# Changelog All notable repository-release changes are tracked here. The main [`README.md`](./README.md) stays focused on the current baseline and entry points; older release history lives in this file. ## Unreleased ## 0.5.1 - 2026-10-13 ### Changed - Builds with the latest MoonBit toolchain (moon 0.1.20261012, moonc 0.10.15): `Kaida-Amethyst/math` is raised from 0.1.20 to 0.1.21 (0.1.20 ships the legacy `moon.mod.json`, which the new toolchain rejects for dependencies) and the sources follow the formatter of that toolchain. - Documented that, except for host `Sqrt`, built-in `Float` and `Double` elementary functions use `Kaida-Amethyst/math` 0.1.20 without a correct-rounding, published error-bound, or cross-backend bit-identity guarantee; replacement work is tracked by [#13](https://github.com/Luna-Flow/arithmetic/issues/13). - **Breaking behavior clarification for 0.5.1:** for checked `Float` and `Double`, NaN divided by zero changes from `Err(DivisionByZero)` to `Ok(NaN)`, and infinity divided by signed zero changes to the correctly signed `Ok(±inf)`. Power results are unchanged: quiet-NaN zero exponents remain `Ok(1)` and other non-zero NaN exponents remain `Ok(NaN)`, as recorded by IEEE 754-2019 §9.2.1. This tier handles all NaN values as quiet because `Float` and `Double` do not distinguish signaling NaN. - Negative-exponent underflow and reciprocal correctness remain tracked by [#1](https://github.com/Luna-Flow/arithmetic/issues/1); this PR does not change those cases. - The test-only dependency `Luna-Flow/luna-generic` is bumped from 0.3.1 to 0.4.0. The tests use only `Zero`, `AddMonoid` and `Num`, so no code change is needed and nothing deprecated in 0.4.0 is used. - Migrated the package to MoonBit 0.10. Trait methods that were implicitly promoted to methods are now promoted explicitly in `src/extends.mbt`: `equal` stays available on `FpClass`, `RoundingMode`, `ArithmeticContext`, `ArithmeticDiagnostics`, `ArithmeticOutcome`, `ArithmeticError`, `ArithmeticErrorKind`, `CertificationStage`, `CertificationFailureReason` and `CertificationFailureDetail`, and now appears in the interface file. - The package imports `moonbitlang/core/debug` for the `Debug` promotions. - Blackbox tests call the package through `@arithmetic`, as MoonBit 0.10 requires. - Documentation rewritten (API, tutorial and design pages for `core`, plus the overview and guides), with complete zh_CN and ja_JP translations. Every runnable example compiles and passes as a test. - Manual brought to the Luna Flow documentation standard: the overview has install, page, export and validation sections; the API page gains purpose and importing sections; the tutorial gains a task table; the design page states its constraints and main decisions. The manual now documents the laws of the integer `Power` instances (and how a wrapped exponent sum breaks them), the limits of Kleene's three-valued logic for correlated conditions, and that `ArithmeticError` and the context types implement neither `Show` nor `Debug`. ### Deprecated - The implicit method forms `x.not_equal(y)` (all public types above) and `x.to_repr()` (`FpClass`, `ArithmeticDiagnostics`, `ArithmeticOutcome`) remain callable but are deprecated and hidden from the documentation. Use `x != y` and `Repr(x)` instead. ## 0.5.0 - 2026-07-18 Latest published release. ### Added - Added `IntegralContextual` for context-dependent integer embedding with diagnostics. - Added `AdjacentContextual` for the next representable value toward positive infinity, negative infinity, or a target value. - Added `ConstantsContextual` for context-dependent `pi`, `tau`, and `e`, with structured certification failure available to proof-backed implementations. - Added `HyperbolicContextual` for contextual `sinh`, `cosh`, and `tanh` outcomes. - Added `Float` and `Double` adapters for integer embedding and fixed-format adjacent values. - Added edge-case coverage for integer conversion, signed zero, NaN, infinity, maximum finite values, and minimum subnormal values. ### Changed - Standardized the Luna Flow test import alias on `lf_alg`. - Synchronized the English, Simplified Chinese, and Japanese API, design, verification, getting-started, and release documentation. ### Compatibility - The four contextual traits are additive public APIs; existing signatures are unchanged. - `IntegralContextual` deliberately accepts MoonBit `Int`. Wider integer sources can be added through separate capabilities without changing this contract. - `Float` and `Double` intentionally do not implement `ConstantsContextual` or `HyperbolicContextual`; those capabilities require a backend that can honor the supplied context and report meaningful diagnostics or certification failures. ### Current Limits - Native `Float` and `Double` adapters retain their fixed IEEE binary formats rather than emulating arbitrary decimal contexts. - `Float` integer embedding reports detectable conversion loss. Native