Hi,
IMHO the compiler should be able to implicitly convert function-pointers into delegate-pointers.
Before the delegate is passed as argument, the context-pointer is filled in.
Now when I pass a function, the context-pointer could simply be filled with null.
A problem will be, where to store the context-pointer in a function, as it has no context-pointer (to preserve compatibility with C).

Or the compiler could implicitly create an anonymous delegate around that function.

Thorsten Kiefer wrote:
> Hi,
> IMHO the compiler should be able to implicitly convert function-pointers into delegate-pointers.
> Before the delegate is passed as argument, the context-pointer is filled in.
> Now when I pass a function, the context-pointer could simply be filled with null.

At the moment it can't, since delegates and function-pointers have incompatible calling conventions.

> A problem will be, where to store the context-pointer in a function, as it has no context-pointer (to preserve compatibility with C).

There's another thread on that, "D 2.0 - dalegates to function"

> Or the compiler could implicitly create an anonymous delegate around that function.

That would be a good transitional solution, but because it involves an extra function call, I'd still hope the delegate and function-pointer calling-convention can be made compatible.

L.

Lionello Lunesu Wrote:

> Thorsten Kiefer wrote:
>> Hi, IMHO the compiler should be able to implicitly convert function-pointers into delegate-pointers.  Before the delegate is passed as argument, the context-pointer is filled in.  Now when I pass a function, the context-pointer could simply be filled with null.
> 
> At the moment it can't, since delegates and function-pointers have incompatible calling conventions.

Well, it certainly used to work under at least some platform/compiler combinations.

I've looked, and it would appear to be a mere matter of whether a context pointer is pushed to the stack as well.  The ABI doesn't make it clear whether parameters/local variables are addressed from the bottom or the top of the stack frame.  From my experiment, it would appear to be the bottom.

<snip>
> That would be a good transitional solution, but because it involves an extra function call, I'd still hope the delegate and function-pointer calling-convention can be made compatible.

Indeed.  And one way that's been suggested is to pass the context pointer in a register rather than pushing it to the stack.

Stewart.