Re: optimization/7247: [3.2 regression] copy constructor missing when inlining enabled for i386

Re: optimization/7247: [3.2 regression] copy constructor missing when inlining enabled for i386

[jbuck]

Synopsis: [3.2 regression] copy constructor missing when inlining enabled for i386

State-Changed-From-To: analyzed->closed
State-Changed-By: jbuck
State-Changed-When: Fri Apr 25 19:51:03 2003
State-Changed-Why:
    Fixed for the next release (3.3).

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7247

Re: optimization/7247: [3.2 regression] copy constructor missing when inlining enabled for i386

[Janis Johnson]

The following reply was made to PR optimization/7247; it has been noted by GNATS.

From: Janis Johnson <janis187@us.ibm.com>
To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, gawrilow@math.tu-berlin.de,
   gcc-bugs@gcc.gnu.org, nobody@gcc.gnu.org
Cc:  
Subject: Re: optimization/7247: [3.2 regression] copy constructor missing when 
 inlining enabled for i386
Date: Tue, 14 Jan 2003 14:57:50 -0800

 This bug was fixed on the mainline with this patch:
 
 --- gcc/cp/ChangeLog ---
 2002-03-11  Dan Nicolaescu  <dann@ics.uci.edu>
           Daniel Berlin  <dan@dberlin.org>
 
       * cp-lang.c (ok_to_generate_alias_set_for_type): New function.
       (cxx_get_alias_set): Use it.
 
 --- gcc/ChangeLog ---
 2002-03-11  Dan Nicolaescu  <dann@ics.uci.edu>
           Daniel Berlin  <dan@dberlin.org>
 
       C++ alias analysis improvement.
       * alias.c (record_component_aliases): Record aliases for base
       classes too.
 
 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7247
 
 
 

Re: optimization/7247: [3.2 regression] copy constructor missing when inlining enabled for i386

[reichelt]

Old Synopsis: copy constructor missing when inlining enabled for i386
New Synopsis: [3.2 regression] copy constructor missing when inlining enabled for i386

State-Changed-From-To: feedback->analyzed
State-Changed-By: reichelt
State-Changed-When: Thu Dec 19 06:19:45 2002
State-Changed-Why:
    Confirmed (just for the record).
    
    Here's a reduced testcase that fails on the 3.2 branch.
    gcc 3.0.x, 3.3 and mainline do not seem to be affected.
    
    --------------------------------snip here---------------------------------
    extern "C" { extern int printf (const char *, ...) throw (); }
    
    struct A {
       const double d;
       A() : d(1e-4) {}
       void dump () const { printf("%e\n", d); }
    };
    
    struct B {
       A a;
       int i;
    
       B(const A& arg) : a(arg) { arg.dump(); a.dump();}
    };
    
    struct C : B {
       C(const A& arg) : B(arg) {}
    };
    
    struct X {
       int i, j, k;
    
       X() : j(0) {}
    };
    
    struct Y {
       template <typename T> Y(int, const T&) {}
    };
    
    inline int foo() { X x; return 0; }
    
    int main() {
       Y y(foo(), C(A()));
       return 0;
    }
    --------------------------------snip here---------------------------------
    
    Just compile with "-O2" or "-O -fschedule-insns2" (on i686-pc-linux-gnu)
    to get the wrong result
    
      1.000000e-04
      1.948841e-314
    
    instead of
    
      1.000000e-04
      1.000000e-04
    
    
    Regards,
    Volker

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7247