result of (int)NaN ?

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

Tom - 14 Jun 2005 00:51 GMT

I have the following instruction in C++.NET :

x = log10(n);
int y = (int)(x * h);

where x, h, n are of type double, and x receives the value NaN
(because the value of number is rather small).

This results in y having value of the largest negative 32-bit integer.

What does the standard specify in this instance?
Should an exception be thrown?
Will the sign of Y always be known?

Right now the workaround is to test the value of n, to make sure that it is
within the range of the log10 function else force a value to y.

-- Tom

Reply to this Message

Mike Welham - 14 Jun 2005 17:22 GMT

Hi Tom,

> I have the following instruction in C++.NET :
>
[quoted text clipped - 3 lines]
> where x, h, n are of type double, and x receives the value NaN
> (because the value of number is rather small).

The conversion from double to int will be compiled to a 'conv.i4' IL
instruction.

> This results in y having value of the largest negative 32-bit integer.
>
> What does the standard specify in this instance?
> Should an exception be thrown?
> Will the sign of Y always be known?

The ECMA spec (partition III, section 3.27) has this to say about conv.i4:

- "Conversion from floating-point numbers to integral values truncates
the number toward zero."

- "If overflow occurs converting a floating-point type to an integer the
value returned is unspecified."

- "No exceptions are ever thrown."

> Right now the workaround is to test the value of n, to make sure that it is
> within the range of the log10 function else force a value to y.

There is a 'conv.ovf.i4' IL instruction that will throw an
OverflowException if the conversion fails.

In C# you can ask the compiler for this behaviour by doing:

int y = checked((int)(x * h));

I don't know how to use checked arithmetic and conversion operations in
managed C++, perhaps somebody else has that answer?

Best Regards

Mike

Reply to this Message

Tom - 15 Jun 2005 01:15 GMT

Hi, Mike.

Could not find the 'checked' modifier in C++.NET, but
I did find the method: static bool Double::IsNAN(double)
so I can test for that. Thanks for your help.

-- Tom

> Hi Tom,
>
[quoted text clipped - 42 lines]
>
> Mike

Reply to this Message

Rate this thread:

result of (int)NaN ?

Free Magazines