Global constructor problem on mips-sgi-irix5.2

Global constructor problem on mips-sgi-irix5.2

[Fred Richardson]

Hi-

It looks like constructors for global objects aren't called on my
mips-sgi-irix5.2 box when using egcs-1.0.2.

The following program:

    #include <iostream.h>
    #include <stdlib.h>
    #include <stdio.h>

    class Foo
    {
    private:
        int     a_int;
        float   a_flt;
    public:
        Foo () :
            a_int(1234), a_flt(56.78) {};
        void print()
        {
            fprintf(stdout, "Foo: a_int=%d, a_flt=%g\n",
                   a_int, a_flt);
            fflush(stdout);
        }
        void set(int _a_int, float _a_flt)
        {
            a_int = _a_int;
            a_flt = _a_flt;
        }

    };

    Foo the_foo;

    int main(int argc, char* argv[])
    {
        the_foo.print();
        the_foo.set(4321, 87.65);
        the_foo.print();
    }

compiled the following way:

    % g++ -v -Wall -o main -v main.cc
    Reading specs from /d4mutl/bybrad/lib/gcc-lib/mips-sgi-irix5.2/egcs-2.90.27/specs
    gcc version egcs-2.90.27 980315 (egcs-1.0.2 release)
     /d4mutl/bybrad/lib/gcc-lib/mips-sgi-irix5.2/egcs-2.90.27/cpp -lang-c++ -v -undef -D__GNUC__=2 -D__GNUG__=2 -D__cplusplus -D__GNUC_MINOR__=90 -Dunix -Dmips -Dsgi -Dhost_mips -DMIPSEB -D_MIPSEB -DSYSTYPE_SVR4 -D_SVR4_SOURCE -D_MODERN_C -D__DSO__ -D_MIPS_SIM=_MIPS_SIM_ABI32 -D_MIPS_SZPTR=32 -D__unix__ -D__mips__ -D__sgi__ -D__host_mips__ -D__MIPSEB__ -D_MIPSEB -D__SYSTYPE_SVR4__ -D_SVR4_SOURCE -D_MODERN_C -D__DSO__ -D_MIPS_SIM=_MIPS_SIM_ABI32 -D_MIPS_SZPTR=32 -D__unix -D__mips -D__sgi -D__host_mips -D__MIPSEB -D__SYSTYPE_SVR4 -Asystem(unix) -Asystem(svr4) -Acpu(mips) -Amachine(sgi) -D__EXCEPTIONS -D__CHAR_UNSIGNED__ -Wall -D__LANGUAGE_C_PLUS_PLUS -D_LANGUAGE_C_PLUS_PLUS -D__SIZE_TYPE__=unsigned int -D__PTRDIFF_TYPE__=int -D__EXTENSIONS__ -D_SGI_SOURCE -D_LONGLONG -D_MIPS_FPSET=16 -D_MIPS_ISA=_MIPS_ISA_MIPS1 -D_MIPS_SZINT=32 -D_MIPS_SZLONG=32 main.cc /var/tmp/cca0040A.ii
    GNU CPP version egcs-2.90.27 980315 (egcs-1.0.2 release) [AL 1.1, MM 40] SGI running IRIX 5.x
    #include "..." search starts here:
    #include <...> search starts here:
     /d4mutl/bybrad/include/g++
     /usr/local/include
     /d4mutl/bybrad/mips-sgi-irix5.2/include
     /d4mutl/bybrad/lib/gcc-lib/mips-sgi-irix5.2/egcs-2.90.27/include
     /usr/include
    End of search list.
     /d4mutl/bybrad/lib/gcc-lib/mips-sgi-irix5.2/egcs-2.90.27/cc1plus /var/tmp/cca0040A.ii -quiet -dumpbase main.cc -Wall -version -o /var/tmp/cca0040A.s
    GNU C++ version egcs-2.90.27 980315 (egcs-1.0.2 release) (mips-sgi-irix5.2) compiled by GNU C version 2.7.2.1.
     /d4mutl/bybrad/mips-sgi-irix5.2/bin/as -v -o /var/tmp/cca0040A1.o /var/tmp/cca0040A.s
    GNU assembler version 980320 (mips-sgi-irix5.2), using BFD version 2.8.1.0.24
     /d4mutl/bybrad/lib/gcc-lib/mips-sgi-irix5.2/egcs-2.90.27/ld -call_shared -no_unresolved -_SYSTYPE_SVR4 -o main /usr/lib/crt1.o -L/d4mutl/bybrad/lib/gcc-lib/mips-sgi-irix5.2/egcs-2.90.27 -L/d4mutl/bybrad/mips-sgi-irix5.2/lib -L/d4mutl/bybrad/lib /var/tmp/cca0040A1.o -lstdc++ -lm -lgcc -lc -lgcc /usr/lib/crtn.o
    %

