While porting Tango to D2 I ran into this issue.

When applying const tags to BitArray, I've found that opApply has two issues.

I have to at least double the number of opApply functions.  The variations are for const or not, and using an index or not.

So I now have 4 opApplys with the following signatures:

int opApply( int delegate(ref bool) dg )
int opApply( int delegate(ref size_t, ref bool) dg )
int opApply( int delegate(ref const(bool)) dg ) const
int opApply( int delegate(ref size_t, ref const(bool)) dg ) const

So first issue, it sucks that I have to put ref on the size_t (index). Builtin arrays don't permit this, so it makes no sense why the compiler can't figure out whether a ref parameter is allowed in the foreach loop by looking at the opApply signature.  I should be able to compile with:

int opApply( int delegate(size_t, ref bool) dg )

And furthermore, to save extra code and headache, I should be able to remove ref altogether when the struct is const:

int opApply( int delegate(size_t, bool) dg ) const

---------------------

Second issue, I get a compile error with the following unittest code:

        BitArray a = [1,0,1];

        int i;
        foreach( b; a )
        {
            switch( i )
            {
            case 0: assert( b == true );  break;
            case 1: assert( b == false ); break;
            case 2: assert( b == true );  break;
            default: assert( false );
            }
            i++;
        }

tango/core/BitArray.d(414): Error: cannot uniquely infer foreach argument types

WTF?  How am I supposed to make const-correct structs/classes that allow setting elements during foreach when the struct/class is not const?

Is this a bug, or is there another way to do this?

I know Walter is working on updating foreach so that it works with ranges, will fixes to these problems also be included in that update?

-Steve

Since nobody responded, I'll just file bugs.

-Steve


"Steven Schveighoffer" wrote
> While porting Tango to D2 I ran into this issue.
>
> When applying const tags to BitArray, I've found that opApply has two issues.
>
> I have to at least double the number of opApply functions.  The variations are for const or not, and using an index or not.
>
> So I now have 4 opApplys with the following signatures:
>
> int opApply( int delegate(ref bool) dg )
> int opApply( int delegate(ref size_t, ref bool) dg )
> int opApply( int delegate(ref const(bool)) dg ) const
> int opApply( int delegate(ref size_t, ref const(bool)) dg ) const
>
> So first issue, it sucks that I have to put ref on the size_t (index). Builtin arrays don't permit this, so it makes no sense why the compiler can't figure out whether a ref parameter is allowed in the foreach loop by looking at the opApply signature.  I should be able to compile with:
>
> int opApply( int delegate(size_t, ref bool) dg )
>
> And furthermore, to save extra code and headache, I should be able to remove ref altogether when the struct is const:
>
> int opApply( int delegate(size_t, bool) dg ) const
>
> ---------------------
>
> Second issue, I get a compile error with the following unittest code:
>
>        BitArray a = [1,0,1];
>
>        int i;
>        foreach( b; a )
>        {
>            switch( i )
>            {
>            case 0: assert( b == true );  break;
>            case 1: assert( b == false ); break;
>            case 2: assert( b == true );  break;
>            default: assert( false );
>            }
>            i++;
>        }
>
> tango/core/BitArray.d(414): Error: cannot uniquely infer foreach argument types
>
> WTF?  How am I supposed to make const-correct structs/classes that allow setting elements during foreach when the struct/class is not const?
>
> Is this a bug, or is there another way to do this?
>
> I know Walter is working on updating foreach so that it works with ranges, will fixes to these problems also be included in that update?
>
> -Steve