hi

From my point of view, the following piece of code does really strange thing:

write("{}", cast(void*)0xFFFF_FFFF_FFFF_FFFF_UL); // prints ffffffffffff, OK
write(cast(ulong*)0xFFFF_FFFF_FFFF_FFFF_UL); // prints ffffffffffff, OK
write("{}", cast(ulong*)cast(void*)0xFFFF_FFFF_FFFF_FFFF_UL); // prints ffffffff,  not OK

Its not a question of output routine, I've checked it in different envirovments and with different values. It simply cuts of the higher 32 bit part of the address. Could, please, somebody explain me that behavior? For me, it looks like a compiler bug, but that seems impossible for such an obvious thing...

my system:
AMD64, Ubuntu Linux,
gdc (GCC) 4.1.2 20070214 ( gdc 0.24, using dmd 1.024)
(according to gdc --version)

Simon Buerger wrote:
> hi
> 
> From my point of view, the following piece of code does really strange thing:
> 
> write("{}", cast(void*)0xFFFF_FFFF_FFFF_FFFF_UL); // prints
> ffffffffffff, OK
> write(cast(ulong*)0xFFFF_FFFF_FFFF_FFFF_UL); // prints ffffffffffff, OK
> write("{}", cast(ulong*)cast(void*)0xFFFF_FFFF_FFFF_FFFF_UL); // prints
> ffffffff,  not OK
> 
> Its not a question of output routine, I've checked it in different envirovments and with different values. It simply cuts of the higher 32 bit part of the address. Could, please, somebody explain me that behavior? For me, it looks like a compiler bug, but that seems impossible for such an obvious thing...
> 
> my system:
> AMD64, Ubuntu Linux,
> gdc (GCC) 4.1.2 20070214 ( gdc 0.24, using dmd 1.024)
> (according to gdc --version)

It's a bug with the integer conversion routines in DMD's frontend.

I quote expression.c: IntegerExp::toInteger
	    case Tpointer:
	    case Tdchar:
	    case Tuns32:	value = (d_uns32) value;	break;

The following (SVN) patch fixes it:

Index: expression.c ===================================================================
--- expression.c        (Revision 231)
+++ expression.c        (Arbeitskopie)
@@ -1128,10 +1128,15 @@
            case Twchar:
            case Tuns16:        value = (d_uns16) value;        break;
            case Tint32:        value = (d_int32) value;        break;
+#ifndef __x86_64__
            case Tpointer:
+#endif
            case Tdchar:
            case Tuns32:        value = (d_uns32) value;        break;
            case Tint64:        value = (d_int64) value;        break;
+#ifdef __x86_64__
+           case Tpointer:
+#endif
            case Tuns64:        value = (d_uns64) value;        break;

            case Tenum:

downs wrote:
> It's a bug with the integer conversion routines in DMD's frontend.

Yeah, that will fix it, thanks a lot. I hope this will be in svn soon, cuz I think it could lead to really hard-to-find errors.