Forum OpenACS Q&A: Re: Tcl weird behaviour with math

Posted by Benjamin Brink on

First, check the value of tcl_precision to verify it is zero. See about half way down page.

Second, this looks like a standard binary math issue that affects most any computer language. See:

When making an application that calculates currency, caution should be used to round off to the lowest denomination at appropriate places to avoid this kind of error. Commonly, a smallest value of 0.01 is assumed (cents in USD for example), but some currencies have a smallest unit of 0.05 for example.

Avoid coding that exacerbates binary errors in calculation, such as using ceiling before completing a rounding error. This is also true if using a currency value for a logical comparison with the value of 0, such as:

if { expr { 2241.57 *100 - 224157 } == 0 } { ... }

as there will be faulty logic. Round to the currency's lowest transferable unit first. Or, in the case of a logical comparison of zero, one can see if the error in comparison is less than the smallest transferable unit of currency; for example, see ec_same_value in ecommerce/tcl at end of


Posted by Antonio Pisano on
Thanks for your fast and clear explanation Benjamin, I will update my procs so they truncate all digits beyond required precision before being ceiled and this should do the trick.

All the best


Posted by Gustaf Neumann on
The best is to calculate internally the only the smallest unit (i.e. cents) in integers, calculate with these, and format these as required to dollar, euro, etc.

In the more general case, you might want to look into the tcl decimal arithmetic package:

In some simple cases, it might be sufficient to work with a comparison operator with an espilon fuzz value (e.g. withinEpsilon is

best regards
-gustaf neumann

Posted by Brian Fenton on
Also, always brace your expressions!