What's this? (Was: DMD 1.028 and 2.012 releases)

[Steven Schveighoffer]

"Aarti_pl" wrote in message
> Did you notice this example in docs?
> (http://www.digitalmars.com/d/2.0/struct.html#AssignOverload)
>
>
> S* opAssign(S s)
> {   S tmp <== *this; // bitcopy *this into tmp
>     *this <== s;     // bitcopy s into *this
>     tmp.~this();     // call destructor on tmp
>     return this;
> }
>
> It seems to be something new in D 2.012. Does it work? How?
>
> BR
> Marcin Kuszczak
> (aarti_pl)

After reading this new feature, I'm really confused.  I like the feature, 
but I'm not sure how it works, or why there are two functions for the same 
thing.

How does one use the postblit, and how does one use the opAssign?  It seems 
they both serve the same function, but opAssign allows more control 
possibly?  And does opAssign get called if the postblit doesn't exist or 
vice versa?

Not to knock what you have done, but it would seem to me more reasonable to 
have a single method, and if you wanted to blit the values before your 
custom code, just do:

S* opAssign(ref const S s)
{
   *this = s;
   ...
}

And this would get called on both copy construction and assignment.  The way 
I see it now (if each are individual operations), blitting won't work if the 
source is const and contains a pointer, so that the following confusing 
situation happens:

struct S
{
   int[] buf
   this(this)
   {
       buf  = buf.dup;
   }

   S* opAssign(ref const S s)
   {
       buf = s.buf.dup;
       return this;
   }
}

const S s;
S t = s; // fails because the blit fails
S u;
u = s; // works because no blit occurred

Also it appears that your "equivalent" statement is wrong, or else the 
opAssign signature is wrong:

t = S.opAssign(s);

Doesn't t need to be a struct pointer?  Or else is opAssign not really 
returning a struct pointer?  Does this work?

S u;
u = t = s;

Finally, I don't understand the reasoning for tmp.  It appears to be an 
unused part of the default opAssign...

-Steve