Re: gl_MANYWARN_ALL_GCC() leads to many spurious warnings

[Samuel Bronson <naesten@gmail.com>]

On Jul 7, 2012, at 3:53 PM, Paul Eggert wrote:

> On 07/06/2012 03:38 PM, Samuel Bronson wrote:
>
>> why does it use flags like -Wunreachable-code (which was so buggy
>> that it eventually got yanked -- it's now silently ignored)
>
> The big picture is that --enable-gcc-warnings hasn't been
> ironed out --with-ns.  This can be fixed, and I expect it
> will find a few real bugs along with many things that are
> just nit cleanups, but it'll take some work.  Is this
> something you'd like to help out with?  It could be combined
> with the idea of maintaining another configure-time option
> useful for finding bugs on Apple.

Yeah, I've actually started trying to get rid of a lot of warnings in  
my own tree already; I've been waiting on my copyright assignment  
papers before actually submitting.

At this point I don't think the warnings will turn up much in the way  
of actual bugs; it is turning up a fair amount of seemingly redundant  
code, though.

I did find one in GCC 4.7.1, though; it gives warnings like this:

/Users/user/hacking/emacs/src/nsterm.m: In function '-[EmacsView  
markedRange]':
/Users/user/hacking/emacs/src/nsterm.m:5041:5: warning: format not a  
string literal and no format arguments [-Wformat-security]

for calls that most assuredly do use string literals:

- (NSRange)markedRange
{
   NSRange rng = workingText != nil
     ? NSMakeRange (0, [workingText length]) : NSMakeRange  
(NSNotFound, 0);
   if (NS_KEYLOG)
     NSLog (@"markedRange request");
   return rng;
}

given this declaration (from NSObjCRuntime.h):

FOUNDATION_EXPORT void NSLog(NSString *format, ...)  
__attribute__((format(__NSString__, 1, 2)));

> I started on this task, checking in the results in trunk bzr
> 108941, but there's lots more where that came from and to be
> honest I don't understand the --with-ns code that well.  I
> did see some unused-macro warnings but they seemed correct
> to me -- perhaps I am missing something?  Anyway, you can
> look at the diffs in 108941 to see how to go about this,
> and/or to let me know what I did wrong.
>
> Anyway, to get back to your question, I suspect
> -Wunreachable-code was there because it is harmless with the
> latest GCC, so nobody bothered to remove it.  Since this
> flag doesn't ever help I removed it in trunk bzr 108941.
>
>> and -Wsync-nand (which is totally irrelevant, and warns that this
>> warning isn't allowed for Objective C code)
>
> Same thing -- this flag never bothered anybody in the
> --without-ns world.  Since it never helps either, I removed
> it too.

Fair enough, I guess.

>> and -Wunused-macros (which seems rather buggy in GCC 4.7.1 -- it
>> doesn't seem to count appearing in an #ifdef as a use?).
>
> I don't observe a bug here.  The following program compiles
> just fine with GCC 4.7.1, with -Wunused-macros.  (This is on
> Fedora 15 x86-64.)
>
>  #define FOO bar
>  #ifdef FOO
>  int main (void) { return 0; }
>  #endif
>
> What bug do you see?

Well, okay, it doesn't seem *quite* buggy in GCC 4.7.1 -- it just  
seems to solve the wrong problem, based on what it does with regex.c  
when building in lib-src:

/sw/bin/ccache /sw/bin/gcc-fsf-4.7 -std=gnu99 -c   -Wall -W -Wformat- 
y2k -Wformat-security -Winit-self -Wmissing-include-dirs -Wunused - 
Wunknown-pragmas -Wstrict-aliasing -Wdeclaration-after-statement - 
Wunsafe-loop-optimizations -Wpointer-arith -Wbad-function-cast -Wcast- 
align -Wwrite-strings -Wstrict-prototypes -Wold-style-definition - 
Wmissing-prototypes -Wmissing-declarations -Wmissing-noreturn - 
Wmissing-format-attribute -Wpacked -Wunreachable-code -Winvalid-pch - 
Wvolatile-register-var -Wdisabled-optimization -Wbuiltin-macro- 
redefined -Wmudflap -Wpacked-bitfield-compat -Wsync-nand -Wattributes - 
Wcoverage-mismatch -Wunused-macros -Wabi -Wcpp -Wdeprecated - 
Wdeprecated-declarations -Wdiv-by-zero -Wdouble-promotion -Wendif- 
labels -Wextra -Wformat-contains-nul -Wformat-extra-args -Wformat-zero- 
length -Wformat=2 -Wmultichar -Wnormalized=nfc -Woverflow -Wpointer-to- 
int-cast -Wpragmas -Wsuggest-attribute=noreturn -Wtrampolines -Wno- 
missing-field-initializers -Wno-sign-compare -Wno-type-limits -Wno- 
switch -Wno-unused-parameter -Wno-format-nonliteral -Wno-logical-op - 
fdiagnostics-show-option -funit-at-a-time  -I. -I../src -I../lib -I/ 
Users/user/hacking/emacs/lib-src -I/Users/user/hacking/emacs/lib- 
src/../src -I/Users/user/hacking/emacs/lib-src/../lib  -DXASSERTS=1 -g  
-O2 -DCONFIG_BROKETS -DINHIBIT_STRING_HEADER \
	  /Users/user/hacking/emacs/lib-src/../src/regex.c
/Users/user/hacking/emacs/lib-src/../src/regex.c:805:0: warning: macro  
"CHARSET_RANGE_TABLE_BITS" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:268:0: warning: macro  
"CHAR_HEAD_P" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:269:0: warning: macro  
"SINGLE_BYTE_CHAR_P" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:1880:0: warning:  
macro "EXTEND_RANGE_TABLE" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:1890:0: warning:  
macro "SET_RANGE_TABLE_WORK_AREA_BIT" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:270:0: warning: macro  
"SAME_CHARSET_P" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:285:0: warning: macro  
"BYTE8_TO_CHAR" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:284:0: warning: macro  
"MAKE_CHAR" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:263:0: warning: macro  
"CHARSET_LEADING_CODE_BASE" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:262:0: warning: macro  
"CHAR_CHARSET" is not used [-Wunused-macros]
/Users/user/hacking/emacs/lib-src/../src/regex.c:1903:0: warning:  
macro "SET_RANGE_TABLE_WORK_AREA" is not used [-Wunused-macros]

(It seems that these macros are only used #ifdef emacs)