[patch] Failed to build mips64-elf with NLS on OSX

[patch] Failed to build mips64-elf with NLS on OSX

[Masaki Muranaka]

Hello,

On mainline, it's failed to compile with a lot of
warnings like
: format not a string literal, argument types not checked
: format not a string literal and no format arguments


In case building with NLS, _("string") was expanded to
gettext("string"), not literal.

The patch is to suppress -Wformat-nonliteral
when you build with NLS.

My environment is OSX (gcc version 4.0.1), configuration
is --target=mips64-elf. I've not tested yet but I guess
it occurs on the another host and target.


- - - - - - - - - - -
2006-06-15  Masaki Muranaka <monaka@monami-software.com>

	* configure.ac: Remove -Wformat-nonliteral
	if USE_NLS = yes
	* configure: regenerate.

- - - - - - - - - - -
Index: configure
===================================================================
RCS file: /cvs/src/src/gdb/configure,v
retrieving revision 1.212
diff -u -p -r1.212 configure
--- configure   31 May 2006 15:14:36 -0000      1.212
+++ configure   15 Jun 2006 08:05:55 -0000
@@ -20277,8 +20277,11 @@ fi
# NOTE: If you add to this list, remember to update
# gdb/doc/gdbint.texinfo.
build_warnings="-Wimplicit -Wreturn-type -Wcomment -Wtrigraphs \
--Wformat -Wparentheses -Wpointer-arith -Wformat-nonliteral \
+-Wformat -Wparentheses -Wpointer-arith \
-Wunused-label -Wunused-function -Wno-pointer-sign"
+if test x"$USE_NLS" != xyes; then
+    build_warnings="$build_warnings -Wformat-nonliteral"
+fi
# GCC supports -Wuninitialized only with -O or -On, n != 0.
if test x${CFLAGS+set} = xset; then
Index: configure.ac
===================================================================
RCS file: /cvs/src/src/gdb/configure.ac,v
retrieving revision 1.33
diff -u -p -r1.33 configure.ac
--- configure.ac        31 May 2006 15:14:37 -0000      1.33
+++ configure.ac        15 Jun 2006 08:05:56 -0000
@@ -1104,8 +1104,11 @@ fi
# NOTE: If you add to this list, remember to update
# gdb/doc/gdbint.texinfo.
build_warnings="-Wimplicit -Wreturn-type -Wcomment -Wtrigraphs \
--Wformat -Wparentheses -Wpointer-arith -Wformat-nonliteral \
+-Wformat -Wparentheses -Wpointer-arith \
-Wunused-label -Wunused-function -Wno-pointer-sign"
+if test x"$USE_NLS" != xyes; then
+    build_warnings="$build_warnings -Wformat-nonliteral"
+fi
# GCC supports -Wuninitialized only with -O or -On, n != 0.
if test x${CFLAGS+set} = xset; then

--
Masaki Muranaka
Monami software

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Andreas Schwab]

Masaki Muranaka <monaka@monami-software.com> writes:

> Hello,
>
> On mainline, it's failed to compile with a lot of
> warnings like
> : format not a string literal, argument types not checked
> : format not a string literal and no format arguments
>
>
> In case building with NLS, _("string") was expanded to
> gettext("string"), not literal.

GCC should recognize gettext specially in the context of format string
tests.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Masaki Muranaka]

On 2006/06/15, at 17:22, Andreas Schwab wrote:
> GCC should recognize gettext specially in the context of format string
> tests.

Regardless GCC should work so, the build was failed.

This is a sample script that reproduce the error.
I tested on gcc-4.0.1/OSX and gcc-3.2.2/Linux.
Results are same.

-8<----8<----8<----8<----8<----8<---
cat <<__EOF > test.c
#include <stdio.h>

# define _(String) gettext (String)

int
main(int argc, char **argv)
{
         printf(_("aaa"));
}
__EOF
gcc -Werror -Wformat -Wformat-nonliteral -c test.c
-8<----8<----8<----8<----8<----8<---


--
Masaki Muranaka
Monami software

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Andreas Schwab]

Masaki Muranaka <monaka@monami-software.com> writes:

> On 2006/06/15, at 17:22, Andreas Schwab wrote:
>> GCC should recognize gettext specially in the context of format string
>> tests.
>
> Regardless GCC should work so, the build was failed.

GCC _does_ work so.

> This is a sample script that reproduce the error.
> I tested on gcc-4.0.1/OSX and gcc-3.2.2/Linux.
> Results are same.
>
> -8<----8<----8<----8<----8<----8<---
> cat <<__EOF > test.c
> #include <stdio.h>
>
> # define _(String) gettext (String)
>
> int
> main(int argc, char **argv)
> {
>         printf(_("aaa"));
> }
> __EOF
> gcc -Werror -Wformat -Wformat-nonliteral -c test.c
> -8<----8<----8<----8<----8<----8<---

This is not a valid testcase since gettext is not declared.

I cannot reproduce the problem with any gcc version, once the missing
declaration is added.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Masaki Muranaka]

On 2006/06/15, at 18:24, Andreas Schwab wrote:
> GCC _does_ work so.

