OK for compiler to not call function with constant return?

[Aaron Lawrence]

Hi all.

Suppose I have

bool CFoo::AMethod()
{
// do lots of side effects...
...
return false; // the only return
}

and later I have:

bool result = AMethod();

Is it OK for the compiler to skip the call to AMethod and just assign
the value false to result? Even if all optimisations are off?

Thanks

Aaron

 

[Amal P]

Hi,
It wont assign false directly. It will only happen if the function
is inline.

Best regards,
Amal P.

 

[Ben Pope]

Hi all.

Suppose I have

bool CFoo::AMethod()
{
// do lots of side effects...
...
return false; // the only return
}

and later I have:

bool result = AMethod();

Is it OK for the compiler to skip the call to AMethod and just assign
the value false to result? Even if all optimisations are off?

Not in general, it depends what you mean by side-effects.

Can you post complete, compilable code that demonstrates the problem?

Ben Pope

 

[Richard Herring]

Hi all.

Suppose I have

bool CFoo::AMethod()
{
// do lots of side effects...
...
return false; // the only return
}

and later I have:

bool result = AMethod();

Is it OK for the compiler to skip the call to AMethod and just assign
the value false to result? Even if all optimisations are off?

Not if the side-effects affect program state. The compiler is allowed to
eliminate some calls to copy constructors, but otherwise the as-if rule
means that if the function has detectable side-effects, they must take
place.

Can you post a short complete program that demonstrates your actual
problem?

 

[Aaron Lawrence]

Not if the side-effects affect program state. The compiler is allowed to
eliminate some calls to copy constructors, but otherwise the as-if rule
means that if the function has detectable side-effects, they must take
place.

That's what I thought. It seemed like a clear case of the optimisation
breaking my reasonable code because I had accidentally returned false in
all cases. Suddenly my function was no longer called.

Can you post a short complete program that demonstrates your actual
problem?

Should be possible. Tomorrow ...

Cheers

Aaron