expression templates

[Robert Jacques]

On Sun, 27 Mar 2011 15:46:54 -0400, enuhtac <enuhtac_lists at gmx.de> wrote:

> Hello everyone,
>
> I'm new to D and to this list (although I've had a look onto D a few  
> years ago). I hope you guys can help me with my questions.
>
> At the moment I'm trying to implement some expression template stuff. My  
> first goal is to encode an expression into a type representing that  
> expression without any additional functionality (like the possibility to  
> evaluate that expression). Actually this is very simple and short in D.  
> This is my approach:
>
> struct OpBinary( string Op, R1, R2 )
> {
>     alias typeof( mixin( "R1.EvalT.init" ~ Op ~ "R2.EvalT.init" ) )  
> EvalT;
>
>     enum string Operator = Op;
> };
>
> struct Constant( T, T v )
> {
>     alias T EvalT;
>
>     enum T value = v;
> };
>
> struct Expr( R )
> {
>     auto opBinary( string Op, R2 )( Expr!R2 )
>     {
>         return Expr!( OpBinary!( Op, R, R2 ) )();
>     }
>
>     auto opBinary( string Op, T )( T v ) if( isNumeric!T )
>     {
>         return Expr!( OpBinary!( Op, R, Constant!( T, v ) ) )();
>     }
>
>     auto opBinaryRight( string Op, T )( T v ) if( isNumeric!T )
>     {
>         return Expr!( OpBinary!( Op, Constant!( T, v ), R ) )();
>     }
> };
>
> But I cannot figure out how to implement expression templates for  
> comparison operators, which is crucial for my purpose. The opCmp  
> function is great for implementing comparison functionality, but when  
> building an expression template tree the information on the actual  
> comparison operator is needed. opCmp just knows that a comparison is  
> going on, the actual type of comparison is unknown.
> What I would like to have is something like this:
>
>     auto opCmp( string Op, R2 )( Expr!R2 )
>     {
>         return Expr!( OpBinary!( Op, R, R2 ) )();
>     }
>
> So opCmp knows about the actual operator and would just use my OpBinary  
> struct to encode it. But this is not possible.
>
> The only workaround for I this problem I can imagine is using my own  
> comparison functions instead of the comparison operators:
> op!"<"( a, b ) instead of a < b.
> Another possibility would be to call opBinary explicitly:
> a.opCmp!"<"( b )
> In this case I would not even have to write additional code.
>
> But these workarounds are ugly, if would greatly prefer the normal  
> comparison operators.
> Does anyone has an idea how to use them?
>
> Regards,
> enuhtac

Hmm... I don't know you're use case exactly, but it sounds like a case of  
operator overload abuse. The design of opCmp was inspired by the amount of  
bug prone repetition that happens with C++ style comparison operators.  
Furthermore, both opCmp and opEquals have fixed return types in order to  
limit abuse. (D also prevents the overload of certain operators for the  
same reason). The main reason behind expression templates is to avoid  
costly intermediates, but the expression is always going to be heavier  
weight than an int, so why can't you evaluated the expression then and  
there?