Re: g++ static data members [solaris27 + gcc-2.95.2/egcs]

[Mike Stump <mrs at windriver dot com>]

> Date: Mon, 6 Mar 2000 10:29:40 -0800
> To: gcc-bugs@gcc.gnu.org, bkoz@cygnus.com
> From: Benjamin Kosnik <bkoz@cygnus.com>

> I believe this problem has to do with cc1plus not putting the data
> into the correct section: here are two snippets from example code
> included below.

Well, if you had filed a complete bug report, it would be easier for
those that can help you to help you.  I came very close to not being
able to help you because of this.  I'd love to help you, as I would
love to see the new library code put into gcc.  I recommend that all
future bug reports include at least the testcase.  (g++ -E).  Luckily
I found a 2.7 machine lying around here to try it one.

> It's clear to me that _S_table should be in .data, not .bss.

That's fine, but wrong:

  3.6.2  Initialization of non-local objects          [basic.start.init]

1 The storage for objects with static storage duration (3.7.1) shall  be
  zero-initialized  (8.5)  before  any other initialization takes place.
  Objects of POD types (3.9) with static  storage  duration  initialized
  with  constant  expressions  (5.19)  shall  be  initialized before any
  dynamic initialization takes place.  Objects of namespace  scope  with
  static  storage  duration  defined  in  the  same translation unit and
  dynamically initialized shall be initialized in  the  order  in  which
  their  definition  appears  in  the  translation  unit.   [Note: 8.5.1
  describes the order in which aggregate members are  initialized.   The
  initialization of local static objects is described in 6.7.  ]

We are allowed to generate this type of code if we want to.  Do you
mean to critique our code generation strategy, or are you just being
confused by it, I can't tell.  My guess is that you are being confused
by it.  You cannot rely upon (in C++) it being initialized until the
dymanic point in program start up is hit: Welcome to C++.

> Any thoughts?  I've been unable to compile CVS egcs/gcc on
> solaris2.7, so anybody with a current g++ binary care to run the
> code below through it, and see what section _S_table is in?

It is run time initialized.  Some changes by Mark seem to have caused
it.(?)  It used to not be run time initialized.  I don't know why Mark
wanted to defer it.  Maybe that was an accident?  If so, I think it
should be moved back to static time initialization.  The change that I
was guessing is related was:

1999-09-30  Mark Mitchell  <mark@codesourcery.com>

> Is there a patch to put data like this into .data and not .bss on
> solaris2.7?

Below is a testcase for the problem:

extern unsigned int     *__ctype_mask;
typedef unsigned int size_t;
 
struct ctype_base {
  typedef unsigned int      mask;
};
 
 
template<typename _CharT>
class ctype   : public ctype_base
{ };
 
template<>
class ctype<char>  : public ctype_base
{
 
  typedef char char_type;
 private:
 
  const mask*               _M_table;
  bool                      _M_del;
  static const mask* const& _S_table;
 
 public:
 
  explicit
    ctype(const mask* __table = 0, bool __del = false,
          size_t __refs = 0) throw()
    : _M_table(__table == 0 ? _S_table: __table),
    _M_del(__table != 0 && __del) { }
};
 
const ctype_base::mask* const& ctype<char>::_S_table = __ctype_mask ;