fix unnecessary rounding to float64 precision when JSON-marshaling durations

[dfawley]

No description provided.

[dfawley]

Ping - this should be pretty trivial.

[dsnet]

It seems Herbie is OOO.

[awalterschulze]

Could we add a test for a negative duration?
I think this might result in -2277.-100626394s for example.

[puellanivis]

“For durations of one second or more, a non-zero value for the nanos field must be of the same sign as the seconds field. Must be from -999,999,999 to +999,999,999 inclusive.” https://developers.google.com/protocol-buffers/docs/reference/google.protobuf#google.protobuf.Duration

It looks like you’re right, @awalterschulze . And, the negative duration test is a good idea in all cases. Additionally, a negative less-than-one-second test is also going to be important as this would currently write as “0.-100626394s” for example.

It looks like we will need to extract the sign and absolute value of both of the integers… some quick and dirty code (not intended as complete solution, I just think in code)…

sign := ""
if s < 0 {
  s = -s
  sign = "-"
}
if ns < 0 {
  ns = -ns
  sign = "-"
}
x := fmt.Sprintf("%s%d.%09ds", sign, s, ns)

Even though it’s more work, in the end, this should still work better (no power-of-ten rounding issues) and be faster than using floating points… (?)

[awalterschulze]

I was just trying to merge this change into gogoprotobuf and ran my tests, which found the bug. I am really glad to hear its not my tests, but an actual bug. Looking forward to merge a fix when its ready.

[dfawley]

This is my fault -- I'll send a PR to fix this in a few minutes.

[dfawley]

#491