Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.c++ > #14882 > unrolled thread

constepxr | Isn't this a constant expression?

Back to article view | Back to comp.lang.c++

Contents

  constepxr | Isn't this a constant expression? Samkit Jain <samkit.feeds@gmail.com> - 2012-03-11 03:40 -0700
    Re: constepxr | Isn't this a constant expression? darylew@gmail.com - 2012-03-12 06:22 -0700
    constepxr | Isn't this a constant expression? Johannes Schaub - litb <schaub.johannes@googlemail.com> - 2012-03-13 23:52 -0700
      Re: constepxr | Isn't this a constant expression? 88888 Dihedral <dihedral88888@googlemail.com> - 2012-03-14 20:03 -0700
        Re: constepxr | Isn't this a constant expression? Samkit Jain <samkit.feeds@gmail.com> - 2012-03-15 10:04 -0700
          Re: constepxr | Isn't this a constant expression? Johannes Schaub <schaub.johannes@googlemail.com> - 2012-03-16 23:16 +0100
            Re: constepxr | Isn't this a constant expression? Samkit Jain <samkit.feeds@gmail.com> - 2012-03-18 08:09 -0700
      Re: constepxr | Isn't this a constant expression? Miles Bader <miles@gnu.org> - 2012-03-16 09:46 +0900

#14882 — constepxr | Isn't this a constant expression?

Compiler: gcc 4.6.1

Code:

struct data
{
    constexpr data(int i) :i_(i) {}
    int i_;
};

template <typename T, int Size>
constexpr void const* get_addr(T const (&elements)[Size]) { return
(void const*)elements; }

template <typename T>
constexpr void const* get_addr(T const& element) { return (void
const*)&element; }

int main()
{
    constexpr data elements[] = { data(10) };
    constexpr data ten(10);
    static_assert(get_addr(elements) == get_addr(elements), "not equal
- 1");   // doesn't compile?
    static_assert(elements == elements, "not equal - 2");
    static_assert(get_addr(ten) == get_addr(ten), "not equal -
3");             // doesn't compile?
    static_assert(&ten == &ten, "not equal - 4");
    return 0;
}

[toc] | [next] | [standalone]

#14910

On Sunday, March 11, 2012 6:40:42 AM UTC-4, Samkit Jain wrote:
> Compiler: gcc 4.6.1
> 
> Code:
> 
> struct data
> {
>     constexpr data(int i) :i_(i) {}
>     int i_;
> };
> 
> template <typename T, int Size>
> constexpr void const* get_addr(T const (&elements)[Size]) { return
> (void const*)elements; }
> 
> template <typename T>
> constexpr void const* get_addr(T const& element) { return (void
> const*)&element; }

IIUC, constexpr stuff is only for value semantics.  Addresses are for identity semantics, so constexpr is ignored.  Compile-time evaluations are done generally before object addresses are finalized, so pointer stuff can't really ever be constexpr.

Daryle W.

[toc] | [prev] | [next] | [standalone]

#14926

The address of local nonstatic variables are not a constant expression. 

[toc] | [prev] | [next] | [standalone]

#14934

Johannes Schaub - litb於 2012年3月14日星期三UTC+8下午2時52分31秒寫道：
> The address of local nonstatic variables are not a constant expression.



Johannes Schaub - litb於 2012年3月14日星期三UTC+8下午2時52分31秒寫道：
> The address of local nonstatic variables are not a constant expression.



Johannes Schaub - litb於 2012年3月14日星期三UTC+8下午2時52分31秒寫道：
> The address of local nonstatic variables are not a constant expression.



Johannes Schaub - litb於 2012年3月14日星期三UTC+8下午2時52分31秒寫道：
> The address of local nonstatic variables are not a constant expression.

Yes, don't pass references and  addresses in the STACK in a function return.

That's quite buggy for the stack frames.
 

[toc] | [prev] | [next] | [standalone]

#14940

The below lines are compiling fine. What does compiler thinks and what
values are evaluated at compile time?

    static_assert(elements == elements, "not equal - 2");
    static_assert(&ten == &ten, "not equal - 4");

Also, this compiles:

    static_assert(get_addr("!23") == get_addr("123"), "not equal -
5");

[toc] | [prev] | [next] | [standalone]

#14952

Am 15.03.2012 18:04, schrieb Samkit Jain:
> The below lines are compiling fine. What does compiler thinks and what
> values are evaluated at compile time?
>
>      static_assert(elements == elements, "not equal - 2");
>      static_assert(&ten ==&ten, "not equal - 4");
>

This works because the address of a local variable is a core constant 
expression (these can be thought of intermediary results within the 
computation of a constant expression). The assignment then evaluates to 
true and yields a constant expression bool integer.

[toc] | [prev] | [next] | [standalone]

#14975

On Mar 17, 3:16 am, Johannes Schaub <schaub.johan...@googlemail.com>
wrote:
> Am 15.03.2012 18:04, schrieb Samkit Jain:
>
> > The below lines are compiling fine. What does compiler thinks and what
> > values are evaluated at compile time?
>
> >      static_assert(elements == elements, "not equal - 2");
> >      static_assert(&ten ==&ten, "not equal - 4");
>
> This works because the address of a local variable is a core constant
> expression (these can be thought of intermediary results within the
> computation of a constant expression). The assignment then evaluates to
> true and yields a constant expression bool integer.

By this argument I should be able to instantiate another template
which expects a <void const*> as non-type template parameter, but I
can't.
Obviously, I am doing something wrong but I want to understand it
according to standard wordings.

template <void const* addr>
struct unique_data
{
};

unique_data<get_addr(ten)> u_obj;

[toc] | [prev] | [next] | [standalone]

#14947

Johannes Schaub - litb <schaub.johannes@googlemail.com> writes:
> The address of local nonstatic variables are not a constant expression. 

I dunno what the standard says, but his confusion seems understandable
to me.

After all, he's not taking the address of a random local variable, but
of a _constexpr_ local variable, and I think the most natural way of
thinking about such a thing is "local in scope but global in extent,"
exactly like a local static variable.

The idea of a variable with a value explicitly fixed at compile-time,
but an address that changes at runtime seems just sort of weird...

-miles

-- 
Friendship, n. A ship big enough to carry two in fair weather, but only one
in foul.

[toc] | [prev] | [standalone]

Back to top | Article view | comp.lang.c++

csiph-web