$Source: /usr/local/cvsroot/unixlib/Docs/Changes,v $
$Date: 1997/09/19 21:43:31 $
$Revision: 1.2 $
$State: Exp $

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

CL0039 15 September 1997
------------------------

More bug fixes rather than anything new :-(

Nick B's new _kernel_* functions had some return codes wrong or lost the
_kernel_oserror whenit was needed as the return value. _kernel_getenv
and _kernel_setenv have been re-written to meet their specification.
This is good, because the SCL's versions don't.

Minor changes to stdio/setbuf.c and sys/exec.c.

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

CL0037 14 September 1997
------------------------

I've modified Nick Clark's last change which introduced Stewart Brodie's
SWI veneer for socket code. I would like to know exactly what Stewart
has to say here regarding which versions of Acorn's stack go horribly
wrong and under what circumstances.

Anyway, the new macro NetSWI replaces CallSWI because it is only of use
for the networking code and not a general SWI wrapper which CallSWI
might have suggested. Further, it is slightly more efficient and
preserves all flags from the socket swi. Anyone know why Stewart put a
no-op after SWI XOS_EnterOS ?

I've also tidied up net error processing by using __net_error which now
saves the kernel_oserror too.

Finally, setting FIOSLEEPTW on a socket is done in socket.c where it
allows _socket.s to become much tidier. As you can see if you look at
the comment there, I'm not convinced this should be automatically
enabled for all new sockets. I suspect Nick Clark added it, but Stewart
Brodie says in c.s.a.programmer

   "It is definitely not a good idea to do this, since some of the
networking SWI calls made by the taskwindow'ed application may not be
re-entract.  NFS SWIs are a particular problem here."

Comments ?