After sent the previous mail, I've built GDB on my Linux box.
The build was succeeded without my patch.
So I believe you are right.
Then, I confused why the build was failed on OSX.

The use of intl/ is a one of the configuration diffence
between my Linux host and my OSX one.
In my Linux box, Gettext has installed. so intl/libintl.h
is not generated.

How do you think possibilities about bugs in intl/ causes
the build failed?

The cpp'ed code under OSX is like this:
internal_error ("../../src/gdb/mips-tdep.c", 273, libintl_gettext  
("bad switch"));
And the definition of internal_error is like this:
extern  void internal_error ( char *file, int line,
                                       char *, ...) __attribute__  
((noreturn)) __attribute__ ((format(printf,  3,  4)));


--
Masaki Muranaka
Monami software

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Masaki Muranaka]

Still I was confused, but I see somehow.

On 2006/06/15, at 19:04, Masaki Muranaka wrote:
> internal_error ("../../src/gdb/mips-tdep.c", 273, libintl_gettext  
> ("bad switch"));

As Schwab says, dgettext/gettext was regarded as constant by gcc.
The information was put on gcc/builtin.def.
So there is no troubles on some targets which expands _() to dgettext().
I checked the code generated on Linux. _() was expanded to
dgettext().

But the another target which expands _() to libintl_gettext(),
the build will be failed. Because there is no information
about libintl_gettext() in builtin.def.
The code which is generated on MacOSX was so.

If my understanding is reasonable, -Wformat-nonliteral
should be off in case GDB is built with intl/, right?
(Fixing GCC or intl is also not impossible, but I don't want to)

--
Masaki Muranaka
Monami software

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Daniel Jacobowitz]

On Thu, Jun 15, 2006 at 08:34:06PM +0900, Masaki Muranaka wrote:
> 
> Still I was confused, but I see somehow.
> 
> On 2006/06/15, at 19:04, Masaki Muranaka wrote:
> >internal_error ("../../src/gdb/mips-tdep.c", 273, libintl_gettext  
> >("bad switch"));
> 
> As Schwab says, dgettext/gettext was regarded as constant by gcc.
> The information was put on gcc/builtin.def.
> So there is no troubles on some targets which expands _() to dgettext().
> I checked the code generated on Linux. _() was expanded to
> dgettext().
> 
> But the another target which expands _() to libintl_gettext(),
> the build will be failed. Because there is no information
> about libintl_gettext() in builtin.def.
> The code which is generated on MacOSX was so.
> 
> If my understanding is reasonable, -Wformat-nonliteral
> should be off in case GDB is built with intl/, right?
> (Fixing GCC or intl is also not impossible, but I don't want to)

But it's right :-)

char *libintl_gettext (const char *) __attribute__((format_arg(1)));

That's enough to fix it.  The example in the manual is even a dgettext
wrapper.  It looks like it's been there since ~ 2001.  There's already
a GNUC version check in libgnuintl.h; adding another one with the
appropriate version ought to fix this.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [patch] Failed to build mips64-elf with NLS on OSX

[Masaki Muranaka]

[-- Attachment #1: Type: text/plain, Size: 513 bytes --]


On 2006/06/15, at 22:51, Daniel Jacobowitz wrote:
> But it's right :-)
>
> char *libintl_gettext (const char *) __attribute__((format_arg(1)));
>
> That's enough to fix it.

Thanks for your comment.
It's shut up the warinings to add the attribute.

An attached file is the patch I was applied.
For MacOSX, it's enough to modify at libgnuintl.h:line 128.
And  I guess it should be applied line 119, too.

Then, what should I do to request for apply?
Need to post to gcc-bugs?

--
Masaki Muranaka
Monami software


[-- Attachment #2: libgnuintl.h.diff --]
[-- Type: application/octet-stream, Size: 1003 bytes --]

Index: libgnuintl.h
===================================================================
RCS file: /cvs/src/src/intl/libgnuintl.h,v
retrieving revision 1.1
diff -u -p -r1.1 libgnuintl.h
--- libgnuintl.h	22 May 2006 15:30:13 -0000	1.1
+++ libgnuintl.h	16 Jun 2006 00:40:02 -0000
@@ -116,7 +116,7 @@ extern "C" {
    LC_MESSAGES locale.  If not found, returns MSGID itself (the default
    text).  */
 #ifdef _INTL_REDIRECT_INLINE
-extern char *libintl_gettext (const char *__msgid);
+extern char *libintl_gettext (const char *__msgid)  __attribute__((format_arg(1)));
 static inline char *gettext (const char *__msgid)
 {
   return libintl_gettext (__msgid);
@@ -126,7 +126,7 @@ static inline char *gettext (const char 
 # define gettext libintl_gettext
 #endif
 extern char *gettext _INTL_PARAMS ((const char *__msgid))
-       _INTL_ASM (libintl_gettext);
+       _INTL_ASM (libintl_gettext)  __attribute__((format_arg(1)));
 #endif
 
 /* Look up MSGID in the DOMAINNAME message catalog for the current