SPREAD: A Reliable Multicast and Group Communication Toolkit
-----------------------------------------------------------

PORTING GUIDE:
--------------

This is definitely a work in progress, so email me any comments you have 
at jonathan@cs.jhu.edu.

Basically Spread requires a BSD sockets networking API, and a few fairly 
standard Posix API calls. It was some work to port to Windows, but not 
much in comparison with porting graphical software.

Each supported architecture has an entry in the arch.h file which defines
all necessary info about each architecture. The arch.h file has three major
sections. First there is detection code which detects what platform is 
currently being built. Second there is the setting of capability flags 
depending on the built architecture. Third there are some general definitions
used by all the architectures. For example, the Linux entry looks like this:

/* detect linux */
#ifdef __linux__
#define ARCH_PC_LINUX
#endif

/* Set flags for linux */
#ifdef ARCH_PC_LINUX
#define         INTSIZE32
#define		ARCH_SCATTER_CONTROL /* should be control if supported */
#define		ARCH_ENDIAN	0x80000080
#define         LOC_INLINE      __inline__
#include        <sys/uio.h>
#define         ARCH_SCATTER_SIZE       UIO_MAXIOV
#define         HAVE_GOOD_VARGS
typedef         int     sockopt_len_t;
#endif /* ARCH_PC_LINUX */

Essentially all that is required is some knowledge about the sizes of the
basic C types int and short, what endianness the machine is, and whether
{recv/send}msg calls use the POSIX msg_control structure or the Solaris like
acc_rights structure. For endianness, if a machine is little-endian then
the ARCH_ENDIAN should be set to 0x80000080, if it is big-endian then 
ARCH_ENDIAN should be 0x00000000. The #include is used to grab the definition
of how large scatters (iovecs) are for each architecture. Since each one stores
the value in a different variable and in a different include file, we have
to do this per architecture. If your platform has <stdarg.h> and a working 
implementation of variable arguments (va_list, va_start, va_end) then 
define HAVE_GOOD_VARGS, otherwise we use a different method that works 
everywhere.

The makefile for each architecture are very similar. The main differences
occur if we can't use gcc on that platform (like IRIX) or if additional
libraries are required for sockets programs (like Solaris)