Yeah, what the subject says.

I want to have a default delegate for a struct, and without a default
constructor, this has to be a compile-time constant. Now, logically,
there should be nothing wrong with storing the function pointer and a
null context pointer at compile-time, but it seems there is. Any ideas?

struct foo {
    void delegate( ) dg = () {}; // Error: non-constant expression
                                 // __dgliteral1
}

-- 
Simen

On 19.07.2010 21:06, Simen kjaeraas wrote:
> Yeah, what the subject says.
>
> I want to have a default delegate for a struct, and without a default
> constructor, this has to be a compile-time constant. Now, logically,
> there should be nothing wrong with storing the function pointer and a
> null context pointer at compile-time, but it seems there is. Any ideas?
>
> struct foo {
>       void delegate( ) dg = () {}; // Error: non-constant expression
>                                    // __dgliteral1
> }
>

I wasn't able to make it work.  The compiler probably sees delegates as something that just can't be created at compile time, since no runtime contexts exist then.  Which is reasonable.

Maybe one of those templates that turn functions into delegates will work?  Otherwise I guess you're back to using a factory function for initializing instances.

Maybe just checking for null pointers before calling those delegates ends up being the easiest solution.

On Mon, Jul 19, 2010 at 22:01, torhu <no@spam.invalid> wrote:

>
>
> I wasn't able to make it work.


Me too :(


> The compiler probably sees delegates as something that just can't be created at compile time, since no runtime contexts exist then.  Which is reasonable.
>

Can you initialize pointers in general, at compile-time?



> Maybe one of those templates that turn functions into delegates will work?
>


I had the same idea and tried to use std.functional.toDelegate, but to no avail.

enum moo = ()  {return 1;};

struct foo {
   int delegate( ) dg = toDelegate(moo);
}

Error: forward reference to inferred return type of function call
toDelegate(delegate int()|

int moo() { return 1;}

struct foo {
   int delegate( ) dg = toDelegate(&moo);
}

Error: forward reference to inferred return type of function call
toDelegate((& moo))|

'auto' strikes again :(
I came to hate these forward reference errors. That and the fact that auto
functions do not appear in the docs, that's enough for me to avoid auto as
much as possible for functions.



> Otherwise I guess you're back to using a factory function for initializing instances.
>
> Maybe just checking for null pointers before calling those delegates ends up being the easiest solution.
>


Philippe

Philippe Sigaud wrote:
> On Mon, Jul 19, 2010 at 22:01, torhu <no@spam.invalid> wrote:
> 
> 
> 
>     I wasn't able to make it work.  
> 
> 
> Me too :(
>  
> 
>     The compiler probably sees delegates as something that just can't be
>     created at compile time, since no runtime contexts exist then.
>      Which is reasonable.
> 
> 
> Can you initialize pointers in general, at compile-time?

You cannot initialize a pointer to runtime-allocated data at compile-time, and will never be able to. This particular example cannot ever work.
You should be able to initialize pointers to static data at compile-time, but currently you can't.

> 
>  
> 
>     Maybe one of those templates that turn functions into delegates will
>     work?  
> 
> 
> I had the same idea and tried to use std.functional.toDelegate, but to no avail.
> 
> enum moo = ()  {return 1;};
> 
> struct foo {
>    int delegate( ) dg = toDelegate(moo);
> }
> 
> Error: forward reference to inferred return type of function call toDelegate(delegate int()|
> 
> int moo() { return 1;}
> 
> struct foo {
>    int delegate( ) dg = toDelegate(&moo);
> }
> 
> Error: forward reference to inferred return type of function call toDelegate((& moo))|
> 
> 'auto' strikes again :(
> I came to hate these forward reference errors. That and the fact that auto functions do not appear in the docs, that's enough for me to avoid auto as much as possible for functions.
> 
>  
> 
>     Otherwise I guess you're back to using a factory function for
>     initializing instances.
> 
>     Maybe just checking for null pointers before calling those delegates
>     ends up being the easiest solution.
> 
> 
> 
> Philippe
>  
> 

Don Wrote:

> Philippe Sigaud wrote:
> > On Mon, Jul 19, 2010 at 22:01, torhu <no@spam.invalid> wrote:
> > 
> > 
> > 
> >     I wasn't able to make it work.
> > 
> > 
> > Me too :(
> > 
> > 
> >     The compiler probably sees delegates as something that just can't be
> >     created at compile time, since no runtime contexts exist then.
> >      Which is reasonable.
> > 
> > 
> > Can you initialize pointers in general, at compile-time?
> 
> You cannot initialize a pointer to runtime-allocated data at compile-time, and will never be able to. This particular example cannot ever work.

You can get pretty close by using structs with opCall defined though.

Sean Kelly <sean@invisibleduck.org> wrote:

> Don Wrote:
>
>> Philippe Sigaud wrote:
>> > On Mon, Jul 19, 2010 at 22:01, torhu <no@spam.invalid> wrote:
>> >
>> >
>> >
>> >     I wasn't able to make it work.
>> >
>> >
>> > Me too :(
>> >
>> >
>> >     The compiler probably sees delegates as something that just can't  
>> be
>> >     created at compile time, since no runtime contexts exist then.
>> >      Which is reasonable.
>> >
>> >
>> > Can you initialize pointers in general, at compile-time?
>>
>> You cannot initialize a pointer to runtime-allocated data at
>> compile-time, and will never be able to. This particular example cannot
>> ever work.
>
> You can get pretty close by using structs with opCall defined though.

Not if the function I got only takes delegates, and needs to only take
delegates.

-- 
Simen

Simen kjaeraas wrote:
> Sean Kelly <sean@invisibleduck.org> wrote:
> 
>> Don Wrote:
>>
>>> Philippe Sigaud wrote:
>>> > On Mon, Jul 19, 2010 at 22:01, torhu <no@spam.invalid> wrote:
>>> >
>>> >
>>> >
>>> >     I wasn't able to make it work.
>>> >
>>> >
>>> > Me too :(
>>> >
>>> >
>>> >     The compiler probably sees delegates as something that just 
>>> can't be
>>> >     created at compile time, since no runtime contexts exist then.
>>> >      Which is reasonable.
>>> >
>>> >
>>> > Can you initialize pointers in general, at compile-time?
>>>
>>> You cannot initialize a pointer to runtime-allocated data at
>>> compile-time, and will never be able to. This particular example cannot
>>> ever work.
>>
>> You can get pretty close by using structs with opCall defined though.
> 
> Not if the function I got only takes delegates, and needs to only take
> delegates.
> 
Eventually, it should be possible to have a compile-time delegate which points to an immutable static variable. Doesn't work at the moment, though.