Is it valid for this to compile:
---
ushort a(ushort b) pure nothrow
	{ return b<<10|b; }
---

And for this to not compile:
---
ushort a(ushort b) pure nothrow
	{ return b<<10; }
---
?

Mon, 06 Apr 2009 08:53:41 -0400, Kagamin wrote:

> Is it valid for this to compile:
> ---
> ushort a(ushort b) pure nothrow
> 	{ return b<<10|b; }
> ---
> 
> And for this to not compile:
> ---
> ushort a(ushort b) pure nothrow
> 	{ return b<<10; }
> ---
> ?

They both compile with 2.026.  What's your problem?

Sergey Gromov Wrote:

> Mon, 06 Apr 2009 08:53:41 -0400, Kagamin wrote:
> 
> > Is it valid for this to compile:
> > ---
> > ushort a(ushort b) pure nothrow
> > 	{ return b<<10|b; }
> > ---
> > 
> > And for this to not compile:
> > ---
> > ushort a(ushort b) pure nothrow
> > 	{ return b<<10; }
> > ---
> > ?
> 
> They both compile with 2.026.  What's your problem?

Could you please check bug http://d.puremagic.com/issues/show_bug.cgi?id=2809 with your dmd 2.026? It can be regression.

Wed, 08 Apr 2009 12:06:48 -0400, Kagamin wrote:

> Sergey Gromov Wrote:
> 
>> Mon, 06 Apr 2009 08:53:41 -0400, Kagamin wrote:
>> 
>>> Is it valid for this to compile:
>>> ---
>>> ushort a(ushort b) pure nothrow
>>> 	{ return b<<10|b; }
>>> ---
>>> 
>>> And for this to not compile:
>>> ---
>>> ushort a(ushort b) pure nothrow
>>> 	{ return b<<10; }
>>> ---
>>> ?
>> 
>> They both compile with 2.026.  What's your problem?
> 
> Could you please check bug http://d.puremagic.com/issues/show_bug.cgi?id=2809 with your dmd 2.026? It can be regression.

Yes the #2809 is present in 2.026.

Actually I didn't realize your example was meant to be compiled with -w switch.  Without -w it simply compiles, no problem.  With -w, it fails while I think it shouldn't.

On Mon, Apr 6, 2009 at 8:53 AM, Kagamin <spam@here.lot> wrote:
> Is it valid for this to compile:
> ---
> ushort a(ushort b) pure nothrow
>        { return b<<10|b; }
> ---
>
> And for this to not compile:
> ---
> ushort a(ushort b) pure nothrow
>        { return b<<10; }
> ---
> ?

There was a terribly long conversation about this and other operations here:

http://d.puremagic.com/issues/show_bug.cgi?id=1977

Basically the idea behind the warning on the left-shift is that you can't know, at compile-time, whether the shift will overflow the size of ushort or not.  If you passed in 0xFFFF, for instance, it would overflow.  So it converts left-shifts to int and complains if you don't have an explicit cast.

But for many bitwise operators, such as | and &, there is no risk of an overflow at runtime, so if your function returned "b & 0x3F00", you wouldn't get such a warning.

That it accepts "b << 10 | b" but rejects "b << 10", however, looks more like a bug.  It's like the compiler isn't doing enough work to find out whether the former can overflow or not.

Jarrett Billingsley wrote:
> On Mon, Apr 6, 2009 at 8:53 AM, Kagamin <spam@here.lot> wrote:
>> Is it valid for this to compile:
>> ---
>> ushort a(ushort b) pure nothrow
>>        { return b<<10|b; }
>> ---
>>
>> And for this to not compile:
>> ---
>> ushort a(ushort b) pure nothrow
>>        { return b<<10; }
>> ---
>> ?
> 
> There was a terribly long conversation about this and other operations here:
> 
> http://d.puremagic.com/issues/show_bug.cgi?id=1977
> 
> Basically the idea behind the warning on the left-shift is that you
> can't know, at compile-time, whether the shift will overflow the size
> of ushort or not.  If you passed in 0xFFFF, for instance, it would
> overflow.  So it converts left-shifts to int and complains if you
> don't have an explicit cast.
> 
> But for many bitwise operators, such as | and &, there is no risk of
> an overflow at runtime, so if your function returned "b & 0x3F00", you
> wouldn't get such a warning.
> 
> That it accepts "b << 10 | b" but rejects "b << 10", however, looks
> more like a bug.  It's like the compiler isn't doing enough work to
> find out whether the former can overflow or not.

cast.c line 50.

		/* This is really only a semi-kludge fix,
		 * we really should look at the operands of op
		 * and see if they are narrower types.
		 * For example, b=b|b and b=b|7 and s=b+b should be allowed,
		 * but b=b|i should be an error.
		 */

Sure, put it in bugzilla.