Contributors: @cuviper, @michaelciraci
Euclid::div_rem_euclid and CheckedEuclid::checked_div_rem_euclid methods compute and return the quotient and remainder at the same time.TotalOrder trait implements the IEEE 754 totalOrder predicate.ConstZero and ConstOne traits offered associated constants, extending the non-const Zero and One traits for types that have constant values.Contributors: @andrewjradcliffe, @cuviper, @tarcieri, @tdelabro, @waywardmonkeys
Contributors: @robamu
ToBytes and FromBytes traits convert to and from byte representations of a value, with little, big, and native-endian options.Float::is_subnormal method checks for subnormal values, with a non-zero magnitude that is less than the normal minimum positive value.Contributors: @ctrlcctrlv, @cuviper, @flier, @GuillaumeGomez, @kaidokert, @rs017991, @vicsn
Euclid trait calculates Euclidean division, where the remainder is always positive or zero.LowerBounded and UpperBounded traits separately describe types with lower and upper bounds. These traits are automatically implemented for all fully-Bounded types.Float::copysign method copies the sign of the argument to to the magnitude of self.PrimInt::leading_ones and trailing_ones methods are the complement of the existing methods that count zero bits.PrimInt::reverse_bits method reverses the order of all bits of a primitive integer.Num::from_str_radix for floats, also ignoring case.Float and FloatCore use more from libm when that is enabled.Contributors: @alion02, @clarfonthey, @cuviper, @ElectronicRU, @ibraheemdev, @SparrowLii, @sshilovsky, @tspiteri, @XAMPPRocky, @Xiretza
Contributors: @cuviper
OverflowingAdd, OverflowingSub, and OverflowingMul traits return a tuple with the operation result and a bool indicating overflow.autocfg probing doesn't work properly.f64 values to f32 now saturate to infinity. They previously returned None because that was once thought to be undefined behavior, but rust#15536 resolved that such casts are fine.Num::from_str_radix documents requirements for radix support, which are now more relaxed than previously implied. It is suggested to accept at least 2..=36 without panicking, but Err may be returned otherwise.Contributors: @cuviper, @Enet4, @KaczuH, @martin-t, @newpavlov
WrappingNeg trait will wrap the result if it exceeds the boundary of the type, e.g. i32::MIN.wrapping_neg() == i32::MIN.SaturatingAdd, SaturatingSub, and SaturatingMul traits will saturate at the numeric bounds if the operation would overflow. These soft-deprecate the existing Saturating trait that only has addition and subtraction methods.FloatConst::{LOG10_2, LOG2_10}.Contributors: @cuviper, @ocstl, @trepetti, @vallentin
Contributors: @cuviper, @m-ou-se
Contributors: @CryZe
libm dependency enables the Float and Real traits in no_std builds.clamp_min and clamp_max limit minimum and maximum values while preserving input NANs.from_str_radix on invalid signs.Contributors: @cuviper, @dingelish, @HeroicKatora, @jturner314, @ocstl, @Shnatsel, @termoshtt, @waywardmonkeys, @yoanlcq
Contributors: @cuviper
CheckedShl and CheckedShr return None.Zero::set_zero and One::set_one will set values to their identities in place, possibly optimized better than direct assignment.PrimInt.Contributors: @cuviper, @dvdhrm, @ignatenkobrain, @lcnr, @samueltardieu
pow(0, 0) returns 1. Mathematically, this is not strictly defined, but the current behavior is a pragmatic choice that has precedent in Rust core for the primitives and in many other languages.WrappingShl and WrappingShr traits will wrap the shift count if it exceeds the bit size of the type.Contributors: @cuviper, @edmccard, @meltinglava
mul_add now clarifies that it's not always faster.FromPrimitive and ToPrimitive are more robust.Contributors: @cuviper, @frewsxcv
i128 crate feature now causes the build script to panic if such support is not detected.Contributors: @cuviper
CheckedNeg and CheckedRem traits perform checked Neg and Rem, returning Some(output) or None on overflow.no_std implementation of FloatCore::to_degrees for f32 now uses a constant for greater accuracy, mirroring rust#47919. (With std it just calls the inherent f32::to_degrees in the standard library.)MulAdd and MulAddAssign traits perform a fused multiply- add. For integer types this is just a convenience, but for floating point types this produces a more accurate result than the separate operations.i128 crate feature. The FromPrimitive and ToPrimitive traits now also have corresponding 128-bit methods, which default to converting via 64-bit integers for compatibility.Contributors: @cuviper, @LEXUGE, @regexident, @vks
None on overflow, avoiding rustc's undefined behavior. This applies to the cast function and the traits NumCast, FromPrimitive, and ToPrimitive.Contributors: @apopiak, @cuviper, @dbarella
FloatCore trait offers a subset of Float for #![no_std] use. This includes everything except the transcendental functions and FMA.Inv trait returns the multiplicative inverse, or reciprocal.Pow trait performs exponentiation, much like the existing pow function, but with generic exponent types.One::is_one method tests if a value equals 1. Implementers should override this method if there's a more efficient way to check for 1, rather than comparing with a temporary one().Contributors: @clarcharr, @cuviper, @vks
std feature, enabled by default, along with the implication that building without this feature makes this a #![no_std] crate.Float and Real traits are only available when std is enabled.Contributors: @cuviper, @termoshtt, @vks
ParseFloatError now implements Display.AsPrimitive trait implements generic casting with the as operator.CheckedShl and CheckedShr traits implement generic support for the checked_shl and checked_shr methods on primitive integers.Real trait offers a subset of Float functionality that may be applicable to more types, with a blanket implementation for all existing T: Float types.Thanks to @cuviper, @Enet4, @fabianschuiki, @svartalf, and @yoanlcq for their contributions!
No prior release notes were kept. Thanks all the same to the many contributors that have made this crate what it is!