Hi Gregory,

Sorry, Gregory, but you think i write this test without read limits.

If you read my title is literal double..., not internal double....

My problem is with routine that convert a literal.
A correct language put a limit to the max number of digits used on conversion
( VFP not put this limit and go in fault).

<snip>

>... Hence, log10(2^52) yields 15.65 digits of precision.

This is incorrect, because any mantissa+exponent rapresentation have the first bit of mantissa set to
1, but it is not stored on the memory.
Hence log10(2^(52+1)) -> 15.95

Infact the max integer value that VFP can use in exact computations is 9007199254740992 (2^53)
( read documentation ).

>I would go for 15 to be on the safe side.

Read next messages for a answer.

Fabio