[Haskell-cafe] custom SQL-to-Haskell type conversion in HDBC

[Erik Hesselink]

On Fri, Aug 19, 2011 at 16:53, Henry House <hajhouse at hajhouse.org> wrote:
> On Friday, 19 August 2011, Erik Hesselink wrote:
>> Why exactly do you need the precision information?
>
> Empirical measurements (e.g., sizes of some fields in hectares) are
> precise only to a certain level of measurement error. Thus, the area
> measurements 1 ha and 1.000 ha are not equivalent or interchangeable.
> Database engines recognize this fact by providing different data types
> for rational numbers and fixed-precision decimal numbers.
>
> The bottom line for me is that the conversion of a fixed-precision
> decimal number as a rational is both throwing away information (the
> precision) as well as introducing bogus information (the notion that the
> result value has greater --- i.e., infinite --- precision that was in
> fact intended when that value was stored).

Note that PostgreSQL also doesn't work with decimals as precision:

postgres=# select 1::decimal(4,2) * 1::decimal(4,2);
?column?
----------
   1.0000
(1 row)

That should be 1.00 instead if you want the precision correctly represented.

Perhaps a solution would be to not treat the database precision as
your primary source of information, but represent that in Haskell
using some data type that correctly propagates precision information,
and marshall your database data to and from that. This means some
duplication of information (precision in both database and Haskell)
but you do the same with NULL and Maybe, etc. I guess that's inherent
to (the way HDBC does) database access.

Erik