Convert units of measurement

This looks ugly:

  with uu(unit, coefficient, u_ord) as (
    select
     unit, 
     coefficient,
     case 
      when log(u.coefficient) < 0 
      then floor (log(u.coefficient)) 
      else ceil(log(u.coefficient)) 
     end u_ord
    from
     unit_conversion u 
  ),
  norm (label, norm_qty) as (
   select
    s.label,
    sum( uc.coefficient * s.quantity ) AS norm_qty
  from
    unit_conversion uc,
    substance s
  where
    uc.unit = s.unit
  group by
    s.label
  ),
  norm_ord (label, norm_qty, log, ord) as (
   select 
    label,
    norm_qty, 
    log(t.norm_qty) as log,
    case 
     when log(t.norm_qty) < 0 
     then floor(log(t.norm_qty)) 
     else ceil(log(t.norm_qty)) 
    end ord
   from norm t
  )
  select
   norm_ord.label,
   norm_ord.norm_qty,
   norm_ord.norm_qty / uu.coefficient val,
   uu.unit
  from 
   norm_ord,
   uu where uu.u_ord = 
     (select max(uu.u_ord) 
      from uu 
      where mod(norm_ord.norm_qty , uu.coefficient) = 0);

but seems to do the trick:

|   LABEL | NORM_QTY | VAL |       UNIT |
-----------------------------------------
| mercury |   1.5e-8 |  15 | microlitre |
|   water |  0.00112 | 112 | centilitre |

You don't really need the parent-child relationship in the unit_conversion table, because the units in the same family are naturally related to each other by the order of coefficient, as long as you have the family identified.

I think, this can be largely simplified.

1. Modify unit_conversion table

Or, if you cannot modify the table, just add the column exp10 for "exponent base 10", which coincides with number of digits to shift in the decimal system:

CREATE TABLE unit_conversion(
   unit text PRIMARY KEY
  ,exp10 int
);

INSERT INTO unit_conversion VALUES
     ('microlitre', 0)
    ,('millilitre', 3)
    ,('centilitre', 4)
    ,('litre',      6)
    ,('hectolitre', 8)
    ,('kilolitre',  9)
    ,('megalitre',  12)
    ,('decilitre',  5);

2. Write function

to calculate the number of positions to shift left or right:

CREATE OR REPLACE FUNCTION f_shift_comma(n numeric)
  RETURNS int LANGUAGE SQL IMMUTABLE AS
$$
SELECT CASE WHEN ($1 % 1) = 0 THEN                    -- no fractional digits
          CASE WHEN ($1 % 10) = 0 THEN 0              -- no trailing 0, don't shift
          ELSE length(rtrim(trunc($1, 0)::text, '0')) -- trunc() because numeric can be 1.0
                   - length(trunc($1, 0)::text)       -- trailing 0, shift right .. negative
          END
       ELSE                                           -- fractional digits
          length(rtrim(($1 % 1)::text, '0')) - 2      -- shift left .. positive
       END
$$;

3. Query

SELECT DISTINCT ON (substance_id)
       s.substance_id, s.label, s.quantity, s.unit
      ,COALESCE(s.quantity * 10^(u1.exp10 - u2.exp10)::numeric
              , s.quantity)::float8 AS norm_quantity
      ,COALESCE(u2.unit, s.unit) AS norm_unit
FROM   substance s 
JOIN   unit_conversion u1 USING (unit)
LEFT   JOIN unit_conversion u2 ON f_shift_comma(s.quantity) <> 0
                              AND @(u2.exp10 - (u1.exp10 - f_shift_comma(s.quantity))) < 2
                              -- since maximum gap between exp10 in unit table = 3
                              -- adapt to ceil(to max_gap / 2) if you have bigger gaps
ORDER  BY s.substance_id
     , @(u2.exp10 - (u1.exp10 - f_shift_comma(s.quantity))) -- closest unit first
     , u2.exp10    -- smaller unit first to avoid point for ties.

Explain:

• JOIN substance and unit tables.
• Calculate ideal number of positions to shift with function f_shift_comma() from above.
• LEFT JOIN to the unit table a second time to find units close to the optimum.
• Pick the closest unit with DISTINCT ON () and ORDER BY.
• If no better unit is found, fall back to what we had with COALESCE().
• This should cover all corner cases and be pretty fast.

-> SQLfiddle demo.