svn commit: r235767 - head/lib/libc/net
Baptiste Daroussin
bapt at FreeBSD.org
Wed May 23 18:44:49 UTC 2012
On Tue, May 22, 2012 at 08:51:44PM +1000, Bruce Evans wrote:
> On Mon, 21 May 2012, Garrett Cooper wrote:
>
> > On Mon, May 21, 2012 at 6:28 PM, Kevin Lo <kevlo at freebsd.org> wrote:
> >> Author: kevlo
> >> Date: Tue May 22 01:28:32 2012
> >> New Revision: 235767
> >> URL: http://svn.freebsd.org/changeset/base/235767
> >>
> >> Log:
> >> Add missing header needed by free()
> >>
> >> Reported by: tinderbox
>
> Please don't use binary characters in mail.
>
> > Thank you for saving my mailbox :)!!!
> >
> > FWIW, it's weird because my STABLE-9 box didn't reproduce this.
>
> This seems to be a bug in both nsparser.y and the new yacc:
> - new yacc: it is incompatible.
>
> Old yacc includes stdlib.h and string.h as the first thing in the
> generated file, so the C code copied from the .y part sees them too.
>
> New yacc puts more its code including its all includes (which still
> involve stdlib.h) at the end of the generated file.
>
> Including stdlib.h nearly first is from Lite2 (skeleton.c 1.7 in 1997).
> Much later, FreeBSD moved it to the very first thing in the generated
> file, so as to use namespace pollution (__unused) from it early.
> __unused was ifdefed and was defined by the yacc skeleton if <stdlib.h>
> didn't define it. It took much churn to produce this, but it seems to
> have been garbage, since old yacc didn't actually generate any use of
> __unused. The definition was apparently compatibilty cruft to hide
> unportabilities in .y files that use __unused (I guess these failed for
> bootstrapping). New yacc doesn't define __unused.
>
> - nsparser.y: it uses free() but never declared it. It depended on
> namespace pollution in the yacc code, and this pollution being in a
> particular order.
>
> The yyparse() incompatibilities seem to be bugs on both sides too:
> - Old yacc didn't declare yyparse() automatically, so .y files had to do
> it, but probably shouldn't have.
>
> - New yacc still doesn't declare yylex() or yyerror() automatically, so
> applications still have to declare them.
>
> - Some of these may be more the responsibility of the yacc library (to
> supply defaults) than others. I couldn't see who is required to declare
> them in POSIX.1-2001. However, a bad example in POSIX.1 has "extern int
> yyparse();" in main() just before calling yyparse(). Badness in this
> example include several style bugs and the declaration being incomplete
> (missing "void").
> Old yacc has about 20 lines of ifdefs half just to supply this "void"
> for STDC but not for K&R. New yacc no longer supports K&R, at least
> here, but it still has 11 lines of ifdefs to get the declaration of
> yyparse() right for bison compatibility).
> I don't know if the badness includes having the home made declaration.
>
> Nearby bugs: all CSRG and FreeBSD changes including copyrights and
> history seem to have been lost. Just after the old include of stdlib.h,
> skeleton.c used to produce a string yyrcsid[] with __FBSDID() for
> skeleton.c itself in it. New yacc generates a string yysccsid with
> a hard-coded berkeley 1993 id in it, and the name in this id doesn't
> even match the file name (it is "yaccpar"). Another FreeBSD change was
> to remove this bogus id from the generated file.
>
> Some relevant FreeBSD changes:
>
> % Index: skeleton.c
> % ===================================================================
> % RCS file: /home/ncvs/src/usr.bin/yacc/skeleton.c,v
> % retrieving revision 1.1.1.2
> % retrieving revision 1.37
> % diff -u -2 -r1.1.1.2 -r1.37
> % --- skeleton.c 6 Jan 1997 03:03:46 -0000 1.1.1.2
> % +++ skeleton.c 12 Feb 2003 18:03:55 -0000 1.37
> % @@ -35,7 +35,12 @@
> % */
> %
> % +#if 0
> % #ifndef lint
> % static char sccsid[] = "@(#)skeleton.c 5.8 (Berkeley) 4/29/95";
> % -#endif /* not lint */
> % +#endif
> % +#endif
>
> FreeBSD normally comments out vendor ids like this. 5.8 is from Lite2;
> Lite1 had 5.7. New yacc doesn't have any of this. It apparently started
> from older yacc (before Lite1; indeed, it agrees with the FreeBSD-1 == Net/2
> version in not having any CSRG copyrights or this sccsid).
>
> % +
> % +#include <sys/cdefs.h>
> % +__FBSDID("$FreeBSD: src/usr.bin/yacc/skeleton.c,v 1.37 2003/02/12 18:03:55 davidc Exp $");
>
> FreeBSD normally adds this to source files. Except in contrib. yacc is
> now mostly in contrib, although it hasn't changed all that much.
>
> %
> % #include "defs.h"
> % @@ -53,31 +58,42 @@
> % /* are conditional. */
> %
> % -char *banner[] =
> % +const char *banner[] =
>
> New yacc has this change too.
