Re: Date-Time dangling unit fix

[ I removed Lockhart, because he's taken no part in Postgres work for
more than twenty years; if that address even still works, you're
just bugging him ]

Alexander Lakhin <exclusion(at)gmail(dot)com> writes:
> In fact,
> SELECT time 'h04mm05s06';
> doesn't work for many years, but
> SELECT time 'h04mm05s06.0';
> still does.

I traced that down to this in DecodeTimeOnly:

if ((fmask & DTK_TIME_M) != DTK_TIME_M)
return DTERR_BAD_FORMAT;

where we have

#define DTK_ALL_SECS_M (DTK_M(SECOND) | DTK_M(MILLISECOND) | DTK_M(MICROSECOND))
#define DTK_TIME_M (DTK_M(HOUR) | DTK_M(MINUTE) | DTK_ALL_SECS_M)

So in other words, this test insists on seeing hour, minute, second,
*and* fractional-second fields. That seems obviously too picky.
It might not matter if we rip out this syntax, but I see other similar
tests so I suspect some of them will still be reachable.

Personally I'd say that hh:mm is a plenty complete enough time, and
whether you write seconds is optional, let alone fractional seconds.
We do accept this:

=> select '12:34'::time;
time
----------
12:34:00
(1 row)

so that must be going through a different code path, which I didn't
try to identify yet.

regards, tom lane