Koenig lookup in D?

[monarch_dodra]

On Tuesday, 27 August 2013 at 10:00:50 UTC, Peter Alexander wrote:
> On Tuesday, 27 August 2013 at 07:45:14 UTC, monarch_dodra wrote:
>> I think that D's non-hijack rules are pretty sweet, but I also
>> think that D should apply Koenig's rules: If a template is
>> dependent on a parameter from a particular module (say "a"), 
>> then
>> the functions inside that module should also be taken into
>> account when building the template instance. It would be
>> *partial* hijack in the sense that you can only hijack the
>> templates that operate on *your* types anyways, but in no case
>> change the behavior of that of others'.
>
> That still makes hijacking a real issue though. It limits the 
> hijacking to template functions, but it's still there.
>
> What if ADL was only used with UFCS on an instance of a 
> template type?
>
> e.g.
>
> void foo(T)(T t)
> {
>     import std.stdio;
>     writeln(t); // always std.stdio
>     t.writeln(); // used ADL, maybe std.stdio, but can be 
> specialised by T
>     "foo".writeln(); // always std.stdio
> }
>
> This kind of makes sense, because T could implement writeln() 
> as a member function anyway, so by using UFCS you are 
> voluntarily opening yourself up to hijacking.


That sounds relativelly safe. An even safer approach might be: 
Allow an ADL call only if there are no local symbols that match.

This way, you could argue that there is no hijacking, if there 
was no matching call.