www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Making Metaprogramming Pleasant and Fun

reply Chad J <chadjoan __spam.is.bad__gmail.com> writes:
So I'm seeing this recurring pattern on the newsgroup:

Person 1:  "Hey we should implement feature X!  It'll be awesome!"
Person 2:  "Bah, you can implement that in current D code with mixins:"

< insert example of typical mixin usage in D >

mixin(Foo!(args,
`//blah blah
`));

OR

mixin Foo!(args,
`//blah blah
`);

Person 3:  "That's hideous!  Let's not do it."
(silence or <insert discussion of mixins and ugliness>)
(maybe insert mention of CTFE gobbling up all of your memory and kittens)

As much as I hate to do it, I'm going to be /that/ guy and make another
feature suggestion.  Or 2 or 3.

----------------------------
Suggestion 1:

For any template whose argument list ends with a string argument, such
as this one:

template foo( T, string arg )
{
...
}

... allow it to be instantiated (or mixed in) like so:

foo!(T)
{
    // Some token string.
}

or

mixin foo!(T)
{
    // Some token string.
}

The contents of the curly braces are treated as a token string, like the
q{ foo + bar } variety, minus the unbalancing q.

This gets us to things like

mixin property!(int,"foo")
{
    get { return $; }
    set { $ = value; }
}

Bonus points: generalize it to IFTI functions.

void foo(T, string code)( T arg )

can be called like so:

foo(arg)
{
    // code goes here.
}

Then use your imagination.

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

Let templates include expression statements.  Optionally make it invalid
to NOT use these as mixin templates.  This would allow us to convert
things in general from

mixin(Foo!(args)
{
    //blah blah
});

to

mixin Foo!(args)
{
    //blah blah
}

Those aforementioned bonus points become much more valuable with this.

void main()
{
    // etc
    mixin foreachReverse!( "ref elem", someRange )
    {
        elem = someFunc(elem);
        // and so on.
    }
}

Admittedly I haven't pondered the negative consequences of this
particular change too much.

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

Bikeshedding: make a mixin operator to complement the keyword.

Thus

mixin Foo!(args) { ... }

becomes

Foo (args) { ... }

Now then...

class SomeElementType
{
    ...

    property (int,"foo")
    {
        get { ... }
        set { ... }
    }

}

void main()
{
    // etc
    foreachReverse ( "ref elem", someRange )
    {
        elem.foo = someFunc(elem);
        // and so on.
    }
}

Smooth.

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

Also, the CTFE memory leak.  But I'm sure everyone who needs to know
about that already knows and is sick of hearing about it.

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

That's all I got for now.
Aug 23 2009
next sibling parent Chad J <chadjoan __spam.is.bad__gmail.com> writes:
Chad J wrote:
 
 ----------------------------
 Suggestion 1:
 
 For any template whose argument list ends with a string argument, such
 as this one:
 
 ...
 
 ----------------------------

 
 Let templates include expression statements.  Optionally make it invalid
 to NOT use these as mixin templates.  This would allow us to convert
 things in general from
 
 ...
 
 ----------------------------

 
 Bikeshedding: make a mixin operator to complement the keyword.
 
 ...
 

Sorry, editing fail.
Aug 23 2009
prev sibling parent Sergey Gromov <snake.scaly gmail.com> writes:
Mon, 24 Aug 2009 00:25:19 -0400, Chad J wrote:

 Suggestion 1:
 
 For any template whose argument list ends with a string argument, such
 as this one:
 
 template foo( T, string arg )
 {
 ...
 }
 
 ... allow it to be instantiated (or mixed in) like so:
 
 foo!(T)
 {
     // Some token string.
 }
 
 or
 
 mixin foo!(T)
 {
     // Some token string.
 }

I agree it would be nice to add some sugar to mixin usage. But I'm afraid this particular proposal has too many corner cases and parsing ambiguities. I cannot point out any right now, it's just an impression.
Aug 30 2009