fix(sensors/darwin): parse sp78 temperatures instead of returning 0.0

[drutsan]

Summary

sensors.TemperaturesWithContext() returns 0.0°C for every sensor on Intel Macs because getTemperature() in sensors_darwin.go explicitly returns 0.0 when it detects the sp78 data type — but sp78 is exactly the format Apple's SMC uses for temperature readings.

Bug

sensors/sensors_darwin.go:108-110:

if result.dataSize == 2 && result.dataType == toUint32(dataTypeSp78) {
    return 0.0
}

This branch is supposed to parse the sp78 reading, not discard it. Looks like an unfinished placeholder.

Fix

Parse the two-byte sp78 response according to spec — high byte = signed integer degrees, low byte = fractional part in 1/256 increments:

if result.dataSize == 2 && result.dataType == toUint32(dataTypeSp78) {
    return float64(int8(result.data[0])) + float64(result.data[1])/256.0
}

Verification

Tested on Mac mini 2018 (Intel Core i5-8500B, T2 chip) running macOS 15.7.5 (Sequoia):

Before:

TC0P: 0.0°C  TM0P: 0.0°C  TW0P: 0.0°C  ... (all 21 keys returned 0.0)

After:

TC0P: 60.3°C  (CPU 0 Proximity)
TM0P: 57.6°C  (Memory Slots Proximity)
TW0P: 49.6°C  (Wireless Module)

Cross-checked against osx-cpu-temp which reports 59.7°C for the same hardware/load — values match.

Other keys (TC0D, TC0H, TG0D, etc.) still legitimately return 0.0 — those sensors aren't physically present on T2-chip Mac mini hardware. The IOKit + SMC stack is fine; only the parsing was broken.

Affected platforms

darwin && !arm64 (Intel Macs). Apple Silicon Macs use a different code path (sensors_darwin_arm64.go) and are not affected.

Test plan

• Build & run on Intel Mac mini 2018 (macOS 15.7.5)
• Verify temperatures match osx-cpu-temp output
• Verify other key types (single-byte readings) still work
• CI to verify cross-compile still succeeds

[shirou]

LGTM. The sp78 decoding is correct — equivalent to interpreting the two bytes as a signed 16-bit big-endian value and dividing by 256, which matches the canonical reading used by the Linux applesmc driver and other established SMC implementations. I have not checked on a real macOS, but verified the formula handles both positive and negative readings (e.g. 0xFE 0x80 → -1.5°C).

Two minor (non-blocking) suggestions:

• The 5-line comment is heavier than the surrounding style; the historical "Previously this branch incorrectly returned 0.0" sentence will rot — git blame already records that. Trimming to a one-liner referencing the format would be enough.
• Extracting the parser into a tiny pure helper (parseSp78(hi, lo uint8) float64) would make it unit-testable on any platform

Would be great to see both addressed in this PR before merge.

Reference implementations consulted:

• https://github.com/torvalds/linux/blob/master/drivers/hwmon/applesmc.c#L820-L836
• https://github.com/aristocratos/btop/blob/main/src/osx/smc.cpp#L78-L81
• https://github.com/beltex/SMCKit