From:	Jason Nye <jnye AT nbnet DOT nb DOT ca>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Solution(?) to STRANGE PROBLEM
Date:	Sat, 18 Oct 1997 21:20:24 -0300
Organization:	NBTel Internet
Lines:	141
Message-ID:	<34495247.8960CC82@nbnet.nb.ca>
NNTP-Posting-Host:	mctnts05c13.nbnet.nb.ca
Mime-Version:	1.0
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

WARNING!!! Long message ahead, I repeat, WARNING!!! Long message ahead!


I have figured out what was going on with my SIGSEGV!!!  It wasn't the
protected mode startup code (poetry).
I didn't think it was because all my other pmode programs still worked
fine while this one didn't.  I would
appreciate it if some would take the time to read this -- especially DJ
Delorie if he could...


My Configuration:
CPU:   P166MMX, 40MB of RAM
DPMI Server:  CWSDPMI.EXE
Compiler version: gcc 2.7.2.1

Summary:

I created a class, say A, which had a helper class, say B.  B was
basically a singleton (except its
members were not static), which means that there would only ever be one
instance of this class.  This
class, B, would be hidden to the user of class A, therefore, I decided
to put it in the cpp file.  I also
instantiated the class in the cpp file like so:


#include <A.hpp>

//
// B's interface is of no interest to the user of A
class B {
public:
    B(int size);
    ...
};

//
// B's functions...
B::B(int size)
{
    ...
}

... etc


//
// Create a static instance of B.
static B workerB(100);


////////////////////////////////////

//
// A's interface is defined in A.hpp
A::A(char c)
{
    //
    // Start using workerB as a helper here...
}

---------------------------------------------------------------------

Now, I would think that this would work absolutely fine, but what
actually happens is once the execution has gone past the construction of
workerB, all Hell breaks loose.  This, apparently is a bug and a damn
hard to find bug at that.  I also tried declaring workerB without the
static storage class and it crashed that way too.

This is how I got around the bug:

-----------------------------------------------------------------------------------------------------------

//
// Header:
#ifndef A_HPP
#define A_HPP

class B;        // Create a forward declaration of class B (compiler
only needs
                    // to know that there exists a class B since we are
only declaring
                    // a pointer to a B -- works if we are declaring a
reference also).

class A {
public:
    A::A(char c);
private:
    B *workerB;
    static bool setup;
};

#endif
-----------------------------------------------------------------------------------------------------------



-----------------------------------------------------------------------------------------------------------

//
// Cpp:
#include <A.hpp>

bool A::setup = false;

//
// B's declaration as above...
//

A::A(char c)
{
    if (!setup) {
        workerB = new B(100);
        setup = true;
    }
}
----------------------------------------------------------------------

This works with no crashes.  Why did the other way not work?  I have no
Idea.  Maybe someone could help me out on why the static initialization
did not work.

Here are my compile options, maybe they are wrong (I am using a P166MMX
with 40MB of RAM):

//
// I wanted to optimize my code for a Pentium (I saw these options in
the Linux kernel makefile except
// Linus used =2 instead of =4 would that cause a problem?).  I didn't
want any optimizations yet (-Ox)
// and I didn't want to compile with debug code (-g), I just use a lot
of print statements when I am
// debugging a class.  I find it easier (personal preference).
CPU = -m486 -malign-loops=4 -malign-jumps=4 -malign-functions=4
OFLAGS = -I$(HPPDIR) -c -Wall -ffor-scope $(CPU)

Any input would be greatly appreciated, thanks for taking time to read
this :).

- Raw text -