On Wed, Jul 04, 2001 at 10:03:23AM +0900, minoura@netbsd.org wrote:
> 
> >Number:         13369
> >Category:       kern
> >Synopsis:       gcc inlines memcpy(3)
> >Confidential:   no
> >Severity:       serious
> >Priority:       high
> >Responsible:    kern-bug-people
> >State:          open
> >Class:          sw-bug
> >Submitter-Id:   net
> >Arrival-Date:   Tue Jul 03 18:01:00 PDT 2001
> >Closed-Date:
> >Last-Modified:
> >Originator:     Minoura Makoto
> >Release:        20010612
> >Organization:
> Minoura Makoto <minoura@netbsd.org>
> The NetBSD Project/Japan NetBSD Users' Group
> >Environment:
> System: NetBSD daisy 1.5W NetBSD 1.5W (DAISY) #190: Mon Jun 25 14:28:07 JST 2001 root@daisy:/usr/obj/sys/arch/i386/compile/DAISY i386
> gcc version egcs-2.91.66 19990314 (egcs-1.1.2 release)
> Architecture: i386
> Machine: i386
> >Description:
> 	gcc -O inlines the following memcpy(3) call (at least on i386):
> 
> #include <string.h>
> 
> extern char buf[];
> 
> void testfunc(char *src)
> {
> 	memcpy(buf, src, 100);
> }
> 
> 	The inlined memcpy() gets insane when the two buffers are
> 	overlapping.
> 
> 	On the other hand, kernel bcopy() is #define'ed as memcpy() in
> 	/sys/sys/systm.h.
> 
> 	This might cause a problem that when the programmer
> 	wants to transfer overlapping strings using bcopy()
> 	they might be destroyed as a result.

Shouldn't we just replace calls to bcopy with memmove or memcpy, as 
appropriate ?

--
Manuel Bouyer <bouyer@antioche.eu.org>
--