GCC 2.95.1 static string won't compile on SCO-OSR5

GCC 2.95.1 static string won't compile on SCO-OSR5

[+ +]

Hellow,

I'm trying to compile the following example: 

----------
#include <string>

class ctest
{
public:
        static string   foo;
};

string  ctest::foo;

void main()
{
        ctest::foo = "bar";
}
----------

Which is no problem using EGCS 1.1.1 on the same machine. GCC 2.95.1 won't
do it however:

----------
[ildev2] /usr/rob> /usr/local/gcc-2.95.1/bin/g++ test.cc
/usr/local/gcc-2.95.1/lib/gcc-lib/i586-pc-
sco3.2v5.0.0/2.95.1/libstdc++.a(cstrmain.o): fatal error: symbol `__as__t12basic_string3ZcZt18string_char_traits1ZcZt24__default_alloc_tem
plate2b0i0PCc` multiply-defined, also in file /usr/tmp/cc4lMcjd.o
collect2: ld returned 1 exit status [ildev2] /usr/rob> ----------

I can't make any sense of this error... Does this mean my GCC install is broken (again), or is this invalid C++?

Thanks!

 Rob Kramer
 robk@cyberway.com.sg




--== Sent via Deja.com http://www.deja.com/ ==--
Share what you know. Learn what you don't.

Re: GCC 2.95.1 static string won't compile on SCO-OSR5

[Robert Lipe]

This is a known (sigh) problem with current g++ and essentially all
the System V linkers.  THey differ in their interpretation of .weak.
The solution on most systems is to use GNU ld.  Unfortunately, GNU ld
doesn't support OSR5 well enough to be an acceptable substutute.

But the good news is that other than templates, things work OK. :-(

RJL


> Which is no problem using EGCS 1.1.1 on the same machine. GCC 2.95.1 won't
> do it however:
> 
> ----------
> [ildev2] /usr/rob> /usr/local/gcc-2.95.1/bin/g++ test.cc
> /usr/local/gcc-2.95.1/lib/gcc-lib/i586-pc-
> sco3.2v5.0.0/2.95.1/libstdc++.a(cstrmain.o): fatal error: symbol `__as__t12basic_string3ZcZt18string_char_traits1ZcZt24__default_alloc_tem
> plate2b0i0PCc` multiply-defined, also in file /usr/tmp/cc4lMcjd.o
> collect2: ld returned 1 exit status [ildev2] /usr/rob> ----------

Re: GCC 2.95.1 static string won't compile on SCO-OSR5

[Robert Lipe]

This is a known (sigh) problem with current g++ and essentially all
the System V linkers.  THey differ in their interpretation of .weak.
The solution on most systems is to use GNU ld.  Unfortunately, GNU ld
doesn't support OSR5 well enough to be an acceptable substutute.

But the good news is that other than templates, things work OK. :-(

RJL


> Which is no problem using EGCS 1.1.1 on the same machine. GCC 2.95.1 won't
> do it however:
> 
> ----------
> [ildev2] /usr/rob> /usr/local/gcc-2.95.1/bin/g++ test.cc
> /usr/local/gcc-2.95.1/lib/gcc-lib/i586-pc-
> sco3.2v5.0.0/2.95.1/libstdc++.a(cstrmain.o): fatal error: symbol `__as__t12basic_string3ZcZt18string_char_traits1ZcZt24__default_alloc_tem
> plate2b0i0PCc` multiply-defined, also in file /usr/tmp/cc4lMcjd.o
> collect2: ld returned 1 exit status [ildev2] /usr/rob> ----------

GCC 2.95.1 static string won't compile on SCO-OSR5

[+ +]

Hellow,

I'm trying to compile the following example: 

----------
#include <string>

class ctest
{
public:
        static string   foo;
};

string  ctest::foo;

void main()
{
        ctest::foo = "bar";
}
----------

Which is no problem using EGCS 1.1.1 on the same machine. GCC 2.95.1 won't
do it however:

----------
[ildev2] /usr/rob> /usr/local/gcc-2.95.1/bin/g++ test.cc
/usr/local/gcc-2.95.1/lib/gcc-lib/i586-pc-
sco3.2v5.0.0/2.95.1/libstdc++.a(cstrmain.o): fatal error: symbol `__as__t12basic_string3ZcZt18string_char_traits1ZcZt24__default_alloc_tem
plate2b0i0PCc` multiply-defined, also in file /usr/tmp/cc4lMcjd.o
collect2: ld returned 1 exit status [ildev2] /usr/rob> ----------

I can't make any sense of this error... Does this mean my GCC install is broken (again), or is this invalid C++?

Thanks!

 Rob Kramer
 robk@cyberway.com.sg




--== Sent via Deja.com http://www.deja.com/ ==--
Share what you know. Learn what you don't.

Re: GCC 2.95.1 static string won't compile on SCO-OSR5

[+ +]

>This is a known (sigh) problem with current g++ and 
>essentially all the System V linkers.  

>But the good news is that other than templates, things 
>work OK. :-(

Hmm. Oh well, at least if I know it won't work, I can work around it. But could you tell me what exactly I can't do? A static 'list <int>' or whatever seems to work fine. What's special about a string?

Thanks!

    Rob Kramer
    robk@cyberway.com.sg



--== Sent via Deja.com http://www.deja.com/ ==--
Share what you know. Learn what you don't.

Re: GCC 2.95.1 static string won't compile on SCO-OSR5

[Robert Lipe]

> >This is a known (sigh) problem with current g++ and essentially all  
> >the System V linkers.                                                
>
> >But the good news is that other than templates, things work OK. :-(
>
> Hmm. Oh well, at least if I know it won't work, I can work around it.
> But could you tell me what exactly I can't do? A static 'list <int>'
> or whatever seems to work fine. What's special about a string?

I can't tell you at a C++ level (I speak only kindergarten C++) exactly
what triggers this though it seems to be template related. At the linker
level, the conflict is that the System V linkers don't allow a strong
definition to follow a weak definition of a symbol in archives and the
G++ template logic depends on the GNU ld behaviour of allowing this.

There are threads in the archives describing this in painful detail.

Re: GCC 2.95.1 static string won't compile on SCO-OSR5

[Robert Lipe]

> >This is a known (sigh) problem with current g++ and essentially all  
> >the System V linkers.                                                
>
> >But the good news is that other than templates, things work OK. :-(
>
> Hmm. Oh well, at least if I know it won't work, I can work around it.
> But could you tell me what exactly I can't do? A static 'list <int>'
> or whatever seems to work fine. What's special about a string?

I can't tell you at a C++ level (I speak only kindergarten C++) exactly
what triggers this though it seems to be template related. At the linker
level, the conflict is that the System V linkers don't allow a strong
definition to follow a weak definition of a symbol in archives and the
G++ template logic depends on the GNU ld behaviour of allowing this.

There are threads in the archives describing this in painful detail.

Re: GCC 2.95.1 static string won't compile on SCO-OSR5

[+ +]

>This is a known (sigh) problem with current g++ and 
>essentially all the System V linkers.  

>But the good news is that other than templates, things 
>work OK. :-(

Hmm. Oh well, at least if I know it won't work, I can work around it. But could you tell me what exactly I can't do? A static 'list <int>' or whatever seems to work fine. What's special about a string?

Thanks!

    Rob Kramer
    robk@cyberway.com.sg



--== Sent via Deja.com http://www.deja.com/ ==--
Share what you know. Learn what you don't.