float_eq

Changed

• Clearer documentation for the FloatEqDebugUlpsDiff and FloatEqUlpsTol traits.
• Updated derive tests to match current compiler output.
• Switched to using github actions and codecov.io for continuous integration and coverage.

Bumped up the version number since this release includes breaking API changes.

Changed

• Renamed many ambiguous uses of epsilon/eps/max_diff to tolerance/tol, which is the mathematical term of art.
• Implementation of traits for arrays of any size using const generics.
• Updated num-complex dependency to 0.4.
• Reworked the documentation, added a tutorial.

Fixed

• Safer behaviour when initialising debug output for array trait impls, including the removal of a source of undefined behaviour.

Bumped up the version number since this release includes breaking API changes to the extension traits.

Changed

• Renamed FloatEqDebug to AssertFloatEq, and similarly FloatEqAllDebug to AssertFloatEqAll. This is to provide a more obvious link between these traits and the assert capability. The FloatDiff trait has been folded into AssertFloatEq both for clarity and because it is geared towards debug context and was not as general as the name implied.
• Expanded rel checks into rmax (aka rel), rmin, r1st and r2nd to select which operand to scale epsilon's granularity to.
• Significant rewrite of the documentation to clarify aspects of comparison and to have a slightly more streamlined structure with better defined sections.
• Updated num-complex dependency to 0.3.

Added

• The derive_float_eq helper macro to simplify deriving traits.

Fixed

• Update README usage documentation to the correct version, correctly documented the "num" feature.

Bumped up the version number since this release includes breaking API changes.

Added

• #[derive] support for all traits on tuple structs and structs with named fields.
• The FloatUlps trait, which is a way to syntactically tie a floating point type to its ULPs representation. The prefered way to express this is now to implement FloatUlps for T and then use Ulps<T> for the ULPs type. The other traits have been reworked to reflect this.
• Blanket trait impls for comparing slices, Option, Vec, VecDeque, LinkedList, BTreeMap and HashMap.

Changed

• Blanket array impls now allow for comparison of arrays of different types.
• FloatEqAll::Epsilon is now AllEpsilon and FloatEqAllDebug::DebugEpsilon is now AllDebugEpsilon to help avoid ambiguity.

Added

• Blanket trait impls for comparing mutable and immutable reference types and the contents of Cell, RefCell, Rc, Arc and Box instances.
• More descriptive documentation for combining types of check.

Bumped up the version number since this release includes breaking API changes.

Added

• The FloatEqAll and FloatEqAllDebug traits, which allow for abs_all, rel_all and ulps_all to be used in the float_eq and assert_float_eq families of macros, which take a single epsilon value to be applied uniformly across all fields being compared. The behaviour of FloatEq has now changed to apply epsilon values structurally, on a per-field basis, since that is a more general behaviour (e.g. all tuples may sensibly be FloatEq but not FloatEqAll). If existing checks against arrays or num::Complex break, switching to use the _all variants ought to fix them.
• Support for tuples of up to size 12 (inclusive).
• Documentation on how to interpret assert error messages.

Changed

• ULPs checks now treat NaN values as not equal, to match the default partial equality behaviour of floats.
• Equality of infinities is now consistent, both internally and with respect to general IEEE floating point behaviour.
• FloatEq now more specifically means equality based on a structurally defined epsilon type. See the notes on FloatEqAll above.
• FloatDiff is much more rigorously defined, in particular ulps_diff now returns Option<T>, see the API documentation for details.
• FloatEq's DiffEpsilon is now Epsilon, and UlpsDiffEpsilon is now UlpsTol. This reduces visual noise around the usage and harmonizes the naming with FloatEqDebug's associated types.
• Directed docs.rs to build documentation for all features.

Bumped up the version number since this release includes breaking API changes.

Added

• Implementation of traits for arrays of size 0 to 32 (inclusive) where the type allows it. Epsilon is assumed to be uniform across the array being compared.
• The 'num' feature, which when enabled provides support for comparison of num::Complex instances.
• Documentation to help with implementing the traits.

Changed

• FloatDiff, FloatEq and FloatEqDebug along with the macros that use them now allow for a different Rhs type to be specified instead of assuming it is always Self.
• The somewhat awkward FloatEq::rel_epsilon was removed in favour of a more equitable FloatEqDebug trait for displaying debug information.
• Asserts now correctly dereference parameters.
• Somewhat more streamlined assert error messages that allow for easier direct comparison of diffs and epsilons.
• Switched back to being 'experimental' on crates.io since the API might change a lot in the next few versions.

Added

• Added codecov.io and coveralls.io support.
• Added 'actively-developed' maintenance badge.
• Basic usage documentation.

Added

• Support for no_std.
• Added the crate to the No Standard Library crates.io category.
• Tests for Float::abs_diff on NaNs.
• Travis CI build.
• A Contributing section to the readme.

Fixed

• FloatDiff tests for f64 no longer test f32 values (oops!).

Added

• Added this changelog.
• Reworked the READMEs to be a bit more granular and provide more appropriate information.
  • The API documentation lib.rs has been pared down a bit and several sections moved to other places.
  • crates-io.md contains a short project introduction, links to alternative crates and future plans.
  • README.md is generated from crates-io.md and LICENSE.md.
• Updated Cargo.toml:
  • Included a link to the documentation.
  • Added the crate to the Development Tools (Debugging) category.

Added

• Initial release.