// Copyright 2019 The Go Authors. All rights reserved. // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. // This file shows some examples of type-parameterized functions. package p // Reverse is a generic function that takes a []T argument and // reverses that slice in place. func Reverse[T any](list []T) { i := 0 j := len(list)-1 for i < j { list[i], list[j] = list[j], list[i] i++ j-- } } func _() { // Reverse can be called with an explicit type argument. Reverse[int](nil) Reverse[string]([]string{"foo", "bar"}) Reverse[struct{x, y int}]([]struct{x, y int}{{1, 2}, {2, 3}, {3, 4}}) // Since the type parameter is used for an incoming argument, // it can be inferred from the provided argument's type. Reverse([]string{"foo", "bar"}) Reverse([]struct{x, y int}{{1, 2}, {2, 3}, {3, 4}}) // But the incoming argument must have a type, even if it's a // default type. An untyped nil won't work. // Reverse(nil) // this won't type-check // A typed nil will work, though. Reverse([]int(nil)) } // Certain functions, such as the built-in `new` could be written using // type parameters. func new[T any]() *T { var x T return &x } // When calling our own `new`, we need to pass the type parameter // explicitly since there is no (value) argument from which the // result type could be inferred. We don't try to infer the // result type from the assignment to keep things simple and // easy to understand. var _ = new[int]() var _ *float64 = new[float64]() // the result type is indeed *float64 // A function may have multiple type parameters, of course. func foo[A, B, C any](a A, b []B, c *C) B { // do something here return b[0] } // As before, we can pass type parameters explicitly. var s = foo[int, string, float64](1, []string{"first"}, new[float64]()) // Or we can use type inference. var _ float64 = foo(42, []float64{1.0}, &s) // Type inference works in a straight-forward manner even // for variadic functions. func variadic[A, B any](A, B, ...B) int { panic(0) } // var _ = variadic(1) // ERROR not enough arguments var _ = variadic(1, 2.3) var _ = variadic(1, 2.3, 3.4, 4.5) var _ = variadic[int, float64](1, 2.3, 3.4, 4) // Type inference also works in recursive function calls where // the inferred type is the type parameter of the caller. func f1[T any](x T) { f1(x) } func f2a[T any](x, y T) { f2a(x, y) } func f2b[T any](x, y T) { f2b(y, x) } func g2a[P, Q any](x P, y Q) { g2a(x, y) } func g2b[P, Q any](x P, y Q) { g2b(y, x) } // Here's an example of a recursive function call with variadic // arguments and type inference inferring the type parameter of // the caller (i.e., itself). func max[T interface{ ~int }](x ...T) T { var x0 T if len(x) > 0 { x0 = x[0] } if len(x) > 1 { x1 := max(x[1:]...) if x1 > x0 { return x1 } } return x0 } // When inferring channel types, the channel direction is ignored // for the purpose of type inference. Once the type has been in- // fered, the usual parameter passing rules are applied. // Thus even if a type can be inferred successfully, the function // call may not be valid. func fboth[T any](chan T) {} func frecv[T any](<-chan T) {} func fsend[T any](chan<- T) {} func _() { var both chan int var recv <-chan int var send chan<-int fboth(both) fboth(recv /* ERROR cannot use */ ) fboth(send /* ERROR cannot use */ ) frecv(both) frecv(recv) frecv(send /* ERROR cannot use */ ) fsend(both) fsend(recv /* ERROR cannot use */) fsend(send) } func ffboth[T any](func(chan T)) {} func ffrecv[T any](func(<-chan T)) {} func ffsend[T any](func(chan<- T)) {} func _() { var both func(chan int) var recv func(<-chan int) var send func(chan<- int) ffboth(both) ffboth(recv /* ERROR cannot use */ ) ffboth(send /* ERROR cannot use */ ) ffrecv(both /* ERROR cannot use */ ) ffrecv(recv) ffrecv(send /* ERROR cannot use */ ) ffsend(both /* ERROR cannot use */ ) ffsend(recv /* ERROR cannot use */ ) ffsend(send) } // When inferring elements of unnamed composite parameter types, // if the arguments are defined types, use their underlying types. // Even though the matching types are not exactly structurally the // same (one is a type literal, the other a named type), because // assignment is permitted, parameter passing is permitted as well, // so type inference should be able to handle these cases well. func g1[T any]([]T) {} func g2[T any]([]T, T) {} func g3[T any](*T, ...T) {} func _() { type intSlize []int g1([]int{}) g1(intSlize{}) g2(nil, 0) type myString string var s1 string g3(nil, "1", myString("2"), "3") g3(&s1, "1", myString /* ERROR does not match */ ("2"), "3") type myStruct struct{x int} var s2 myStruct g3(nil, struct{x int}{}, myStruct{}) g3(&s2, struct{x int}{}, myStruct{}) g3(nil, myStruct{}, struct{x int}{}) g3(&s2, myStruct{}, struct{x int}{}) } // Here's a realistic example. func append[T any](s []T, t ...T) []T { panic(0) } func _() { var f func() type Funcs []func() var funcs Funcs _ = append(funcs, f) } // Generic type declarations cannot have empty type parameter lists // (that would indicate a slice type). Thus, generic functions cannot // have empty type parameter lists, either. This is a syntax error. func h[] /* ERROR empty type parameter list */ () {} func _() { h /* ERROR cannot index */ [] /* ERROR operand */ () } // Parameterized functions must have a function body. func _ /* ERROR missing function body */ [P any]()