Re: [i386-PE] bug with ordering of init_priority vs default ctors (and possible fix)

[Christopher Faylor <cgf at alum dot bu dot edu>]

On Sat, Apr 17, 2004 at 08:07:17PM +1200, Danny Smith wrote:
>The g++ extension init_priority does not work correctly
>on cygwin or mingw, causing long-time failure of g++.dg/special/initp1.C.
>
>The bug appears to be due to the way ctors are ordered by ld.
>
>A simpler testcase. reproduced below, was reported to cygwin list 
>
>// init_priority.cpp
>// Written by Rene Prillop
>// http://www.cygwin.com/ml/cygwin/2003-01/msg01155.html
>// 
>#include <stdio.h>
>
>class CIntHolder{
>public:
>	CIntHolder(int iValue) { m_iValue=iValue; }
>	int m_iValue;
>};
>
>static int g_Counter;
>
>CIntHolder g_Object3(++g_Counter);
>CIntHolder __attribute__ ((init_priority(102))) g_Object2(++g_Counter);
>CIntHolder __attribute__ ((init_priority(101))) g_Object1(++g_Counter);
>CIntHolder g_Object4(++g_Counter);
>
>int main(){
>	printf("%d %d %d %d\n", g_Object1.m_iValue, 
>				g_Object2.m_iValue,
>				g_Object3.m_iValue,
>			        g_Object4.m_iValue);
>	return 0;
>}
> 
>When this is compiled and executed, it ouputs
>
>3 4 1 2
>
>rather than the expected 1 2 3 4
>
>Examination of the map file produced by ld, shows that the ctors
>with default priority are run before the init_priority ctors:
>
>                0x004015a0                ___CTOR_LIST__ = .
>                0x004015a0                __CTOR_LIST__ = .
>                0x004015a0        0x4 LONG 0xffffffff
> *(SORT(.ctors.*))
> .ctors.65433   0x004015a4        0x4 D:\TEMP/ccAhaaaa.o
> .ctors.65434   0x004015a8        0x4 D:\TEMP/ccAhaaaa.o
> *(.ctors)
> .ctors         0x004015ac        0x4 D:\TEMP/ccAhaaaa.o
> *(.ctor)
>                0x004015b0        0x4 LONG 0x0
>
>
>
>The following patch to ld/scriptempl/pe.sc fixes the above testcase
>as well as g++.dg/special/initp1.C.  Is this the right thing to do?

Looks ok to me.  Please check in.

cgf