adjacent selection is exact in the fixed representable set and returns empty diagnostics. ## 0.4.0 - 2026-07-15 Previous repository release. ### Added - Added `CertificationStage` and `CertificationFailureReason` to distinguish where and why proof-backed numerical evaluation could not certify a result. - Added `CertificationFailureDetail`, retaining the operation, target and working precision, and refinement count for callers that need to decide whether to retry with more resources or report the failure. - Added `ArithmeticErrorKind::CertificationFailure` and corresponding `ArithmeticError` construction, inspection, and predicate APIs. - Added the localized getting-started, architecture, and verification guides, with synchronized English, Simplified Chinese, and Japanese counterparts. - Added CI coverage for all-target builds/checks, JavaScript and native tests, public-interface snapshots, and formatting. ### Changed - Standardized the manual publish workflow on the release contract used by `mare_mark`: an explicit version input must match `moon.mod`, verification precedes publication, and a GitHub release is created after publication. - Added the standard `Debug` derivation to `FpClass`. ### Compatibility - `CertificationFailure` is an additive `ArithmeticErrorKind` variant. Existing callers should include a fallback branch when matching error kinds so future categories remain representable. - The new detail types are a capability boundary; built-in `Float` and `Double` operations do not currently produce certification failures. ## 0.3.0 - 2026-07-12 Previous repository release. ### Added - Added optional exponent limits and clamping to `ArithmeticContext`, together with decimal32, decimal64, and decimal128 context presets. - Added `ArithmeticDiagnostics` for immutable `inexact`, `rounded`, `overflow`, `underflow`, `subnormal`, and `clamped` status flags, including explicit flag combination. - Added `ArithmeticOutcome[T]` to keep successful values and diagnostic flags in one result. - Added contextual capabilities for addition, subtraction, multiplication, division, absolute value, square root, exponential, and numeric-format facts. - Added `Float` and `Double` implementations for the contextual capability surface. - Added a repository changelog and aligned the English, Chinese, and Japanese documentation with the current API surface and documentation standard. ### Changed - Expanded `ArithmeticContext::new` with optional `e_min`, `e_max`, and `clamp` arguments while retaining source compatibility for existing constructor calls. - Regenerated the public package interface for the new context fields, result types, contextual traits, and built-in implementations. ### Current Limits - Built-in `Float` and `Double` contextual operations preserve native scalar behavior and return exact diagnostics for successful operations. - These built-in instances do not apply arbitrary decimal precision, directed rounding, exponent clamping, or full IEEE status-flag detection. - Contextual division and square root reuse the existing checked validation paths before wrapping successful values. ## 0.2.2 - 2026-07-07 Previous repository release. ### Highlights - Checked `Float` and `Double` square root began reporting `DomainError` for negative real inputs while preserving negative zero and NaN behavior. - Checked division distinguished indeterminate `0 / 0` and infinity divided by infinity from ordinary non-zero division by zero. - Checked signed powers handled the minimum `Int` exponent without directly negating it. - Package source configuration moved to the current `source = "src"` syntax. ## 0.2.1 - 2026-06-07 Previous repository release. ### Highlights - Refined the checked capability release boundary for downstream Luna Flow numeric packages. - Documented that downstream floating implementations should reuse the shared arithmetic context and error types rather than defining ecosystem-local replacements. - Aligned repository metadata with the Apache-2.0 license baseline. ## 0.2.0 - 2026-06-07 Previous repository release. This release established the checked arithmetic capability boundary. ### Highlights - Added `ArithmeticContext`, `ArithmeticError`, error categories, floating-point classification, and rounding-mode vocabulary. - Added checked square root, division, comparison, integer-power, and parsing traits. - Added enclosure-style containment and definite or possible relation traits. - Added checked `Float` and `Double` implementations and expanded trait-level test coverage. ## 0.1.0 - 2026-06-06 Initial repository baseline. ### Highlights - Added small elementary capability traits for radicals, exponentials, logarithms, powers, trigonometric functions, hyperbolic functions, and constants. - Added default `Float` and `Double` implementations backed by `Kaida-Amethyst/math`. - Added exact integer-family implementations for operations that remain closed on their concrete types.