fugit

Added

• The Monotonic instant kind (together with the old Wrapping kind), for timelines that do not wrap. Before, the Instant started to get methods that were related to Monotonic instants - this is now split.
• MonotonicInstant, MonotonicTimerInstant, MonotonicTimerInstantU32 and MonotonicTimerInstantU64 aliases.
• Instant::is_before and Instant::is_after, wrap-aware on Wrapping, a plain tick compare on Monotonic.
• Instant now implements Hash.
• Instant's Debug output names the kind: WrappingInstant { ticks: 7 }.

Fixed

• Issue #84: Instant comparison no longer reports Equal for instants that are not equal.
• Cross-base operations on u32 storage truncated their u64 conversion constants, panicking with a divide-by-zero or returning silently wrong results.

Breaking Changes

• BREAKING: Operations between time bases differing by more than the storage type's maximum are now a compile-time error. Only u32 storage with a base ratio above u32::MAX is affected.
• BREAKING: Issue #83: wrapping instants are no longer Ord, that compare is not transitive. PartialOrd is kept, so <, >, <= and >= are unaffected. Monotonic instants are Ord.
• BREAKING: Issue #84: on wrapping instants const_cmp is replaced by const_partial_cmp, which returns None for instants exactly half the tick range apart. Monotonic instants keep const_cmp, where a total order makes it sound.
• BREAKING: Rename checked_sub_duration to convert_sub_duration and checked_add_duration to convert_add_duration with improved docs.
• BREAKING: Instant takes a fourth type parameter naming its timeline kind. Instant<u32, 1, 1_000> becomes WrappingInstant<u32, 1, 1_000>, or Instant<u32, 1, 1_000, kind::Wrapping>.
• BREAKING: The TimerInstant, TimerInstantU32 and TimerInstantU64 aliases are replaced by WrappingTimerInstant, WrappingTimerInstantU32 and WrappingTimerInstantU64.
• BREAKING: duration_since_epoch is removed from wrapping instants, where a fixed epoch is meaningless. Use Duration::from_ticks(i.as_ticks()) for the old ticks-to-Duration shorthand.

Added

• Duration::ZERO and Duration::MAX constants
• Duration::checked_mul, checked_div, div_ceil methods
• Duration::saturating_add, saturating_sub, saturating_mul methods
• Duration::from_secs_f32 and from_secs_f64 constructors with rounding
• Duration::as_secs_f32 and as_secs_f64 conversions
• Support for picosecond-level precision (enabled by u64 const generics)
• picos() shorthand method to ExtU32, ExtU64, ExtU32Ceil, and ExtU64Ceil traits
• Duration::from_picos, as_picos, from_picos_at_least methods
• PicosDuration<T>, PicosDurationU32, PicosDurationU64 aliases
• Gigahertz<T>, GigahertzU32, GigahertzU64 aliases (the defmt/Display impls already formatted this base as GHz)
• core::ops::Rem and core::ops::RemAssign trait implementations for Rate and Duration (fixes #41)
• core::ops::SubAssign for Rate (was missing; asymmetric with Add/AddAssign/Sub)
• Conversion to/from core::time::Duration

Fixed

• Issue #50: Rate conversions now use half-up rounding instead of truncating for better accuracy
• Issue #53: Overflow detection in shorthand conversions (e.g., NanosDurationU32.minutes())
• Added #[track_caller] to all panicking functions and operator impls (including *Assign wrappers, Mul/Div/Rem on integers, and the cross-type u32<->u64 ops) so panic locations point at the user's call site.
• Documentation clarifications across Instant::checked_*_duration, Instant::checked_duration_since, Duration::checked_* / Rate::checked_*, the try_from_*/try_to_* conversions, and the const_partial_cmp/const_eq methods. Several methods previously claimed to "check for overflow" or "check for divide-by-zero" while the actual contract was different (e.g. wrapping ticks, returning None on cross-base conversion overflow).
• Duration::const_try_from/const_try_into/convert (and the same on Rate) no longer overflow u64 silently when ticks are close to u64::MAX. The rounding step (lh + d/2) / d was replaced with an overflow-free divmod-based form that produces the same rounded result without ever wrapping.
• Duration::Hz/kHz/MHz, Rate::Hz/kHz/MHz, Rate::to_Hz/to_kHz/to_MHz, Duration::try_from_rate, and Rate::try_from_duration now produce a compile-time error when the conversion constants don't fit the storage type, instead of silently truncating u64 constants on a u32 cast.
• The Duration shorthand methods (as_*, from_*, from_*_at_least) and the Rate::Hz/kHz/MHz/to_Hz/to_kHz/to_MHz family now panic on multiplication overflow instead of silently wrapping in release builds. All gained #[track_caller] so the panic points at the user's call site.

Changed

• Modernized CI
• Updated gcd dependency to 2.3
• Updated defmt dependency to 1.0

Breaking Changes

• BREAKING: Duration::const_try_from now rounds-to-nearest, before it was truncating
• BREAKING: Standardized API naming conventions
  • Constructors now use from_* prefix (e.g., from_ticks, from_rate)
  • Getters now use to_* or as_* prefix (e.g., to_rate, as_ticks)
• BREAKING: Changed NOM and DENOM const generic parameters from u32 to u64 across all types (Duration, Rate, Instant)
  • This enables support for higher precision time units (e.g., picoseconds with denominator 1_000_000_000_000)
  • Type aliases with FREQ_HZ parameters now use u64 instead of u32
  • Compile-time calculations use u128 intermediates with overflow checks, but results are guaranteed to fit in u64 for embedded compatibility

Added

• Optional serde support
• Optional postcard max size support

Fixed

Changed

Added

• Add Div implementation for Duration and Rate
• Add nanos() methods and NanosDuration aliases alongside other units.
• Implement AddAssign and SubAssign for Instant and Duration, and MulAssign and DivAssign for Duration.
• Add to_nanos(), to_micros(), to_millis(), to_secs(), to_minutes(), and to_hours() methods to Duration<u32, NOM, DENOM> and Duration<u64, NOM, DENOM> types to easily convert to integer time units.
• Support for Rate (Hertz, Kilohertz, ...)

Fixed

Changed