typesafe time units: proc sleep(t: Duration), allowing sleep(1.second), sleep(1.millisecond) etc

[timotheecour]

as noted in nim-lang#17149 (comment), sleep is confusing:

• posix.sleep takes seconds
• os.sleep takes milliseconds

this is confusing, not self documenting, error prone and not particularly typesafe

proposal

add proc sleep(t: Duration)
example:

sleep(1.second)
sleep(1.millisecond)
sleep(100.millisecond + 1.second)
sleep(1.5.hour)

note

std/times is a heavy dependency, ideally we'd be able to do this without depending on std/times, refs nim-lang/RFCs#55, via a new std/durations module (imported by std/times)

note

TimeInterval is not good, this has overhead because it's calendar based
we can define millisecond, second (etc) without s to avoid confusion with the existing milliseconds + friends.
eg:

proc second*(a: int | float): Duration
proc millisecond*(a: int | float): Duration
# etc

[ringabout]

Now, we could have

sleep(1'second)
sleep(1'millisecond)
sleep(100'millisecond + 1'second)
sleep(1.5'hour)

Consider following APIs

sleepFor

proc sleepFor(t: Duration)

The thread blocks for at least specified duration.

sleepUntil

proc sleepUntil(t: DateTime)

The thread blocks until specified time arrived.

[timotheecour]

Now, we could have

yes.

sleepFor, sleepUntil

maybe sleepMin, sleepMax is less ambiguous? but also, can you explain the semantics of those 2 procs, and in which case you'd end up sleep more than sleepFor or less than sleepUntil?

also: std/durations should be separate from std/times and be imported/re-exported from std/times
(because std/times is a monster already and durations can be selfcontained and small); related to nim-lang/RFCs#55

[timotheecour]

closing, as I'm opening the corresponding RFC nim-lang/RFCs#383 (because I'm seeing PRs that introduce duration params as int)