mborland in #872
• Fix slowdown with 64 bit mul by @mborland in #873
• Prune some low hanging charconv optimization branches by @mborland in #874
• Improve d128 mul by @mborland in #875
• Improve Charconv coverage by @mborland in #876
• Fix division near 0 by @mborland in #877
• Fix next after part 2 by @mborland in #878
• Separate GCC 13 32 bit and 64 bit drone runs by @mborland in #882
• Small optimizations for frexp10 by @mborland in #881
• Remove duplicate digit counting from charconv by @mborland in #879
• 128-Bit division simplifications by @mborland in #884
• Reduce digit counting in constructors by @mborland in #883
• Add retry clone to jsonnet by @mborland in #889
• Update ci by @mborland in #896
• Change 128-bit backend to cppalliance/int128 by @mborland in #897
• Set width of istream by @mborland in #898
• Fix comparisons to zero by @mborland in #901

Full Changelog: v3.0.1...v4.0.0

v3.0.1

Fixes segfault in to_chars when using chars_format::fixed with unspecified precision

v3.0.0

What's Changed

• Add additional fast math support by @mborland in #749
• Enable interoperability with std:: types from <charconv> by @mborland in #752
• Add literals for fast types by @mborland in #753
• Add module support to the fast types by @mborland in #754
• Enable 80 and 128 bit long double conversions by @mborland in #756
• Replace macos-12 with 15 by @mborland in #755
• Update GDB run in metal CI by @ckormanyos in #760
• Refactor rescale by @mborland in #762
• Remove IBM test cases to be safe on licensing by @mborland in #763
• Add financial calculations to examples by @mborland in #765
• Remove cruft by @mborland in #766
• Add string and string_view interfaces to from_chars by @mborland in #768
• Fix deprecated whitespace in literal definition by @mborland in #771
• Add 32-bit FMA implementation by @mborland in #770
• Update fuzz by @mborland in #772
• Re-activate MSVC 14.3 and Clang-CL testing and Improve Intel Testing by @mborland in #775
• Improve Coverage by @mborland in #776
• Add example using Boost.Math by @mborland in #774
• Fix for 777 by @mborland in #778
• Support for by @mborland in #363
• Implement proper 64 bit FMA by @mborland in #779
• Fix macro naming for format by @mborland in #780
• Capital format by @mborland in #781

Full Changelog: v2.4.0...v3.0.0