rust_decimal

• Performance improvements for to_f64 when using a scale > 0.

Special thank you to @hengchu who discovered and resolved the issue!

• Allow uppercase "E" in scientific notation.
• Allow scientific notation in dec! macro.
• Deprecate set_sign and replace with set_sign_positive and set_sign_negative. This is intended to improve the readability of the API.
• Fixes precision issue when parsing f64 values. The base 2 mantissa of the float was assuming guaranteed accuracy of 53 bit precision, however 52 bit precision is more accurate (f64 only).
• Removes deprecated usage of Error::description.

• Replace num dependency with num_trait - implemented Signed and Num traits.

• Fixes issue whereby overflow would occur reading from PostgreSQL with high precision. The library now handles this by rounding high precision numbers as they're read as opposed to crashing (similar to other underflow situations e.g. 1/3).

• Retain trailing zeros from PostgreSQL. This ensures that the scale is maintained when serialized into the Decimal type.
• Fixes issue where -0 != 0 (these are now equivalent - thank you @hengchu for discovering).
• Improve hashing function so that the following property is true: k1 == k2 -> hash(k1) == hash(k2)
• Update normalize function so that -0 normalizes to 0.

Special thanks to @hathawsh for their help in this release!

• Update to Postgres 0.17 and add postgres async/await support via tokio-pg
• Added option for serializing decimals as float via serde-float

Special thanks to @pimeys and @kaibyao!

Updates dependencies to prevent build issues.

Bug fix release:

• Fixes issue where scaling logic produced incorrect results when one arm was a high precision zero. Thank you @KonishchevDmitry!

Bug fix release:

• Fixes issue where ToSql was incorrectly calculating weight when whole portion = numeric portion.
• Fixes issue where Decimal::new incorrectly handled i64::max_value() and i64::min_value().
• Fixes issue where rem operation incorrectly returned results when scale was required.

This release represents the start of semantic versioning and allows the library to start making fundamental improvements under the guise of V2.0. Leading up to that I expect to release 1.x versions which will include adding various mathematical functions such as pow, ln, log10 etc.

Version 1.0.0 does come with some new features:

• Checked Operations! This implements checked_add, checked_sub, checked_mul, checked_div and checked_rem.
• Fixes overflow from max_value() and min_value() for i32 and i64.
• Minor documentation improvements and test coverage.

Special thanks to @0e4ef622 for their help with this release!

• Add prelude to help num trait inclusion (use rust_decimal::prelude::*)
• Add Default trait to the library. This is equivalent to using Decimal::zero()
• Added assignment operators for references.

Special thanks to @jean-airoldie for his help with this release!

• Fall back to from_scientific when from_str fails during deserialization. Thanks @mattjbray!
• Added basic Sum trait implementation

• Fixes a bug in floor and ceil where negative numbers were incorrectly handled.

• Macros are now supported on stable. This does use a hack for the meantime so due diligence is required before usage.
• Fixes issue when parsing strings where an underscore preceded a decimal point.
• const_fn support via a feature flag. In the future this will be the default option however in order to support older compiler versions is behind a feature flag.

• Macros (nightly) now output structural data as opposed to serialized data. This is fully backwards compatible and results in some minor performance improvements. Also, removed feature gate so that it can be compiled in stable.
• Fixes a string parsing bug when given highly significant numbers that require rounding.

• Bumped dependencies to remove some legacy serialization requirements.

Special thanks to @xilec, @snd and @AndrewSpeed for their help with this release.

• New rounding strategies introduced via round_dp_with_strategy. Previously default rounding support used bankers rounding by default whereas now you can choose to round the half way point either up or down.
• PostgreSQL write performance improved so that it is at least 3 times faster than the previous implementation.
• Debug trait now outputs the actual decimal number by default to make it more useful within consuming libraries ( e.g. criterion.rs). To get something similar to the previous functionality you can use the unpack argument - this is likely for core rust-decimal library maintainers.
• Various other performance improvements for common operations such as rescale, sub and div.

• Performance optimization for add.

• Introduces the Neg trait to support the ability to use -decimal_variable.
• Fixes bug with underflow on addition.

