Array of interfaces static initialization

[Steven Schveighoffer]

"Yossarian" wrote
> Hello.
> I've got following array:
>
> typedef iObject[] paramList;
>
> and following object architecture:
>
> abstract class iObject { ... };
> class associativeArray : iObject { ... };
> class block : iObject { ... };
> class cIrc   : block { ... };
> class command : associativeArray { ... };
>
> and have function
> Run(paramList);
>
> ..
>
> when I use (z.B.)
>
> command c;
> cIrc this;
>
> Run([c, this]);
>
> irc.d(278): Error: cannot implicitly convert expression (this) of type 
> ghhd.irc.
> cIrc to ghhd.irc.command
>
> both command and irc could be easily upcast to iObject,
> so I think that this is absolutely legal usage (bug ?), or I've badly 
> understood the array initializer?
>

Not so badly :)  Basically, the expression [c, this] is ambiguous to the 
compiler.  So it makes a decision and says whatever the type of the first 
element is the type of the array.  So it types this as command[2].  Then it 
can't implicitly convert cIrc to command automatically so it pukes.

Janice's solution tells the compiler how to declare the literal (as an 
iObject[])

What you don't realize is that the D language is structured to have 
context-free parsing, which means just because [c, this] is used as an 
argument to Run, which takes iObject[], doesn't matter.  The compiler looks 
at [c, this] without context and tries to determine the type just from those 
tokens.  This makes the compiler implementation much simpler, but makes the 
compiler less 'intuitive.'

-Steve