Hello D-experts,

I created a little test program which IMHO shows
not the expected behaviour.
Is this a bug or not ?

module testshift;

import std.stdio;

void main()
{
  uint val_a = (1 << 32);

  uint shift = 32;
  uint val_b = (1 << shift);

  writefln("Result: ", val_a, "  ", val_b);
  assert(val_a == val_b);
}

Output is as follows:
Result: 0  1
Error: AssertError Failure testshift(14)

KlausO wrote:
> Hello D-experts,
> 
> I created a little test program which IMHO shows
> not the expected behaviour.
> Is this a bug or not ?
> 
> module testshift;
> 
> import std.stdio;
> 
> void main()
> {
>   uint val_a = (1 << 32);
> 
>   uint shift = 32;
>   uint val_b = (1 << shift);
> 
>   writefln("Result: ", val_a, "  ", val_b);
>   assert(val_a == val_b);
> }
> 
> Output is as follows:
> Result: 0  1
> Error: AssertError Failure testshift(14)


The problem here is: what values would you expect?

1<<32 is too big for an uint, so the 0 from the constant folding is correct, BUT the "shl" instructions (the one << translates into) only looks at the lower 5 bits, which are 0 (32 & 0x1F == 0), so also the 1 is correct.

I don't think we'd want extra overhead for something like <<, so I suppose we let "<<" behave like the instruction "shl". Which means there's only one solution: let the compiler complain.

Come to think of it, it should already have complained: 1<<32 does not fit into an uint! But DMD gives no error, not even a warning (-w)...

L.

http://www.digitalmars.com/d/expression.html#ShiftExpression
It's illegal to shift by more bits than the size of the quantity being
shifted.

That means, 32 is legal for shifting. And therefore it should evaluate to the correct result.

So I think the example is a bug.

Frank Benoit wrote:
> http://www.digitalmars.com/d/expression.html#ShiftExpression
> It's illegal to shift by more bits than the size of the quantity being
> shifted.
> 
> That means, 32 is legal for shifting. And therefore it should evaluate
> to the correct result.
> 
> So I think the example is a bug.

I think 32 should be illegal also. It would append 32 0-bits, which means the result (uint) will always be 0.

L.

Lionello Lunesu wrote:
> 
> I think 32 should be illegal also. It would append 32 0-bits, which means the result (uint) will always be 0.
> 

Are there practical uses for shifting more than 32 bits ?

Example: Calculating a mask
  uint mask = (1 << nbits) - 1;

Of course, you could express this also as
  uint mask = (0xffffffff >> (32 - nbits));