[Bug c++/12370] New: [3.4 regression] wrong code involving friends

[Bug c++/12370] New: [3.4 regression] wrong code involving friends

[reichelt at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12370

           Summary: [3.4 regression] wrong code involving friends
           Product: gcc
           Version: 3.4
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: critical
          Priority: P1
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: reichelt at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org,rth at redhat dot com
  GCC host triplet: i686-pc-linux-gnu

The following code snippet miscompiles with mainline on i686-pc-linux-gnu:

=====================================================================
template <typename T> struct A
{
    T x;
    A(T t) : x(t) {}
    friend A<int> foo (const A<unsigned>&);
};

A<int> foo (const A<unsigned>& a) { A<int> res(a.x); return res; }

int main()
{
    return foo(A<unsigned>(0)).x;
}
=====================================================================

It should return "0" but returns "16" instead.
If I delete the friend declaration, the code compiles fine.

This regression was introduced between 2003-08-29 22:00 and 23:30.

Richard, there were 5 commits from you in this timeframe

  http://gcc.gnu.org/ml/gcc-cvs/2003-08/msg00940.html
  http://gcc.gnu.org/ml/gcc-cvs/2003-08/msg00942.html
  http://gcc.gnu.org/ml/gcc-cvs/2003-08/msg00943.html
  http://gcc.gnu.org/ml/gcc-cvs/2003-08/msg00945.html
  http://gcc.gnu.org/ml/gcc-cvs/2003-08/msg00947.html

and one from Mark (which is about labels and shouldn't be the problem here).
Could you please have a look?

Thanks,
Volker

[Bug c++/12370] [3.4 regression] wrong code involving friends

[bangerth at dealii dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12370


bangerth at dealii dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-09-22 21:15:15
               date|                            |
   Target Milestone|---                         |3.4


------- Additional Comments From bangerth at dealii dot org  2003-09-22 21:15 -------
Confirmed. There's something odd going on here. It's not
a problem with signed vs unsigned, btw, since we have the
same behavior with this testcase:
------------------------------
struct B {
    int x;
};

template <int=0> struct A
{
    int x;
    friend B foo (const A<>&);
};

B foo (const A<>& a) { B res = {a.x};  return res; }

int main()
{
  return foo((A<>){0}).x;
}
---------------------------------
This returns 120 on my system (and the original testcase returned 84 here,
so this seems to be mostly random). I don't know what the friend declaration
can have possibly to do with the result, but removing it really makes
the bug go away. Likewise for making A a non-template...

W.

[Bug c++/12370] [3.4 regression] wrong code when adding friend

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12370


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[3.4 regression] wrong code |[3.4 regression] wrong code
                   |involving friends           |when adding friend


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-09-23 02:12 -------
>From looking into the RTL:
(insn 16 15 17 (set (mem/f:SI (reg/f:SI 56 virtual-outgoing-args) [0 S4 A32])
        (addressof:SI (reg/v:SI 61) 60 0x400e28dc)) -1 (nil)
    (nil))

61 is not used any more after that function or before it either.

Getting rid of the friend:
(insn 16 15 17 (set (mem/f:SI (reg/f:SI 56 virtual-outgoing-args) [0 S4 A32])
        (reg/v/f:SI 58 [ <anonymous> ])) -1 (nil)
    (nil))
Which is right.  Looks some thing is going wrong because of the friend, maybe another GC 
problem again?

[Bug c++/12370] [3.4 regression] wrong code when adding friend

[rth at redhat dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12370



------- Additional Comments From rth at redhat dot com  2003-09-26 22:56 -------
Subject: Re:  [3.4 regression] wrong code when adding friend

On Tue, Sep 23, 2003 at 02:12:25AM -0000, pinskia at gcc dot gnu dot org wrote:
> Looks some thing is going wrong because of the friend, maybe another GC 
> problem again?

Not a gc problem, exactly.  It is fallout from one of my patches however.
We were losing track of the fact that nrvo was to be applied, by losing
track of the cached information in cfun.

The following cures the given test case; general testing to follow.


r~


	* decl.c (duplicate_decls): Copy DECL_SAVED_INSNS too.

Index: decl.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/cp/decl.c,v
retrieving revision 1.1133
diff -c -p -d -u -r1.1133 decl.c
--- decl.c	26 Sep 2003 15:21:30 -0000	1.1133
+++ decl.c	26 Sep 2003 22:52:54 -0000
@@ -3206,7 +3206,10 @@ duplicate_decls (tree newdecl, tree oldd
 	  if (CAN_HAVE_FULL_LANG_DECL_P (newdecl)
 	      && DECL_LANG_SPECIFIC (newdecl)
 	      && DECL_LANG_SPECIFIC (olddecl))
-	    DECL_SAVED_TREE (newdecl) = DECL_SAVED_TREE (olddecl);
+	    {
+	      DECL_SAVED_TREE (newdecl) = DECL_SAVED_TREE (olddecl);
+	      DECL_SAVED_INSNS (newdecl) = DECL_SAVED_INSNS (olddecl);
+	    }
 	}
 
       /* Merge the section attribute.

[Bug c++/12370] [3.4 regression] wrong code when adding friend

[cvs-commit at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12370



------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-10-13 23:06 -------
Subject: Bug 12370

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	reichelt@gcc.gnu.org	2003-10-13 23:06:38

Modified files:
	gcc/cp         : ChangeLog 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/g++.dg/other: friend2.C 

Log message:
	PR c++/12370
	* g++.dg/other/friend2.C: New test.
	* ChangeLog: Add PR number to patch for PR c++/12370.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/ChangeLog.diff?cvsroot=gcc&r1=1.3725&r2=1.3726
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.3118&r2=1.3119
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/g++.dg/other/friend2.C.diff?cvsroot=gcc&r1=NONE&r2=1.1