Limit subseconds precision

[fulmicoton]

I am trying to use time to format UTC into rfc3339.
Unfortunately the nanoseconds precision is a bit too high and appears very verbose in the logs.
I would love to either have a way to limit the precision of a format to milliseconds precision,
or at least make it possible to define it using a custom format.

With time's DSL, I don't think I can express "%Y-%m-%dT%H:%M:%S%.3f"

[jhpratt]

format_description!("[year]-[month]-[day]T[hour]:[minute]:[second].[subsecond digits:3]")

I'm assuming you need something like RFC3339, not RFC3339 itself, given that you mentioned the possibility of custom formats.

[fulmicoton]

rfc3339 does not mention anything about the precision of time-secfrac though.

[jhpratt]

It indicates one or more digits. When parsing this is done by consuming all digits (even if those after the ninth are functionally ignored). When formatting the logical thing to do is to produce as many digits as possible.

[fulmicoton]

Understood. I'll rely on the format_description macro then. Thank you!

In case this comes up again, my use case is logging with tokio-tracing, and 9 digits precision is hurting the readability of the logs.

[andrewhickman]

Would you be open to making this configurable through an enum like https://docs.rs/chrono/latest/chrono/enum.SecondsFormat.html?

Perhaps we could add a new well-known time format like Rfc3339WithOptions { ... }?

My own use case is implementing serialization of the protobuf google.protobuf.Timestamp type. The spec requires emitting 3, 6 or 9 fractional digits depending on the value.

[jhpratt]

The thing is, "RFC3339 with options" is not RFC3339, so even including the name is misleading. It's already possible to set the number of digits, as was mentioned earlier in this discussion. SubsecondDigits exists if you need to do something manually.