Rationale for inet_ntop(3) using socklen_t instead of size_t

[Alejandro Colomar]

Hi Jakub,

inet_ntop(3):

  SYNOPSIS
        #include <arpa/inet.h>

        const char *inet_ntop(int af, const void *restrict src,
                              char dst[restrict .size], socklen_t size);

  STANDARDS
        POSIX.1‐2001, POSIX.1‐2008.  Note that  RFC 2553  defines  a  prototype
        where  the  last  argument size is of type size_t.  Many systems follow
        RFC 2553.  Glibc 2.0 and 2.1  have  size_t,  but  2.2  and  later  have
        socklen_t.


Glibc did the change here (with no rationale):

  commit a1303dc8dc57dbcd4605992dc177d11ae4d45aab
  Author: Ulrich Drepper <drepper@redhat.com>
  Date:   Thu Apr 6 05:04:50 2000 +0000

     Update.

     2000-04-05  Cristian Gafton  <gafton@redhat.com>

             * ctype/ctype.h (__isctype_l): Define even if not __OPTIMIZE__
             because it is used in generic code.
             * wctype/wctrans.c: Define __ctype32_toupper and __ctype32_tolower.
             * wcsmbs/wcwidth.h: Declare __ctype32_b.
             * intl/gettext.c: Include <locale.h>.
             * intl/ngettext.c: Likewise.

     2000-04-05  Jakub Jelinek  <jakub@redhat.com>

             * resolv/inet_ntop.c (inet_ntop4, inet_ntop6, inet_ntop): Use
             socklen_t type for size.
             * sysdeps/sparc/sparc32/dl-machine.h (RTLD_START): Rewrite for new
             init function interface.
             * sysdeps/sparc/sparc64/dl-machine.h (RTLD_START): Likewise.
             * sysdeps/unix/sysv/linux/sparc/bits/setjmp.h: Don't include
             sys/ucontext.h.
             (__jmp_buf): Declare the whole type structure here, don't use
             ucontext.

             * iconv/skeleton.c (put16u): Fix typo.
             Reported by Bruno Haible <haible@ilog.fr>.


That's a year before POSIX.1-2001.  I'd like to understand why was this change 
done.  Especially, considering that socklen_t for a string and in this context 
is *very* confusing.  I guess this is due to historic design mistakes, and 
compatibility with other systems that had it wrong?  Maybe obscure POSIX 
standards prior to 2001 required socklen_t already?

Thanks,

Alex

P.S.: AFAIK, I can't search in the glibc mailing list history; something I can 
do in the lore.kernel.org list archives.  Is it correct?  Or maybe I just don't 
know how?  <https://sourceware.org/pipermail/libc-alpha/>

-- 
<http://www.alejandro-colomar.es/>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20221222/829e2cb9/attachment.sig>