Re: lossage from b... functions change

[Dave Love]

Richard Stallman <address@hidden> writes:

> I can't access the repository from here.  Can you show me the change,
> and the change log entry?  With that, I can search my old mail
> to see why it was made.

I guess this was it.

2003-03-03  Rob Browning  <address@hidden>

        * configure.in: add AC_CHECK_FUNCS for memcmp and memmove.
        (AH_BOTTOM): switch tests to prefer memcmp, memset, and memmove
        over bzero, bcmp, and bcopy.

Index: configure.in
===================================================================
RCS file: /cvsroot/emacs/emacs/configure.in,v
retrieving revision 1.327
retrieving revision 1.328
diff -u -p -c -r1.327 -r1.328
cvs server: conflicting specifications of output style
*** configure.in        21 Feb 2003 08:07:29 -0000      1.327
--- configure.in        3 Mar 2003 18:29:42 -0000       1.328
*************** strerror fpathconf select mktime euidacc
*** 2209,2215 ****
  utimes setrlimit setpgid getcwd getwd shutdown strftime getaddrinfo \
  __fpending mblen mbrlen mbsinit strsignal setitimer ualarm index rindex \
  sendto recvfrom getsockopt setsockopt getsockname getpeername \
! gai_strerror mkstemp getline getdelim mremap memmove fsync bzero memset)
  
  AC_CHECK_HEADERS(sys/un.h)
  
--- 2209,2216 ----
  utimes setrlimit setpgid getcwd getwd shutdown strftime getaddrinfo \
  __fpending mblen mbrlen mbsinit strsignal setitimer ualarm index rindex \
  sendto recvfrom getsockopt setsockopt getsockname getpeername \
! gai_strerror mkstemp getline getdelim mremap memmove fsync bzero \
! memset memcmp memmove)
  
  AC_CHECK_HEADERS(sys/un.h)
  
*************** typedef unsigned size_t;
*** 2826,2839 ****
  #  endif
  #endif
  
! #ifndef HAVE_BCOPY
! #define bcopy(a,b,s) memcpy (b,a,s)
! #endif
! #ifndef HAVE_BZERO
! #define bzero(a,s) memset (a,0,s)
! #endif
! #ifndef HAVE_BCMP
! #define BCMP memcmp
  #endif
  
  #endif /* EMACS_CONFIG_H */
--- 2827,2842 ----
  #  endif
  #endif
  
! /* avoid deprecated functions */
! #ifdef HAVE_MEMCPY
! #  define bcopy(a,b,s) memcpy (b,a,s)
! #endif
! #ifdef HAVE_MEMSET
! #  define bzero(a,s) memset (a,0,s)
! #endif
! #ifdef HAVE_MEMCMP
! #  define BCMP memcmp
! #  define bcmp memcmp
  #endif
  
  #endif /* EMACS_CONFIG_H */

>                                             as I said before, mapping
>     bcopy to memcpy is dubious.
>
> I missed that.  What is the issue?

Here's the message.

> --- Begin Message ---
>
>
>
> Subject: 
>
> b... functions in configure.in
>
>
>
>
> Date: 
>
> Sun, 18 May 2003 21:20:56 +0100
>
>
>
>
> User-agent: 
>
> Gnus/5.1002 (Gnus v5.10.2) Emacs/21.2 (gnu/linux)
>
>
>
>
> I don't understand the purpose of the following fragment in
> configure.in, and I'm not convinced it's correct.
>
> What's the point of defining bcopy away if you have it, especially at
> the expense of confusion when debugging?  What about the case now
> removed for the system not having bcopy or memmove?  (If that's
> irrelevant now, stuff in sysdep.c can be cleaned up.)  Is it clear
> that bcopy is never used with overlapping args, and what's to prevent
> someone doing that in future?  If it's always used with
> non-overlapping args and you want to outlaw it, I think it should just
> be replaced globally with memcpy (which has the advantage of being
> inlined by gcc) to avoid confusion.  What's BCMP for, which doesn't
> seem to be used anywhere.
>
> /* avoid deprecated functions */
> #ifdef HAVE_MEMCPY
> #  define bcopy(a,b,s) memcpy (b,a,s)
> #endif
> #ifdef HAVE_MEMSET
> #  define bzero(a,s) memset (a,0,s)
> #endif
> #ifdef HAVE_MEMCMP
> #  define BCMP memcmp
> #  define bcmp memcmp
> #endif
>
> --- End Message ---
>       I previously cleaned up the
>     treatment of bcopy & al so that the configuration should have worked
>     generally.
>
> Which cleanup was that?  If you could show me the change log entry for
> that, it would help me understand the situation.  (The date is also
> relevant.)

I guess it was these, avoiding previous problems on solaris and irix
which required bcopy & al to be undefed explicitly.

2002-11-15  Dave Love  <address@hidden>

        * configure.in: Tidy up various quoting issues throughout.  Use
        AC_GNU_SOURCE.
        (AH_BOTTOM): Fix #endif protecting config.h.  Maybe include
        alloca.h.  Define GC_SETJMP_WORKS, GC_LISP_OBJECT_ALIGNMENT.
        Maybe define bcopy, bzero, bcmp.

2002-10-30  Dave Love  <address@hidden>

        * configure.in:  Require autoconf 2.53.  Test for pty.h,
        sys/mman.h, sys/param.h, mremap, memmove.
        (AH_BOTTOM): Maybe include strings.h.  Add local variables for
        mode.

along with changes in src/s at 2002-10-30.  The changes avoided having
to do anything special in the config files.  I noticed this change

2003-05-24  Andreas Schwab  <address@hidden>

        * configure.in (AH_BOTTOM) [MAC_OSX]: Do not redefine bcopy,
        bzero, and bcmp.

which implies there's a problem with more current systems than sunos4.
That's the sort of thing that I removed from the src/s files in the
cleanup above.