daumas@frensl61.bitnet (Marc Daumas) (04/08/91)
I need HELP, I have been developping a program and I found that the two
programs below are not equivalents.
The first one causes a Segmentation fault with GNU C++ compiler
on Sparc Sation and the second one works normaly.
PLEASE, COULD SOMEONE EXPLAIN IT TO ME
Please reply to me DIRECTLY,
daumas@frensl61.bitnet
I'll sum up for the net if this seems interesting.
/-/-/-/-/-/-/-/-/-/-/ prog 1
#include <stream.h>
class Classe
{
public:
int classe;
Classe (int);
};
Classe :: Classe (int c)
{ classe = c; cout << "toto\n"; }
Classe *classe = new Classe (6);
main ()
{
}
/-/-/-/-/-/-/-/-/-/-/ prog 2
#include <stream.h>
class Classe
{
public:
int classe;
Classe (int);
};
Classe :: Classe (int c)
{ classe = c; cout << "toto\n"; }
Classe *classe;
main ()
{ classe = new Classe (6);
}
-- Marc Daumasdavew@panache.cs.umd.edu (David G. Wonnacott) (04/09/91)
In article <1991Apr8.010055.7672@lip.ens-lyon.fr> daumas@frensl61.bitnet (Marc Daumas) writes: >I need HELP, I have been developping a program and I found that the two >programs below are not equivalents. > >The first one causes a Segmentation fault with GNU C++ compiler >on Sparc Sation and the second one works normaly. > >/-/-/-/-/-/-/-/-/-/-/ prog 1 > >#include <stream.h> > >Classe *classe = new Classe (6); > >main () >{ >} >/-/-/-/-/-/-/-/-/-/-/ prog 2 > >#include <stream.h> > >Classe *classe; > >main () >{ classe = new Classe (6); >} > The problem you are having occurs because the initialzation for one of your global variables (classe) depends on the value of another global (cout). All global initializations occur before the first statement of main. In this case, classe is getting initialized before cout. Thus, the classe constructor is trying to do cout << "toto\n" before cout has been initialized, which causes a core dump. The "backtrace" command in gdb shows some of what happens during global initialization: (gdb) run Starting program: /panache/davew/tests/one-orig Program received signal 11, Segmentation fault Reading in symbols for streambuf.cc...done. sputs__9streambufPCc ($this=(streambuf *) 0x0, s=(char *) 0x20a8 "toto\n") (streambuf.cc line 92) streambuf.cc: No such file or directory. (gdb) backtrace #0 sputs__9streambufPCc ($this=(streambuf *) 0x0, s=(char *) 0x20a8 "toto\n") (streambuf.cc line 92) #1 0x20e4 in __6Classei (...) (...) #2 0x214a in _GLOBAL_$I$__6Classei (...) (...) #3 0x48c4 in __do_global_init (...) (...) #4 0x489c in __main (...) (...) #5 0x2122 in main (...) (...) (gdb) Note that main has gotten into __do_global_init, which is responsible for doing all global initialization. Note further that the call to sputs__9streambufPCc (presumably the "sputs" operation for class streambuf) has "this" = 0. So, you ask, "what can I do about it?" The first thing is to try to avoid having the initialization of one global depend on the value of another global variable (i.e., use version two). If you can not do this, you must find some way to force the initializations to occur in the right order. As I recall, the AT&T C++ system ensures that initialization of globals goes in order within each source file, but makes no guarantees about variables in different files (but I don't have my A.R.M. here to check -- please correct me if I'm wrong). The AT&T iostream library relies on this fact to ensure that cout, etc. are initialized before any global variables from files that include iostream.h (since the globals in iostream.h will be initialized before they are). I don't know whether g++ enforces the same ordering of global initialization, or whether its stream library is written the way AT&T's is. So I'm not sure if there is a way to fix it. Can any of the GNU folks shed some light on this? -- David Wonnacott davew@tove.cs.umd.edu Although the moon is smaller than the earth, it is further away.