when 32 bit ms counter folded, led may be ON by mistake? how to avoid that? about candlelight_fw

[fenugrec]

Thanks. You are right, there is a bug there.
I think a better way would be to cast the subtraction result to a signed type,
if ((int32_t) (now - led->off_until) >= 0)

a bit similar to what you did but IMO a bit cleaner.
but also need to guarantee that off_until is never set to more than INT32_MAX in the future (25 days), which should be straightforward to enforce.

[daniel-santos]

Yes, wrap around happens. This is a common occurrence in the Linux kernel and even must be considered for transient time values spanning years -- a Linux machine might be up for that long! In such cases, it's better to just think of the counter/timer value as meaningless until compared (via subtraction) to a target time/count and, as @fenugrec illustrated, treat them as signed values -- only worry about the difference, and you'll stay out of trouble.

Also, signed math is far fewer instructions. You just have to change the way you are thinking about your timers and time values.

[fenugrec]

yes. Will fix when I get a chance; leaving this open as a reminder

[fenugrec]

I cobbled up a fix (no time to test on hardware yet but compiles) in PR #78 , if anyone would like to review / test

[fenugrec]

Updated the fix in #78, and tested on hardware. Looks fine, but feedback would be appreciated.