time: print more than just "0" for a zero duration (revert?)

[robpike]

http://play.golang.org/p/pIagpaYGak

package main

import "fmt"
import "time"

func main() {
    var t time.Duration
    t = 1e3
    fmt.Printf("%s\n", t)
    t -= 1e3
    fmt.Printf("%s\n", t)
    t -= 1e3
    fmt.Printf("%s\n", t)
    t = 1e6
    fmt.Printf("%s\n", t)
    t -= 1e6
    fmt.Printf("%s\n", t)
    t -= 1e6
    fmt.Printf("%s\n", t)
    t = 1e9
    fmt.Printf("%s\n", t)
    t -= 1e9
    fmt.Printf("%s\n", t)
    t -= 1e9
    fmt.Printf("%s\n", t)
}

prints

1µs
0
-1µs
1ms
0
-1ms
1s
0
-1s

It should show either ns because that's the internal resolution or s because that's the SI unit.

[distributed]

(Somehow Google Groups ate my message, so here I go again)

+1. Personally I favor s for two reasons: 1.) As an engineer if in doubt I lean towards SI units and 2.) As a user of software written in Go I don't know about the internal resolution; I feel that s as a unit will be recognized as a duration with higher likelihood.

[tarndt]

It should show either ns because that's the internal resolution or s because that's the SI unit.

My vote is ns because it relays useful information- the precision of the quantity at hand.

Also, the adjacent values on the number line are labeled as ns, I would argue that fact favors zero being labeled similarly: http://play.golang.org/p/nWZN7w-2G9

[kkirsche]

I would vote ns because that's the internal resolution. To me that makes sense as it's the lowest denominator and if a person is working in ns, µs or ms it seems odd to grow in magnitudes of the duration indicator when really it's 0 of the lowest possible value since the closest value would be 1ns not 1s

[mdempsky]

In general, time.Duration.String does not use its internal resolution to indicate significant figures. For example, it formats time.Second as "1s", not as "1.000000000s" or "1000000000ns". Expecting the zero time.Duration's formatting to reflect its precision seems arbitrary and inconsistent.

Also, the godocs are explicit that the reason time.Nanosecond is formatted as "1ns" is to avoid the leading zero in "0.000000001s". There will be leading zeros in either "0s" and "0ns", so that does not seem like a valid argument to favor "0ns" either.

Lastly, time.Minute + 1 is formatted as "1m0.000000001s", not as "1m1ns", so it seems in general we favor formatting time.Duration values in units of seconds, not nanoseconds.

[rsc]

I suspect this should be 0s. Any SI prefix is distracting. There's no guarantee that it really is 0 to nanosecond precision. That's up to the use case.

I understand why 0 had no suffix here, but I agree that it's a bit too irregular for its own good. It reminds me of ANSI C printf %#x formatting (only) 0 without the 0x prefix, which has caused no end of formatting and parsing surprises. (Go did not adopt this bug.)

[gopherbot]

CL https://golang.org/cl/22357 mentions this issue.

[rsc]

I'm seeing a lot of things break as a result of this change. I'm not convinced it is worth the pain to our users. Even if we can technically do it, it burns some goodwill for not much benefit, since programs will still need to think about older versions that returned "0". Thoughts?

[bradfitz]

I'm in favor of reverting. Similar to locking down testing/quick and continuing to support lax URL parsing, I think we should just accept the old "0" output and not try to fix past mistakes.

[robpike]

The only breaks I've seen are things like golden files for usage messages, which are both unimportant and easy to fix.

I believe that the previous behavior may have been deliberate but it was certainly a bug and should be fixed.

[minux]

Reverting SGTM. We could argue that the unit doesn't matter for zeros.