time: Parse of RFC3339 is incorrect for fractional second seperator

[dsnet]

RFC 3339, section 5.6 specifies that the time-secfrac field must use a period.

However, Parse permits the use of commas:

Parse(time.RFC3339, "0000-01-01T00:00:00,0Z")

reports nil rather than an error.

[dsnet]

Unfortunately, this behavior is specified in the docs:

A comma or decimal point followed by one or more zeros represents a fractional second,
printed to the given number of decimal places.
A comma or decimal point followed by one or more nines represents a fractional second,
printed to the given number of decimal places, with trailing zeros removed.
For example "15:04:05,000" or "15:04:05.000" formats or parses with millisecond precision.

[seankhliao]

That is a relatively recent addition #6189, at some point in the past, we didn't support it #24860
Maybe it would be ok to roll it back for RFC3339

[dsnet]

Yuck. In my opinion #6189 was a mistake. We probably should instead have allowed the format string to have a , where it means either , or ., while keeping the presence of . to only mean ..

[gopherbot]

Change https://go.dev/cl/425045 mentions this issue: time: fix Parse regression of sub-second separator in RFC3339

[gopherbot]

Change https://go.dev/cl/444277 mentions this issue: time: implement strict RFC 3339 during marshal and unmarshal

[dsnet]

Closing as the allowance of commas was a deliberate decision in #6189.

The Unmarshal methods have been modified to perform strict checking. That change is more justifiable since the methods are explicitly documented as using RFC 3339 and the Marshal methods already have some checks for strict RFC 3339 compliance.