spec: taking address of a string literal

[gopherbot]

by bronze1man:

What steps will reproduce the problem?
If possible, include a link to a program on play.golang.org.
http://play.golang.org/p/aI6gKoGP_3

What is the expected output?
compilation should success.

What do you see instead?
prog.go:8: cannot take the address of []byte literal
prog.go:9: cannot take the address of string("a")
prog.go:10: cannot take the address of "a"
prog.go:11: cannot take the address of t1("a")

Please provide any additional information below.

http://golang.org/ref/spec#Address_operators
In the spec it not forbid to get an address of a string literal

But I can get an address of a string literal,and I do not need to care about free that
pointer.
        tmp:="a"
    fmt.Println(&tmp)
http://play.golang.org/p/B4Owwtsqn9

Just a syntactic sugar.

[gopherbot]

Comment 1 by bronze1man:

I think taking address of a string literal is just like &[]byte{0} which malloc a
variable which value is []byte{0}
I think I should also taking address of a int literal, taking address of a func literal
...

[cznic]

Comment 2:

You can take only address of a "home" (variable, field, argument, ...). Literals are
"homeless".
However, the legal composite literal address taking syntax `foo := &T{whatever}` is an
exception. It's a handy shortcut for
    foo = new(T); foo = whatever
The difference to your proposal is that this is allowed only for _named types_. AFAICS,
allowing the same for anonymous type would introduce a new ambiguity to the grammar. I
suppose that's the reason why the &T{} is an "exception" and not a general concept.
#WorkingAsIntended
PS: Please send language _proposals_ to golang-nuts as the first option; to be discussed
there before optionally entering the issue tracker as an agreed on change.

[robpike]

Comment 3:

In general, one cannot take the address of a constant.

Status changed to WorkingAsIntended.