SignedDuration's Display should use upper case letter

[tisonkun]

It seems in the definition of ISO 8601 (RFC-3339), it wrotes: https://datatracker.ietf.org/doc/html/rfc3339#appendix-A

Durations:

   dur-second        = 1*DIGIT "S"
   dur-minute        = 1*DIGIT "M" [dur-second]
   dur-hour          = 1*DIGIT "H" [dur-minute]
   dur-time          = "T" (dur-hour / dur-minute / dur-second)
   dur-day           = 1*DIGIT "D"
   dur-week          = 1*DIGIT "W"
   dur-month         = 1*DIGIT "M" [dur-day]
   dur-year          = 1*DIGIT "Y" [dur-month]
   dur-date          = (dur-day / dur-month / dur-year) [dur-time]

Note that the unit letter here is upper case, while jiff's Display prints lower case.

It would fail when using format("{}", dur) to cast the repr to a PostgreSQL's Interval: https://www.postgresql.org/docs/current/datatype-datetime.html#DATATYPE-INTERVAL-INPUT

... where only upper case letters are accepted:

scopedb_meta=# SELECT INTERVAL 'PT10m';
ERROR:  invalid input syntax for type interval: "PT10m"
LINE 1: SELECT INTERVAL 'PT10m';
                        ^
scopedb_meta=# SELECT INTERVAL 'PT10M';
 interval 
----------
 00:10:00
(1 row)

I'm not sure if it's intended or if somewhere there only accepts lower case letters. But the described PG interval integration is a failed case I encountered.

[BurntSushi]

This was intentional behavior. And it was brought up in #22.

My main issue was that the uppercase variant is one that I find hideously difficult to read. And I find the lowercase variant somewhat less hideous to read since the letters and numbers are easier (for my eyes) to tease apart. And since it was the only way Jiff had to convert a duration (whether a Span or a SignedDuration) to a string, I wanted it to be easier to read.

I think I also under-estimated just how much more broadly supported uppercase unit labels are.

Now, technically speaking, the behavior was correct according to Jiff's documented behavior. Namely, the jiff::fmt::temporal docs specifically point out that Temporal's ISO 8601 specification is followed. This is not the same as ISO 8601, but an extension to it. And I guess it is not strictly a subset, since I cannot find any text in ISO 8601-1 or ISO 8601-2 that permits lowercase unit designator labels. But Temporal's ISO 8601 specification clearly does.

With that said, now that Jiff as a "friendly" format that's easy to access (e.g., format!("{dur:#}")) and easy to read, I feel more comfortable leaning more heavily on prioritizing interoperability. So in #191, I've switched the default to using uppercase unit designator labels. But I did add a jiff::fmt::temporal::SpanPrinter::lowercase knob to enable the previous behavior.

Why does Temporal support lowercase if ISO 8601 doesn't? I wasn't fully sure where that came from, so I dug.

• This seems to be the commit switching support from just uppercase to both uppercase and lowercase: tc39/proposal-temporal@941ba81
• Here's the PR containing that commit: Properly spec ISO 8601 parsing tc39/proposal-temporal#404
• Here is the first mention of using upper and lowercase unit designator labels: Point to ISO standard for Date string syntax tc39/proposal-temporal#198 (comment)
• And this seems to confirm it: Point to ISO standard for Date string syntax tc39/proposal-temporal#198 (comment)

But I can't find anything about where exactly that comes from, other than perhaps "good sense." (Which I wholeheartedly agree with.)

I've otherwise looked through both ISO 8601-1 and ISO 8601-2 and I can't find anything that says the unit designator labels in a duration may be lowercase. Unless I'm missing something in the standard, my guess is that the lowercase alternative evolved organically over time and enjoys spotty support.

[tisonkun]

Thanks for the quick fix!

I agree that systems like PostgreSQL should accept lowercase somehow, but reporting or contributing there becomes more difficult :\

[tisonkun]

And FWIW, yes the friendly format is accepted by PG and I can use format!("{:#}", dur) to work it around on 0.1.18.

[BurntSushi]

This should now be fixed in jiff 0.1.19 on crates.io.