[Bug c++/19628] New: g++ no longer accepts __builtin_constant_p in constant-expressions

[Bug c++/19628] New: g++ no longer accepts __builtin_constant_p in constant-expressions

[austern at apple dot com]

The following test case is accepted by 3.3 (both C and C++) and by 4.0 C, but not by 4.0 C++.
#define FOO(x) (__builtin_constant_p(x) ? 1 : 2)

int foo(int n) {
  switch(n) {
  case FOO(3):
    return 1;
  default:
    return 2;
  }
}

Was this change intentional?  Even if it was, the C/C++ incompatibility is unfortunate.  This sort of 
construct is likely to appear in headers used by both languages.  (And we can't very well point to 
language standards in deciding whether __builtin_constant_p should be considered a constant-
expression, since it's a GNU extension.)

-- 
           Summary: g++ no longer accepts __builtin_constant_p in constant-
                    expressions
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: austern at apple dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: powerpc-apple-darwin7.7.0
  GCC host triplet: powerpc-apple-darwin7.7.0
GCC target triplet: powerpc-apple-darwin7.7.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-25 20:44 -------
: Search converges between 2003-01-27-trunk (#173) and 2003-02-03-trunk (#174).

I should note this was not accepted before:
: Search converges between 2001-01-14-trunk (#2) and 2001-01-21-trunk (#3).

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |rejects-valid
            Summary|g++ no longer accepts       |[3.4/4.0 Regression] g++ no
                   |__builtin_constant_p in     |longer accepts
                   |constant-expressions        |__builtin_constant_p in
                   |                            |constant-expressions
   Target Milestone|---                         |3.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[austern at apple dot com]

------- Additional Comments From austern at apple dot com  2005-01-28 20:17 -------
This was undoubtedly caused by Mark's large checkin on 2003-01-29, which included changes to 
constant-expression handling.

I do not believe the __builtin_constant_p change was deliberate.  The intent behind this patch, as far as 
I can tell, was to reject constructs that the C++ standard explicitly says are not constant-expressions, 
such as "(1, 2)".  It looks to me like rejecting GNU extensions like __builtin_constant_p in constant-
expressions was just a side effect. 

As far as I can tell, rejecting __builtin_constant_p is a change that (a) was not deliberate; (b) is 
undocumented; (c) introduced a C/C++ incompatbility; and (d) causes previously working code to 
break.  I think that justifies calling it a bug.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|rejects-valid               |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-29 17:31 -------
Patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-01/msg02167.html>.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |patch, rejects-valid
   Last reconfirmed|0000-00-00 00:00:00         |2005-01-29 17:31:57
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-30 19:56 -------
New patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-01/msg02212.html>.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[mmitchel at gcc dot gnu dot org]

------- Additional Comments From mmitchel at gcc dot gnu dot org  2005-01-31 01:39 -------
I agree that __builtin_constant_p should be allowed in an integral
constant-expression.

The only thing I don't like about the patch is that the documentation for
builtin_is_constant_p doesn't obviously match the name; the documentation says
it checks for built-ins that can appear in consant-expressions, but the name
might seem to refer exclusively to __builtin_constant_p.  A name like
"builtin_valid_in_constant_expr_p" would seem better.

The patch is OK with that change.

(Matt, I think you should also check what happens when __builtin_constant_p is
used as a template argument with dependent arguments, like:

  template <int I>
  void f(int &[__builtin_constant_p (I) + 2]);

Do we issue errors at substitution time?  If not, do we emit reasonable
manglings?  Etc.  We don't have to fix these issues before re-enabling the core
functionality, but if there are problems we should open new PRs for them.)

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[austern at apple dot com]

------- Additional Comments From austern at apple dot com  2005-01-31 02:38 -------
Subject: Re:  [3.4/4.0 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

> it checks for built-ins that can appear in consant-expressions, but 
> the name
> might seem to refer exclusively to __builtin_constant_p.  A name like
> "builtin_valid_in_constant_expr_p" would seem better.

I agree, that's a better name.  I'll change it.

> (Matt, I think you should also check what happens when 
> __builtin_constant_p is
> used as a template argument with dependent arguments, like:
>   template <int I>
>   void f(int &[__builtin_constant_p (I) + 2]);

Actually, it ICEs.  That would be a regression.  I think I'd better fix 
it before checking this in.

             --Matt



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628

[Bug c++/19628] [3.4 Regression] g++ no longer accepts __builtin_constant_p in constant-expressions

[mmitchel at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4.4                       |3.4.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19628