>
> % {
> % + "#include <stdlib.h>",
>
> See above.
>
> % "#ifndef lint",
> % - "static char yysccsid[] = \"@(#)yaccpar 1.9 (Berkeley) 02/21/93\";",
>
> Was in Net/2 (?), FreeBSD-1, Lite1 and Lite2. Came back with new yacc.
> It's interesting that "yaccpar 1.9 (Berkeley) 1.9 02/21/93" is only 3 months
> before the correct id "skeleton.c 5.7 (Berkeley) 5/24/93".
>
> Hmm, the FreeBSD-1 history is even more instructive than I first
> thought. FreeBSD-1 started with the CSRG version of skeleton.c on
> 1993/06/12. This had CSRG copyrights and ids, and yaccpar was at 1.8
> 01/20/90. But this was upgraded to "the newest version on
> vangogh.cs.berkeley.edu" 17 days later on 1993/06/29. This version
> has no CSRG copyrights or ids, except for the above rotted one for
> yaccpar (the 1.9 02/21/93 one). I think it was only missing CSRG
> copyrights and ids because CSRG only added these to full releases.
> This resulted in the version in FreeBSD-1 not having any copyright
> notices inside most files, and none for the directory either. The
> new byacc seems to be the same as the older yacc here too.
>
> % + "#ifdef __unused",
> % + "__unused",
> % + "#endif",
> % + "static char const ",
> % + "yyrcsid[] = \"$FreeBSD: src/usr.bin/yacc/skeleton.c,v 1.37 2003/02/12 18:03:55 davidc Exp $\";",
> % "#endif",
> % - "#include <stdlib.h>",
>
> See above.
>
> % "#define YYBYACC 1",
> % "#define YYMAJOR 1",
> % "#define YYMINOR 9",
> % - "#define yyclearin (yychar=(-1))",
> % + "#define YYLEX yylex()",
> % + "#define YYEMPTY -1",
> % + "#define yyclearin (yychar=(YYEMPTY))",
> % "#define yyerrok (yyerrflag=0)",
> % - "#define YYRECOVERING (yyerrflag!=0)",
> % + "#define YYRECOVERING() (yyerrflag!=0)",
> % + "#if defined(__cplusplus) || __STDC__",
> % + "static int yygrowstack(void);",
> % + "#else",
> % + "static int yygrowstack();",
> % + "#endif",
>
> % 0
> % };
>
> Hrmph. It was FreeBSD that added the STDC ifdefs, to avoid breaking
> support for generating K&R code (older yacc generated fairly pure K&R
> code without even "const"). For yygrowstack(), this was sloppy:
> it only changed the declaration -- the function body still doesn't
> say "void".
>
> % @@ -151,13 +174,35 @@
> % "#define YYACCEPT goto yyaccept",
> % "#define YYERROR goto yyerrlab",
> % + "",
> % + "#ifndef YYPARSE_PARAM",
> % + "#if defined(__cplusplus) || __STDC__",
> % + "#define YYPARSE_PARAM_ARG void",
> % + "#define YYPARSE_PARAM_DECL",
> % + "#else /* ! ANSI-C/C++ */",
> % + "#define YYPARSE_PARAM_ARG",
> % + "#define YYPARSE_PARAM_DECL",
> % + "#endif /* ANSI-C/C++ */",
> % + "#else /* YYPARSE_PARAM */",
> % + "#ifndef YYPARSE_PARAM_TYPE",
> % + "#define YYPARSE_PARAM_TYPE void *",
> % + "#endif",
> % + "#if defined(__cplusplus) || __STDC__",
> % + "#define YYPARSE_PARAM_ARG YYPARSE_PARAM_TYPE YYPARSE_PARAM",
> % + "#define YYPARSE_PARAM_DECL",
> % + "#else /* ! ANSI-C/C++ */",
> % + "#define YYPARSE_PARAM_ARG YYPARSE_PARAM",
> % + "#define YYPARSE_PARAM_DECL YYPARSE_PARAM_TYPE YYPARSE_PARAM;",
> % + "#endif /* ANSI-C/C++ */",
> % + "#endif /* ! YYPARSE_PARAM */",
> % + "",
> % "int",
> % - "yyparse()",
> % + "yyparse (YYPARSE_PARAM_ARG)",
> % + " YYPARSE_PARAM_DECL",
> % "{",
> % - " register int yym, yyn, yystate;",
> % + " int yym, yyn, yystate;",
> % "#if YYDEBUG",
> % - " register char *yys;",
> % - " extern char *getenv();",
> % + " const char *yys;",
> % "",
> % - " if (yys = getenv(\"YYDEBUG\"))",
> % + " if ((yys = getenv(\"YYDEBUG\")))",
> % " {",
> % " yyn = *yys;",
>
> The STDC support was especially messy for yyparse().
>
> Bruce
Thanks for all the remarks, I'll try to fix all issues you have spotted as soon
as possible.
regards,
Bapt
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/svn-src-head/attachments/20120523/84c1cee6/attachment.pgp
More information about the svn-src-head
mailing list