"Kris" <fu@bar.com> wrote in message news:d6lfa6$1j0r$1@digitaldaemon.com...
> Right; but isn't that trying to salvage a use-case that's already broken?
I
> mean, you're working with a invalid null instance at that point - yes?

Casting a null class reference to an interface should produce a null, not a small offset value.

"Kris" <fu@bar.com> wrote in message news:d6lfpg$1jiu$1@digitaldaemon.com...
> Shall I test it against Mango?

I want to look at the package problem first.

"Walter" <newshound@digitalmars.com> wrote in message news:d6lg1n$1jrk$1@digitaldaemon.com...
>
> "Kris" <fu@bar.com> wrote in message
news:d6lfa6$1j0r$1@digitaldaemon.com...
> > Right; but isn't that trying to salvage a use-case that's already
broken?
> I
> > mean, you're working with a invalid null instance at that point - yes?
>
> Casting a null class reference to an interface should produce a null, not
a
> small offset value.

Arguably so. But then what should one do with -release? This can lead to quite a bit of bloat, and the extra jumps don't help any with pipeline bubbles in todays CPUs ... seems a shame to really penalize the usage of interfaces in this manner, especially with code that (a) has no null object references, or (b) does explicit checks for them in the most appropriate places, such as DbC; (c) the library function used for casting should do this check, but inline code? with full optimization enabled? We have to wonder ...

The codegen used to be a whole lot better in these cases.

For example, here's the codegen distinction between where CR is an object and where CR is an interface ~ CR implements an interface in both cases, which is expected by the called method:

2449:         Stdout.put (CR);
0041ECD3   push        dword ptr
[_D5mango2io6Writer2CRC5mango2io5model7IWriter14INewlineWriter (00550e38)]
0041ECD9   mov
eax,[_D5mango2io6Stdout6StdoutC5mango2io6Stdout13ConsoleWriter (005512b0)]
0041ECDE   mov         ecx,dword ptr [eax]
0041ECE0   call        dword ptr [ecx+38h]


2449:         Stdout.put (CR);
0041F093   cmp         dword ptr
[_D5mango2io6Writer2CRC5mango2io6Writer13NewlineWriter (00551e38)],0
0041F09A   je          _D8unittest14testTextWriterFZv+1Ah (0041f0a6)
0041F09C   mov
eax,[_D5mango2io6Writer2CRC5mango2io6Writer13NewlineWriter (00551e38)]
0041F0A1   lea         ecx,[eax+10h]
0041F0A4   jmp         _D8unittest14testTextWriterFZv+1Ch (0041f0a8)
0041F0A6   xor         ecx,ecx
0041F0A8   push        ecx
0041F0A9   mov
eax,[_D5mango2io6Stdout6StdoutC5mango2io6Stdout13ConsoleWriter (005522b0)]
0041F0AE   mov         edx,dword ptr [eax]
0041F0B0   call        dword ptr [edx+38h]


These used to be *almost* identical. Should people avoid using interfaces, since their usage inccurs two wholly unecessary jumps for every function call that accepts one?