Here's a modification to:
1) hide the intermediate struct (as usual in std.algorithm, I
forgot what this trick is called)
2) work with ranges, not just arrays (ie will work with iota, see
unittest)
3) accept input without "in" attribute;
4) accept arbitrary predicate, not just "==x"
Please comment if I'm doing anything not casher.
5) In particular, is there a way to avoid the dummy argument in
this(int ignore)?
6) for (3), I didn't use "in": I believe that is the usual
behavior in std.algorithm, where the caller is responsible to
pass in items.dup or items.save if he so wishes. Is that true?
7) I had to remove nothrow and pure, is that needed in the
general range case? how to fix it if it is?
----
unittest{
assert(iota(10).findIndexes!`a%3==1`==[1,4,7]);
}
auto findIndexes(alias pred,R)(R items) if(isInputRange!R) {
import std.functional:unaryFun;
struct FindIndexes(alias pred2) {
private size_t _front;
this(int ignore=0) {//TODO:this() not allowed so... anything
better?
while(!items.empty){
if(pred2(items.front)) break;
_front++;
items.popFront;
}
}
@property bool empty() const {
return items.empty;
}
@property size_t front() const {
return _front;
}
void popFront() {
_front++;
items.popFront;
while(!items.empty){
if(pred2(items.front)) break;
_front++;
items.popFront;
}
}
}
return FindIndexes!(unaryFun!pred)();
}
----
