Re: ao record convert overflow

["Schoeneburg, Bernd" <bernd.schoeneburg@desy.de>]

Hi Andrew,
I think the part "value >= 0" works.
The add/substract of 0.5 before or after checking is a question of 
taste. In my proposal, 0.5 would be needed only once and after the "if" 
only the cast is to do.
The part "value < 0":
The test "value > (0.5 - 0x80000000)" means "value > (0.5 - 
(-2147483648)" == "value > +2147483648.5" if 0x80000000 is interpreted 
as 32-Bit long constant. If long is 64 bit on this system, the test is 
"value > (-2147483647.5)", which is OK.
If you write "value > (0.5 - 0x80000000LU)" this should work always.

Why did you change the type of ROFF? Since value is double when ROFF is 
substracted it doesn't cause problems even if ROFF is <0, I think.
What do you think of defining constants like "epicsInt32_MAX" and 
"..MIN" in epicsTypes.h?
Bernd