by daniel.crettol:

package main

import "fmt"

type A struct {
    i int
}

func new1() *A {
    return &A{10}
}

func new2() *A {
    p := A{}
    p.i = 11
    return &p
}

func dummy(p *A) {
}

func new3() *A {
    p := A{}
    p.i = 12
    dummy(&p)
    return &p
}

func main() {
    a1 := new1()
    fmt.Printf("a1=%p %+v\n", a1, a1)

    a2 := new2()
    fmt.Printf("a2=%p %+v\n", a2, a2)

    a3 := new3()
    fmt.Printf("a3=%p %+v\n", a3, a3)
}

$ go build -gcflags '-N' bugtst.go && ./bugtst
a1=0xc208000150 &{i:10}
a2=0x7f31b606ddc0 &{i:11}
a3=0xc2080001e0 &{i:12}

$ go build -gcflags '' bugtst.go && ./bugtst
a1=0xc208000150 &{i:10}
a2=0xc2080001a8 &{i:11}
a3=0xc2080001e8 &{i:12}

$ go version
go version go1.3.1 linux/amd64

I had this pattern in a program and the struct generated by the new2 function was
corrupted after returned.

It seems to me that the compiler does not generate a copy of the structure when using
the new2 function but does it when we call the new1 and new2 function.

As the -N -L flags seems to be required to debug with gcc, it's really annoying.

Happy debugging :)