no exceptions are thrown in math operations

[Fabian Wein]

Hi,

I had one case, where gcc 4.1.0 on an opteron gave 0.0 in case of an
division by 0. I wrote an simple example where I get now inf.
When i do a square-root of a negative number via pow() I get nan.

Now my question is, is there a way that in such cases exceptions are
thrown? At least for the function call this would be nice!

Thanks a lot,

Fabian

double x = 62.5e-6;
double t = x / 0.0;
std::cout << x << " divided by 0.0 = " << t << std::endl;
// -> 6.25e-05 divided by 0.0 = inf

double p = pow(-0.0003, 0.5);
std::cout << "pow = " << p << std::endl;
// -> pow = nan

 

[Alf P. Steinbach]

* Fabian Wein:

I had one case, where gcc 4.1.0 on an opteron gave 0.0 in case of an
division by 0. I wrote an simple example where I get now inf.
When i do a square-root of a negative number via pow() I get nan.

Now my question is, is there a way that in such cases exceptions are
thrown? At least for the function call this would be nice!

In practice it's likely that your C++ compiler supports throwing
floating point operations, but this is not supported by the standard.

If you make floating point operations throw, there is a good chance
you'll have to deal with asynchronous non-C++ operations, and
performance may suffer.

The C library mechanism is the 'signal' function, still in the C++
library, but since a signal handler is invoked asynchronously there are
severe restrictions on what it can do without possibly screwing up
things: in a C++ program setting a flag is in practice the most.

 

[Fabian Wein]

In practice it's likely that your C++ compiler supports throwing
floating point operations, but this is not supported by the standard.

Thanks for the answer! I found no switch up to now for gcc but I know
more now

Cool would be such a feature for valgrind

Fabian