If the unit is already known to be invalid, but there is a validation that includes a numericality check we try to instantiate the measurement to do comparisons. But then that would raise.
So just skip the numericality checks if the unit is already known to be invalid. Can't do anything about it anyway.
@cyprusad @garethson @RichardBlair @mdking
Found the bug. Follow up to #20.
If the unit is already known to be invalid, but there is a validation that includes a numericality check we try to instantiate the measurement to do comparisons. But then that would raise.
So just skip the numericality checks if the unit is already known to be invalid. Can't do anything about it anyway.