This release updates the published documentation only and is a no-op for functionality.

• Introduces from_scientific allowing parsing of scientific notation into the Decimal type.
• Fixes a bug when formatting a number with a leading zero's.

• Fixes bug in rescale whereby scaling which invoked rounding incorrectly set the new scale for the left/right sides.

• Fixes bug in cmp whereby two negatives would return an incorrect result.
• Further documentation examples
• Small improvements in division logic
• New abs, floor and ceil functions.

This is a minor version bump as we slowly build our way towards 1.0. Thank you for everyone's support and help as we get there! This has a few notable changes - also introducing a few new interfaces which is the reason for the version bump:

• from_parts function to allow effective creation of Decimal's without requiring binary serialization. An example of this benefit is with the lazy static group initializers for Postgres.
• normalize function to allow stripping trailing zero's easily.
• trunc function allows truncation of a number without any rounding. This effectively "truncates" the fractional part of the number.
• fract function returns the fractional part of the number without the integral.
• Minor improvements in some iterator logic, utilizing the compiler for further optimizations.
• Fixes issue in string parsing logic whereby _ would cause numbers to be incorrectly identified.
• Many improvements to mul. Numbers utilizing the lo portion of the decimal only will now be shortcut and bigger numbers will now correctly overflow. True overflows will still panic, however large underflows will now be rounded as necessary as opposed to panicing.
• Hash was implemented by convention in 0.6.5 however is reimplemented explicitly in 0.7.0 for effectiveness.
• PostgreSQL read performance improved by pre-caching groups and leveraging normalize (i.e. avoiding strings). Further optimizations can be made in write however require some div optimizations first.
• Added short circuit write improvement for zero in PostgreSQL writes.
• Benchmarks are now recorded per build so we can start tracking where slow downs have occurred. This does mean there is a performance hit on Travis builds however hopefully the pay off will make it worthwhile.

Fixes issue with rescale sometimes causing a silent overflow which led to incorrect results during addition, subtraction and compare. Consequently Decimal now rounds the most significant number so that these operations work successfully.

In addition, Decimal now derive's the Hash trait so that it can be used for indexing.

Fixes silent overflow errors when parsing highly significant strings. from_str will now round in these scenario's, similar to oleaut32 behavior.

Fixes a regression in ordering where by different scales would be rescaled towards losing precision instead of increasing precision. Have added numerous test suites to help cover more issues like this in the future. Also fixes an issue in parsing invalid strings whereby the precision exceeded our maximum precision. Previously, this would work with unintended results however this now returns an Error returned from FromStr.

Fixes an issue with division of rational numbers allowing results greater than MAX_PRECISION. This would ultimately cause issues for future operations on this number. In addition, in some cases transitive operations would not be equal due to overflow being lost.

This minor release is purely to expose rust_decimal_macros for use on the nightly channel. Documentation has been updated accordingly.

This release has a few major changes to the internal workings of the Decimal implementation and consequently comes with a number of performance improvements.

• Floats can now be parsed into a Decimal type using from_f32 and from_f64.
• add, sub, mul run roughly 1500% faster than before.
• div run's roughly 1000% faster than before with room for future improvement.
• Also get significant speed improvements with cmp, rescale, round_dp and some string manipulations.
• Implemented *Assign traits for simpler usage.
• Removed BigInt and BigUint as being intermediary data types.

Minor bug fix to prevent a panic from overflow during comparison of high significant digit decimals.

Minor bux fix to prevent panic upon parsing an empty string.

• Removes postgres from default feature set.
• bincode support for serde
• Better support for format strings
• Benchmarks added to tests

Fixes bug in cmp whereby negative's were not being compared correctly.

Minor bug fix to support creating negative numbers using the default constructor.

This release is a stylistic cleanup however does include some minor changes that may break existing builds.

Changed

• Serde is now optional. You can enable Serde support within features using the keyword serde.
• Serde now returns errors on invalid input as opposed to 0.
• f64 conversion support has been added.
• Update Postgres dependency to use v0.15.

This is a documentation release that should help with discoverability and usage.

Changed

• Removed trait ToDecimal and replaced with builtin From trait (#12)