The errors raised by the numeric-numeric conversion function in std.conv are inaccurate in two ways:

1. Both underflowing and overflowing conversions raise a ConversionOverflowError, with the following respective messages presented to the user:
    std.conv.ConvOverflowError: Error: overflow Conversion underflow
    std.conv.ConvOverflowError: Error: overflow Conversion overflow

2. Converting a floating-point NaN to an integer type gives an underflow error.


Suggestions for simple fixes:
1. At the very least, the "overflow" string which is added to both over- and underflow errors should be removed from ConvOverflowError.

2. Put the following at the top of the "to" function definition:

    static if (isFloatingPoint!S && !isFloatingPoint!T)
        if (isNaN(value)) ConvOverflowError.raise("Cannot convert "
            ~S.stringof~".nan to "~T.stringof);

(Of course, it is arguable whether the slightly more accurate error message is worth the performance penalty of the additional check...)