Gives the following output:

    % main
    Foo: a_int=0, a_flt=0
    Foo: a_int=4321, a_flt=87.65
    %

Re: Global constructor problem on mips-sgi-irix5.2

[Fred Richardson]

>>         It looks like constructors for global objects aren't called on my
>>         mips-sgi-irix5.2 box when using egcs-1.0.2.
>> 
>> Yes, there is something wrong, but it is not clear what is wrong.

I found the problem.  Sorry I didn't report back!  It turns out that
I was specifying " --with-gnu-ld" on the SGI's even though GNU LD is
not supported here (although GNU AS is).  So, getting rid of this flag
to configure fixed everything for me.

                    -Fred

Re: Global constructor problem on mips-sgi-irix5.2

[Jim Wilson]

	It looks like constructors for global objects aren't called on my
	mips-sgi-irix5.2 box when using egcs-1.0.2.

Yes, there is something wrong, but it is not clear what is wrong.

	GNU assembler version 980320 (mips-sgi-irix5.2), using BFD version 2.8.1.0.24

I see that you are using a funny version of binutils.  I think that is one
of H.J. Lu's linux binutils releases.  Perhaps it doesn't work right under
Irix 5.  It was probably never tested there.  Try using binutils-2.8.1 instead.

I am guessing that there might be some confusion over whether pointers are
32 bit or 64 bits in the ctor/dtor handling code in binutils.  If binutils
puts a 64 bit value in the ctor list, and then gcc's ctor code read a 32 bit
value, it would see a zero and think that there were no ctors.  If this is
the problem, then you will have to rebuild all of egcs with a different
binutils.

Jim

Re: Global constructor problem on mips-sgi-irix5.2

[H.J. Lu]

> 
> 	It looks like constructors for global objects aren't called on my
> 	mips-sgi-irix5.2 box when using egcs-1.0.2.
> 
> Yes, there is something wrong, but it is not clear what is wrong.
> 
> 	GNU assembler version 980320 (mips-sgi-irix5.2), using BFD version 2.8.1.0.24
> 
> I see that you are using a funny version of binutils.  I think that is one
> of H.J. Lu's linux binutils releases.  Perhaps it doesn't work right under
> Irix 5.  It was probably never tested there.  Try using binutils-2.8.1 instead.
> 

FYI, my binutils is binutils snapshot + ARM + some x86 changes. I
don't think they should affect mips/irix. Try binutils 2.9 prereleae
instead. Quite a few bugs have been fixed since 2.8.1.


H.J.

Re: Global constructor problem on mips-sgi-irix5.2

[Jim Wilson]

	FYI, my binutils is binutils snapshot + ARM + some x86 changes.

That begs the question as to whether the binutils snapshots have been tested
on a mips/irix5 machine anytime recently though.

	I don't think they should affect mips/irix.

FYI mips/irix is ambiguous.  irix4 is different from irix5 is different from
irix6.

Jim

Re: Global constructor problem on mips-sgi-irix5.2

[Alexandre Oliva]

H.J. Lu wrote:

> 	FYI, my binutils is binutils snapshot + ARM + some x86 changes.

Jim Wilson writes:

> That begs the question as to whether the binutils snapshots have been tested
> on a mips/irix5 machine anytime recently though.

I've been successfully running binutils-2.8.1.0.18 on IRIX 5.2.

-- 
Alexandre Oliva
mailto:oliva@dcc.unicamp.br mailto:aoliva@acm.org
http://www.dcc.unicamp.br/~oliva
Universidade Estadual de Campinas, SP, Brasil