Adam posted this in beerconf, as an alternative to std.meta.NoDuplicates:

template NoDuplicates(T...) {
        import std.meta;
        string getIndexesCode() {
                size_t[string] things;
                foreach(idx, t; T)
                        things[t.mangleof] = idx;
                string code;
                foreach(k, v; things) {
                        if(code.length) code ~= ", ";
                        code ~= "T[" ~ cast(char) (v + '0') ~ "]";
                }
                return "AliasSeq!(" ~ code ~ ")";
        }
        alias NoDuplicates = mixin(getIndexesCode());
}

Now, this works only on types, because of t.mangleof. So for instance if you do 3.manglof, it gives "i", the mangle for int. But I figured out, if I combine both the mangleof and stringof, you get something that is bound to be unique for each value AND type.

Here's an alternate:

template NoDuplicates(T...)
{
    import std.meta : AliasSeq;

    string getIndexesCode()
    {
        size_t[string] things;
        static foreach (idx, alias t; T)
            things[t.mangleof ~ t.stringof] = idx;
        string code;
        foreach (k, v; things)
        {
            code ~= "/*" ~ k ~ "*/ T[" ~ cast(char)(v + '0') ~ "],";
        }
        return code;
    }

    // pragma(msg, getIndexesCode()); // uncomment to see the keys and indexes
    mixin("alias NoDuplicates = AliasSeq!(" ~ getIndexesCode() ~ ");");
}

I tested this a bit, and it works pretty well.

I think the only thing that doesn't work with this is a lambda, which __traits(isSame) treats specially. Any other problems with this?

Another way to solve this is just to use CTFE brute force:

template NoDuplicates(T...)
{
    import std.meta : AliasSeq;

    string getIndexesCode()
    {
        string code;
        static foreach(i; 0 .. T.length)
        {{
            bool includeit = true;
            static foreach(j; i + 1 .. T.length)
            	static if(__traits(isSame, T[i], T[j]))
                   includeit = false;
            if(includeit)
                code ~= "T[" ~ i.stringof ~ "],";
        }}
        return code;
    }

    //pragma(msg, getIndexesCode());
    mixin("alias NoDuplicates = AliasSeq!(" ~ getIndexesCode() ~ ");");
}

I'm not sure which one would perform better. One has better complexity, but the other is more correct (you could possibly split out lambdas or other problematic things into their own sub-function). But both I think might be a step up over the current NoDuplicates, which is horrendously complicated.

-Steve

On 9/26/20 11:13 PM, Steven Schveighoffer wrote:


Oops, posted in wrong forum.. sorry.

-Steve