I have always felt it rather insecure or at the very least risky to
force all packages to link to -lintl, when all of them do not.
I have for some time successfully compiled a complete NLS/libintl free
system and have moved back and forth to compare performance and bugs.
During all of this I made notes on what did and did not like external
libintl.
Out of a list of 639 programs that I successfully compiled under my
HLFS based system, the following programs where the only ones with
apparent problems:
Linux-PAM
shadow
psmisc
ms-sys
alsa-utils
aspell
gftp
linux (the kernel)
tvtime
php
lbreakout2
wesnoth
That's it! I have not tested yet, whether or not the other 600+
programs did not disable NLS with an external libintl. However, in
order to disable nls entirely I had to hack/modify/patch at least ~30
projects. I also passed --enable-nls to every applicable project.
All of the mentioned projects above could be fixed by simply
executing: " make LDFLAGS='-lintl' ", except for the following:
gftp, which failed because it re-defined gettext AFTER including the
appropriate nls/gettext headers! My fix was: sed -i -e '/#define
gettext(str) (str)/d' src/gtk/misc-gtk.c
the linux kernel, which only uses nls for it's scripts. The hack was
a patch I called skip_nls:
--- linux-2.6.17.4/scripts/kconfig/Makefile.orig 2006-07-24
11:10:39.000000000 -0500
+++ linux-2.6.17.4/scripts/kconfig/Makefile 2006-07-24
11:12:16.000000000 -0500
@@ -122,9 +122,7 @@
if echo "\#include <libintl.h>" | $(HOSTCC) $(HOSTCFLAGS) -E - >
/dev/null 2>&1 ; \
then echo yes ; \
else echo no ; fi)
-ifeq ($(KBUILD_HAVE_NLS),no)
HOSTCFLAGS += -DKBUILD_NO_NLS
-endif
# generated files seem to need this to find local include files
HOSTCFLAGS_lex.zconf.o := -I$(src)
Any thoughts on this?
--
Kevin Day
--
http://linuxfromscratch.org/mailman/listinfo/hlfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
