Pricelist computations need to consider 2 different Units
of Measure:
- The default product UoM (product.uom_id), used as reference
for the various quantities and amounts specified in each
pricelist rules.
- The `context UoM` is the UoM in which the result is requested,
that is the list price UoM.
For example the 'price_min_margin' amount is meant for the unit
price of 1 x default UoM. When the context UoM is not the default
product UoM, it can be any UoM of the same UoM Category, and the
various quantities and amounts specified on the rule need to
be adapted accordingly:
- min_quantity (expressed in terms of the default UoM)
- price_surcharge (specified for 1 x default UoM)
- price_min_margin (specified for 1 x default UoM)
- price_max_margin (specified for 1 x default UoM)
The UoM corrections were not done consistently and resulted in
wrong prices when computing the price using a non-default UoM.
The cases were a conversion was needed or not were not properly
identified within the _price_rule_get_multi().
After this commit, the various code branches in _price_rule_get_multi
always ensures that:
- price requested for: `qty` of `qty_uom_id`
- `qty_in_product_uom` is the requested `qty` converted to default UoM
- current (intermediary) price: `price` for `price_uom_id`
Therefore `price` and `price_uom_id` are always in sync, and `price_uom_id`
can always be compared with `qty_uom_id' in order to know whether
a conversion is still needed.
This patch also corrects and extends the regression tests
introduced at revision 79ebe10.
Remove the intermediate rounding inside _compute_qty(), as it
is not necessary after rev. fa2f7b86 and has undesired side-effects.
An extra float_round() operation inside _compute_qty()
had been added at rev. 311c77bb to avoid a float representation
error in UoM factors that could bias the ceiling() operation
done as the last conversion step.
Example 1:
Dozen has a factor of 1/12, which was previously stored in the
database with a decimal accuracy of 12 significant decimal digits.
This meant the factor was exactly stored as 0.08333333333333.
When reading this back into a Python float, the precision was not
sufficient, and the UoM conversion of 1 Dozen to Units gave a
result of 12.00000000000047961...
After the final ceiling() operation to Unit's rounding, the
converted value ended up as 13.
This problem was initially solved using an extra rounding.
However at revision fa2f7b86 the decimal precision used to store
UoM factors was increased to preserve all significant digits.
This added the extra precision necessary to read the Dozen factor
back into an accurate float value of 1/12, and the conversion of
1 Dozen now gives 12.0 Units, even without the intermediate
rounding operation. (Works for other factor values too)
At the same time that extra rounding operation has undesired
side-effects, as it requires a fixed precision derived from
the rounding precisions of the UoMs. But there is no given precision
that would work in all cases for this intermediate value. It is
always possible to find a valid combination of UoM roundings
that breaks that intermediate step, e.g. by forcing integer
roundings.
Example 2:
Let Grams have a rounding precision set to 1 because no smaller
quantities are allowed, and Kilograms a rounding of 0.001 to allow
representing 1 Gram. (gram factor = 1000 and kilogram rounding = .001
by default)
If we try to convert 1234 Grams into Kilograms, the extra rounding
introduced in 311c77bb will cause a rounding of 1234.0/1000.0 at
the precision of Grams (1), which gives 1.0 as a result.
The net result of this conversion gives 1234.0 Gram = 1.0 Kilogram,
while the correct result (1.234 Kilogram) is perfectly compatible
with the UoM settings.
Similar errors could be triggered with various rounding settings, as
long as the intermediate rounding needs a finite precision.
Two extra tests have been added to cover Example 1 and Example 2.
--
Related to #2072, #1125, #1126, #2672Closes#2495, #2498
The price_surcharge attribute must be computed based on the reference unit of measure (divided by the factor).
This is to make sure than 12 units and 1 dozen have the same price after pricelist computation (opw 599727).
Added test checking the correctness of pricelist computation based on unit of measures.
Xavier Morel
Olivier Dony
Martin Trigaux
Martin Trigaux