Instant::toISOString() is not RFC 3339 compliant

[JanTvrdik]

Instant::toISOString() / ZonedDateTime::toISOString returns value that is not RFC 3339 compliant, because it will omit seconds, when its value is zero.

This is causing portability issues, for example, such value is not accepted by java.time.Instant.parse.

Java sandbox

RFC 3339 format description

[BenMorel]

Good point, thanks for reporting it @JanTvrdik!

I'm surprised to see that the behaviour is different for LocalDateTime and LocalTime: sandbox.

This is probably the source of confusion when I first implemented ZonedDateTime::__toString(), where I just concatenated the output of the LocalDateTime with the timezone.

I'll work on a fix, however this will target version 0.8.0 as this is technically a BC break.

What about parsing, should we force seconds as well? I'm leaning towards yes.

[JanTvrdik]

What about parsing, should we force seconds as well?

Java allows missing seconds when parsing ZonedDateTime, LocalDateTime and LocalTime. The missing seconds are allowed by ISO 8601. You also have the problem, that you would be unable to parse string that were generated by brick/date-time < 0.8. So overall I think it's better to keep the parser unchanged.