feat!: available forecast window is now dependent the current hour (instead of 12 AM of the day) and implemented quantitive changes to let the OpenMeteo forecast and reflect the amount of sunshine that passes through clouds

[lmarkmann]

What changed

Request window. forecast_hours=168&past_hours=24&timeformat=unixtime replaces forecast_days=7. The old window started at local midnight, so an evening launch had already spent most of its first day and could scrub six days ahead; the new one is 24 hours back and 168 forward from now, and --at -6h lands inside it. Timestamps are unix seconds, so the local-string parse and its Decode path are gone.

Sun through cloud. The analytic sky's weight was (1 - cover), which switched the physical model off under any cloud, a cirrus veil included. It is now the model's direct_normal_irradiance_instant over Meinel's clear-sky beam at that elevation, and the same number is Sun.strength, which scales the disc and its glow. This closes the "known compromise in the analytic blend" note.

Turbidity from aerosol optical depth. New weather::aerosol fetches CAMS AOD at 550 nm beside the forecast (scoped thread, same latency); Atmosphere::from_readings maps it through Preetham's beta relation and falls back to visibility when the request fails or past the five-day horizon. Visibility still drives the near-ground haze.

Precipitation. rain, showers and snowfall are read separately: sleet draws both renderers, freezing rain draws rain, any showers make the low deck cumulonimbus. precipitation_probability scales what is drawn (floor 0.3). Storm intensity is cape / 2500 when CAPE is present.

Fog. Code 45/48, or humidity >= 97% with visibility under 1 km, becomes a full-frame veil (the fog draft's haze, darkened at night), pulls the gradient toward a new Palette::Fog by daylight, and hides stars and meteors.

README credits Open-Meteo (CC BY 4.0) and GeoNames, as their licence asks.

Not in this change

minutely_15=lightning_potential is regional and needs a second time axis. No --model flag and no previous-runs request; forecast confidence is expressed through precipitation probability only.

Release

Breaking on the public API (HourlyArrays.time and the daily timestamps are i64, HourlyArrays and Sun gain fields), so this is 0.8.0. cargo-semver-checks reports the added fields; the type change it cannot see is named in the BREAKING CHANGE trailer. Goldens are unchanged: no scene carries an analytic sky and Sun.strength defaults to 1.0, so the render is bit-identical there.

Verification

just ci and just policy green locally; 210 tests. Live: --plain now, --at -6h, --at -23h, --at +7d, --at +167h, --frame all return the expected sky.

[codspeed]

Merging this PR will improve performance by 52.95%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
✅ 71 untouched benchmarks
⏩ 10 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	at_far	212 µs	73.9 µs	×2.9
⚡	Simulation	at_near	82.5 µs	73.1 µs	+12.91%
⚡	Simulation	timeline_168h	5.1 ms	4.6 ms	+10.52%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing weather-intake (6aba832) with main (f546234)}

Footnotes

1. 10 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