www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Division by zero - why no exception?

reply A Bothe <info alexanderbothe.com> writes:
Hey everyone,
I noticed something:

double a=0.0;
double f=1/a;
writeln(f);

... prints 2.66825e-284

normally it shall throw a exception or do something else ... but it doesn't ...
why?
Jan 10 2010
parent reply bearophile < bearophileHUGS lycos.com> writes:
A Bothe:

 double a=0.0;
 double f=1/a;
 writeln(f);
 ... prints 2.66825e-284


To me this code prints "nan" (I am using the latest D2 compiler, but D1 too
prints nan):

import std.stdio: writeln;
void main() {
    double a = 0.0;
    double f = 1 / a;
    writeln(f);
}

If you want the division by zero error you have to activate it (but be careful,
this works only in the main(), probably to keep tidy the semantics of pure
functions):

import std.stdio: writeln;
import std.math: FloatingPointControl;

void main() {
    FloatingPointControl fpc;
    fpc.enableExceptions(FloatingPointControl.severeExceptions);

    double a = 0.0;
    double f = 1 / a;
    writeln(f);
}

Prints at run time:
object.Error: Float Divide by Zero

But be careful, that works at run time only, so if you use compile time
constants:

import std.stdio: writeln;
import std.math: FloatingPointControl;

void main() {
    FloatingPointControl fpc;
    fpc.enableExceptions(FloatingPointControl.severeExceptions);

    enum double a = 0.0;
    enum double f = 1 / a;
    writeln(f);
}

It prints nan again. Eventually the behaviour of D floating point operations at
compile time will need to become much more similar to their run-time behaviour,
to simplify language semantics and to have more "deterministic" results. You
can see a similar problem here, some floating point circus, that I've recently
shown in D.learn:

import std.math: sqrt, PI;
import std.stdio: writefln;

void main() {
  const double x1 = 3.0 * PI / 16.0;
  writefln("%.17f", sqrt(1.0 / x1));

  double x2 = 3.0 * PI / 16.0;
  writefln("%.17f", sqrt(1.0 / x2));

  real x3 = 3.0 * PI / 16.0;
  writefln("%.17f", sqrt(1.0 / x3));

  real x4 = 3.0L * PI / 16.0L;
  writefln("%.17f", sqrt(1.0L / x4));
}

Output with various D compilers:

DMD1:
1.30294003174111994
1.30294003174111972
1.30294003174111979
1.30294003174111979

DMD2:
1.30294003174111972
1.30294003174111972
1.30294003174111979
1.30294003174111979

LDC:
1.30294003174111994
1.30294003174111994
1.30294003174111972
1.30294003174111972

Bye,
bearophile
Jan 10 2010
parent reply bearophile <bearophileHUGS lycos.com> writes:
 (but be careful, this works only in the main(), probably to keep tidy the
semantics of pure functions):


That was not precise, the FPC struct is restored at the end of the main()
scope, so if you call a function inside main it will keep the same FPC struct,
and this raises the division by zero error:

import std.stdio: writeln;
import std.math: FloatingPointControl;

void foo() {
    double a = 0.0;
    double f = 1 / a;
    writeln(f);
}

void main() {
    FloatingPointControl fpc;
    fpc.enableExceptions(FloatingPointControl.severeExceptions);
    foo();
}

Also, you can't catch it:

void foo() {
    double f = 0;
    try {
        double a = 0.0;
        f = 1 / a; // object.Error: Float Divide by Zero
    } catch (Error e) {
    }
    writeln(f);
}

Bye,
bearophile
Jan 10 2010
parent A Bothe <info alexanderbothe.com> writes:
OK, thanks!
Jan 10 2010