SNIP
>You are right; BCD can be either packed or unpacked. I had quite forgotten about the details, since I never used them in actual practice.
>
>So, for calculations that are 100% accurate (at least for plus and minus), you have to use BCD - it doesn't matter so much whether it is packed or not.
>
But is it not the case that BCD is not available generally, at least on the Intel platform?

On IBM mainframes, and I believe in their AS/400 series too, "packed decimal" is fully integrated into the INSTRUCTION SET in the hardware. In other words, there are specific instructions to ADD PACKED, SUBTRACT PACKED, etc.

cheers