D2 code, foo4() shows that you can't define a pure function pointer in a function signature: pure int sqr(int x) { return x * x; } pure int foo1(TF)(TF func, int x) { // OK return func(x); } pure int foo2(typeof(&sqr) func, int x) { // OK return func(x); } alias pure int function(int) FN; pure foo3(FN func, int x) { // OK, from Simen kjaeraas return func(x); } pure int foo4(pure int function(int) func, int x) { // line 14, ERR return func(x); } void main() { assert(foo1(&sqr, 5) == 25); assert(foo2(&sqr, 5) == 25); assert(foo3(&sqr, 5) == 25); assert(foo4(&sqr, 5) == 25); } DMD 2.047 prints: test.d(14): basic type expected, not pure test.d(14): found 'pure' when expecting ')' test.d(14): semicolon expected following function declaration test.d(14): no identifier for declarator int function(int) test.d(14): semicolon expected, not 'int' test.d(14): semicolon expected, not ')' test.d(14): Declaration expected, not ')' test.d(16): unrecognized declaration Also, it generates too many error messages.
I think this is a problem of the keyword pure trying to act as a type specifier for the return type ( See also my bug 4734 for a similar issue). If you put pure after the function definition but before the identifier, like so: pure int foo4(int function(int) pure func, int x) { then your example compiles and all asserts pass.
*** Issue 10603 has been marked as a duplicate of this issue. ***
As mentioned by Andrej, it's possible if you put the keyword on the RHS, which is the recommended way to do it nowadays. I don't think that's an issue worth fixing, so marking as WONTFIX.