[Linux-ia64] Problem to build Kernel 2.4.0 with gcc version "2.96-ia64-000717"

[Linux-ia64] Problem to build Kernel 2.4.0 with gcc version "2.96-ia64-000717"

[Faccini, Bruno]

[-- Attachment #1: Type: text/plain, Size: 1933 bytes --]


When trying to build Kernel 2.4.0 (from
"http://www.kernel.org/pub/linux/kernel/v2.4/linux-2.4.0.tar.gz" +
"http://www.kernel.org/pub/linux/kernel/ports/ia64//linux-2.4.0-ia64-010109.
diff.gz" downloads) with gcc version "2.96-ia64-000717" + snap "001117" we
experienced compilation errors with the specific
"net/ipv4/netfilter/ip_tables.c" source file.
After some investigations, it appeared to be due to a problem/regression
with version "2.96-ia64-000717" (+ snap "001117" applied, the gcc pkg
shipped as part of RedHat64 "December-2000" Beta release) to deal with the
following macro in "include/linux/netfilter_ipv4/ip_tables.h" header : 
--------------------------------
#define IPT_MATCH_ITERATE(e, fn, args...)	\
({						\
	unsigned int __i;			\
	int __ret = 0;				\
	struct ipt_entry_match *__m;		\
						\
	for (__i = sizeof(struct ipt_entry);	\
	     __i < (e)->target_offset;		\
	     __i += __m->u.match_size) {	\
		__m = (void *)(e) + __i;	\
						\
		__ret = fn(__m , ## args);	\
		if (__ret != 0)			\
			break;			\
	}					\
	__ret;					\
})
--------------------------------
where "## args" substitution was wrongly done (the first argument in the
"...[,...]" list was garbled depending on its first character, "*" or "&"
are missing if specified and if none the full argument/variable name is
fully missing !!, leading to the compiler error because of 2x"," in a row ).
The simplest way to bypass this was to generate a pre-processed only C
source file, to manually edit+fix the problems/missing characters/names and
to use it as the original source file.
This points to some regression between latest gcc versions vs Linux Kernel
source files content and the syntax used (there was a lot of warnings about
"depracated" use of "##" constructs during the Kernel build but the only
error found) and may be highlights the need of a "generic" gcc ship+use for
Kernel builds ...
Bruno.


 <<Bruno Faccini (E-mail).vcf>> 

[-- Attachment #2: Bruno Faccini (E-mail).vcf --]
[-- Type: application/octet-stream, Size: 847 bytes --]

BEGIN:VCARD
VERSION:2.1
N:Faccini;Bruno
FN:Bruno Faccini (E-mail)
ORG:Intel;Technical Marketing/Content Group EMEA
TITLE:Tech. Mktg. Engr.
TEL;WORK;VOICE:+33 (0) 146947228
TEL;WORK;VOICE:+33 (0) 155498427
TEL;WORK;FAX:+33 (0) 146947068
ADR;WORK;ENCODING=QUOTED-PRINTABLE:;Munich, Germany.;Intel Corporation=0D=0ATour Chenonceaux=0D=0A204 Rond Poin=
t Du Pont De Sevres;Boulogne Billancourt;Cedex;92516;France
LABEL;WORK;ENCODING=QUOTED-PRINTABLE:Munich, Germany.=0D=0AIntel Corporation=0D=0ATour Chenonceaux=0D=0A204 Rond =
Point Du Pont De Sevres=0D=0ABoulogne Billancourt, Cedex 92516=0D=0AFrance
ADR;HOME:;;20 Rue Du Marechal Joffre;Brie Comte Robert;;77170;France
LABEL;HOME;ENCODING=QUOTED-PRINTABLE:20 Rue Du Marechal Joffre=0D=0ABrie Comte Robert 77170=0D=0AFrance
EMAIL;PREF;INTERNET:bruno.faccini@intel.com
REV:20001111T162700Z
END:VCARD

Re: [Linux-ia64] Problem to build Kernel 2.4.0 with gcc version "2.96-ia64-000717" + snap "001117"

[Jim Wilson]

>experienced compilation errors with the specific
>"net/ipv4/netfilter/ip_tables.c" source file.

See the file ftp.cygnus.com:pub/ia64-linux/snap-001117/WORKAROUND.cpp.kernel,
available since Nov 17.

The gcc c preprocessor was rewritten 3 or 4 times this year, and this
particular bug was fixed in the last rewrite which was sometime in late
October I think.  However, since there is no guarantee that the final rewrite
is perfect, so it is safer to work around the problem for now.

>This points to some regression between latest gcc versions vs Linux Kernel
>source files content and the syntax used (there was a lot of warnings about
>"depracated" use of "##" constructs during the Kernel build but the only
>error found) and may be highlights the need of a "generic" gcc ship+use for
>Kernel builds ...

As part of the many rewrites, the C preprocessor now has better compliance
to ISO C, and some GNU extensions have been deprecated.  However, you shouldn't
pay too much attention to the warnings you get from this July FSF snapshot,
because this cpp behaviour changed several times after that.  Better to
try the current FSF gcc sources and worry about those warnings.  I'd suspect
other people have already done this.

Your comment about a "generic" gcc for kernel builds makes no sense.  This
is the gcc to use for kernel builds.  There is no other IA-64 gcc that is
usable.

Jim