conversion of integer values in number literals gives warning for large magnitudes

[paciorek]

 f <- nFunction(
      fun = function(x='numericScalar') {
          z <- 1e300
          returnType(double(0))
          return(z)
      })

  cf <- nCompile(f)

Warning message:
In compile_generateCpp(code$args[[2]], symTab) :
  probable complete loss of accuracy in modulus

This occurs because value %% 1 doesn't handle large magnitudes in compile_generateCpp.

I will work out a fix.

[paciorek]

Integers 2^53 and lower can be represented exactly in a double. After 2^53, some integers cannot be represented exactly and are rounded to the nearest representable integer. And every number larger than 2^53 is represented as an integer-valued number.

I thought that was going to correspond to when value %% 1 started to give the warning, but the warning appears around the value 1e19, whereas 2^53 ~= 1e16.

In any event, I think the solution here is do the formatting either if we exceed 2^53 or if %%1 is 0:

else if (identical(code$type$type, 'double') &&
                 (value > 9007199254740992  ||  identical(value %% 1, 0)))
          ## If value is a whole number we need to add "." or ".0" to the end
          ## or use scientific notation so C++ knows it's actually a double.
          format(x = value, nsmall = 1)