very often I would wish that sort (+ related functions) could take unaryFun arguments and convert them to binaryFun as follows:

//pseudocode:
template unaryToBinaryComp(alias foo) {
bool unaryToBinaryComp(T)(T a, T b) if (__traits(compiles,foo(a) < foo(b))
) {
return foo(a) < foo(b);
}
}


Using this we could support much nicer syntax for sorting with unary functions, for example:

sort!"foo(a)" <=>sort!(unaryToBinaryComp!(unaryFun!"foo(a)"))
<=> sort!"foo(a) < foo(a)"
sorting in reverse order is easy: just use sort!"-foo(a)" (works for
signed, unsigned, floating point types etc).

Examples of use:

E1)

struct A{
double score;
int index;
}

[A.init,A.init].sort!"a.score";
instead of:
[A.init,A.init].sort!"a.score < b.score";

E2)

mytuple.sort!"a[0]"
instead of:
[A.init,A.init].sort!"a[0]<b[0]"

I find that in large majority of cases, binary fun is used in that way. So let's support both unaryFun and binaryFun.

On Tuesday, 28 May 2013 at 08:16:47 UTC, Timothee Cour wrote:
> sort!"foo(a)" <=>sort!(unaryToBinaryComp!(unaryFun!"foo(a)"))
> <=> sort!"foo(a) < foo(a)"
> sorting in reverse order is easy: just use sort!"-foo(a)"

In this case, schwartzSort might actually be more appropriate.

But in general, I see your point.

David

On 5/28/13 4:16 AM, Timothee Cour wrote:
> very often I would wish that sort (+ related functions) could take
> unaryFun arguments and convert them to binaryFun as follows:
>
> //pseudocode:
> template unaryToBinaryComp(alias foo) {
> bool unaryToBinaryComp(T)(T a, T b) if (__traits(compiles,foo(a) <
> foo(b)) ) {
> return foo(a) < foo(b);
> }
> }

I actually have a branch in my code that does exactly that. I abandoned it because the code, error messages, and documentation got really confusing.

It may be worth to add a keySort algorithm that only works with unary keys, but then I thought people can always write a small lambda.


Andrei