Re: Interval support for Postgres

On Thu, 21 Apr 2005, Oliver Siegmar wrote:

> Well, actually there is almost no implementation at all. Parsing strings in
> Java in 2005 is a royal pain in the ass - doing that for 0.x% users suck even
> more.

Well, just be thankful the 8.0 release finally dropped JDK 1.1 support.

> Let's assume I'd write a JDBC2 version with old-school string parsing - would
> you apply it to upstream, or are there any other show stoppers?
>

We definitely want something better for PGInterval.  I do have some
questions about the implementation you've got.

1) You have an "int milliseconds" field.  PG supports microsecond
precision in intervals.  Is there any reason not to go with a plain "float
seconds" field instead of splitting these up?

2) getValue() calls nf.format(Math.abs(milliseconds)), if milliseconds is
2 won't the NumberFormat(ter) turn that into .02 ?

3) In general the whole "before" API is confusing.  Do you have any better
ideas on how to handle this?
  a) Note getValue() returns the wrong value if hours == 0 and minutes < 0.
  b) Doesn't clone() need to copy the before setting?
  c) After a setValue() if someone does setYear() don't they need to
     know the before setting (which is protected).


Kris Jurka