tcpstream.getline() problem on Solaris

[Costas Kotsokalis]

Heya,
  working with the CVS version of CC++ on Solaris 2.7/Sparc, g++ 2.95.1.
The following code will work fine on Linux; however, on Solaris as soon as a connection is made the program receives a SIGILL and dies:

#include <cc++/socket.h>
#include <cc++/thread.h>

#ifdef  CCXX_NAMESPACES
using namespace std;
using namespace ost;
#endif

#include <iostream>

int main(int argc, char *argv[]){
    char *line = (char *)malloc(200);

    InetAddress addr = "0";
    TCPSocket *sock = new TCPSocket(addr, 9000);
    while (1){
        if (sock->isPendingConnection()){
            tcpstream tcp;
            tcp.open(*sock);
            tcp.getline(line, 200);
            cout << line << endl;
            tcp.close();
        }
    }
}

The socket is not closed before getline() is called, I have checked that. It is still open, yet the call to getline() fails. I tried to use gdb having compiled CC++ with debug info but I only managed to find that the problem takes place when TCPStream::underflow is called. It's weird though that gdb could not step into underflow(), it could only get to the point that it's called.

Any ideas? Thanks!

  Costas

[Frediano Ziglio]

Very strange problem...
I added to all TCPStream methods a line such
  cout << "funcname" << endl;
And TCPStream::underflow is NOT called...
Is seem a library bug...
If you change che calling to underflow in TCPStream::uflow from underflow to TCPStream::underflow it run correctly (it exec underflow)...
I watch header but the code is strange. It use a _IO_FILE C structure and some static C function like __underflow...
Perhaps it's a Gcc 2.95.1/2.95.2 library problem ???

Why the behavior change calling underflow from a TCPStream and TCPStream::underflow ?? Perhaps the compiler ?? (streambuf::underflow is public and TCPStream::underflow protected as standard?). I tried to remove virtual from TCPStream::underflow declaration but nothing change...

freddy77

[Frediano Ziglio]

Fixed in CVS..
Many hours are gone...
This is a g++ bug (should be libio). libio (a library used for FILE* and streambuf I/O) is a C library that use C++ virutal table making some assumption. Perhaps these assumption are not very correct for some platform...
streambuf MUST BE the first superclass on those version...