Add Sleep.seconds!

[dmisiuk]

Adds Sleep.seconds!, closes #409.

Discussed on Zulip — @Anton-4 preferred F64 over U64 so that one function covers whole and fractional durations without having to switch back to millis!.

Implementation

seconds! converts to milliseconds and delegates to the existing Host.sleep_millis!, so there are no host or glue changes.

The conversion is the only subtle part, so it lives in a pure top-level helper, millis_from_seconds, which keeps it unit-testable. F64.round_to_u64 crashes on negative, NaN, infinite, and out-of-range values, which isn't acceptable for a signature with no error channel. F64.to_u64_try rejects exactly that set, so the helper separates the cases by sign: a positive value reaching the Err branch can only be one that overflowed the millisecond range.

Edge-case behavior

This wasn't part of the Zulip discussion, so flagging it explicitly for review:

Input	Behavior
1.5	sleeps 1500 ms — fractional part truncated to whole ms
0.0004	truncates to 0 ms rather than rounding up
0	returns immediately
negative	returns immediately
NaN	returns immediately
inf, or beyond U64.highest ms	saturates at U64.highest ms

Returning immediately on negative and NaN follows Go's time.Sleep. Happy to change either behavior if you'd prefer something different.

Testing

Nine expect blocks in platform/Sleep.roc cover millis_from_seconds across all three branches — the ordinary conversion, the negative/NaN case, and saturation. Each was verified to fail when its expected value is altered, so they aren't vacuously green.

Full ./scripts/test.py passes — host build plus all examples.

The effectful wrapper was also measured against a local build of the platform before being reduced to a one-liner: 1.5 → 1500 ms, 0.25 → 253 ms, 0 → 0 ms, -5 → 0 ms, NaN → 0 ms. The saturating branch can't be exercised at runtime, since it sleeps for roughly 584 million years; it is covered by the expect on millis_from_seconds instead.

examples/time.roc now uses Sleep.seconds!(1) in place of Sleep.millis!(1000) and its # 1000 ms = 1 second comment, which is the ergonomic problem the issue describes. Its one-second-sleep case asserts output shape rather than the duration value, so scripts/test_spec.json is unchanged.

[Anton-4]

Thanks @dmisiuk!