Jim,

The help topic does mention "integer":

To store values that are both dates and times or only dates, use the DateTime data type. A DateTime value is stored in eight bytes — two four-byte integers.

And the datetime expression topics implicitly state it based on the standard naming conventions:

nSeconds = tTime1 - tTime2

Granted n doesn't necessarily imply integer, but it's certainly not float.

Datetime arithmetic does support fractional seconds though:

lt1 = {^2003-10-16 00:00:00}
? lt1 && :00
lt1 = lt1 + 1.5
? lt1 && :02 - display rounds up
lt1 = lt1 + 1.5
? lt1 && :03

lt1 = {^2003-10-16 00:00:00}
? lt1 && :00
lt1 = lt1 + 1.4
? lt1 && :01 display rounds down
lt1 = lt1 + 1.4
? lt1 && :03 display rounds up

>You seem to know that Datetimes are limited to accepting integers, but no one else would KNOW that from the documentation. I simply cannot find such a restriction in the Help and the tiny bit on "Date and time operators" makes no mention of the fact.
>
>Now why would it be unusual for someone using datetimes to simply use the memvar they already have, that might be real or whatever, in a calculation? Nothing, I should say, UNLESS it is documented as problematic. And if it truly is problematic, wouldn't it be deserving of an error message?