Re: [Mingw-w64-public] Cygwin mingw64-i686-gcc-4.5.3-4: -Wformat warnings broken in C++

[JonY <jo...@us...>]

On 1/6/2012 04:42, Kai Tietz wrote:
> 2012/1/5 Christian Franke <Chr...@t-...>:
>> [Originally sent to Cygwin mailing list]
>>
>> When printf/scanf functions from MinGW runtime are selected via
>> __USE_MINGW_ANSI_STDIO, then format string checking is broken. This only
>> affects the C++ compiler:
>>
>> $ cygcheck -f /usr/bin/i686-w64-mingw32-g++
>> mingw64-i686-gcc-g++-4.5.3-4
>>
>> $ cat testfmt.c
>> #define __USE_MINGW_ANSI_STDIO 1
>> #include <stdio.h>
>>
>> void myprintf(const char *, ...)
>>   __attribute__((format(gnu_printf,1,2)));
>>
>> int main()
>> {
>>   long long x = 42;
>>   printf("%lld\n", x); // C++: Bogus warning
>>   myprintf("%lld\n", x); // No warning (OK)
>>   printf("%I64d\n", x); // Warning (OK)
>>   return 0;
>> }
>>
>> $ i686-w64-mingw32-gcc -Wformat -c testfmt.c
>> testfmt.c: In function 'main':
>> testfmt.c:12:3: warning: format '%I64d' expects type 'int', but argument
>> 2 has type 'long long int'
>>
>> $ i686-w64-mingw32-g++ -Wformat -c testfmt.c
>> testfmt.c: In function 'int main()':
>> testfmt.c:10:21: warning: unknown conversion type character 'l' in format
>> testfmt.c:10:21: warning: too many arguments for format
>> testfmt.c:12:22: warning: format '%I64d' expects type 'int', but
>> argument 2 has type 'long long int'
>>
>>
>> Interestingly the bogus warning only occurs for standard functions like
>> printf(). These are replaced by inline functions in MinGW stdio.h if
>> __USE_MINGW_ANSI_STDIO is set. Probably a subtle bug in the handling of
>> functions known by the compiler.
> 
> Issue is fixed on trunk.  Version 2 branch still has those issues, as
> fixes to stdio.h and wchar.h were not merged back.
> 

Looks like I need to do a new header/crt release soon.

>> Downgrading to previous version fixes the problem:
>>
>> $ cygcheck -f /usr/bin/i686-w64-mingw32-g++
>> mingw64-i686-gcc-g++-4.5.3-3
>>
>> $ i686-w64-mingw32-g++ -Wformat -c testfmt.c
>> testfmt.c: In function 'int main()':
>> testfmt.c:12:22: warning: format '%I64d' expects type 'int', but
>> argument 2 has type 'long long int'
>>
>>
>> Another observation: Include files from C++ standard library (e.g.
>> <string>) now silently set __USE_MINGW_ANSI_STDIO:
>>
>> /usr/lib/gcc/i686-w64-mingw32/4.5.3/include/c++/i686-w64-mingw32/bits/os_defines.h:
>>
>> ...
>> #undef __USE_MINGW_ANSI_STDIO
>> #define __USE_MINGW_ANSI_STDIO 1
>>
>> Is this done to ensure that iostream templates use the correct
>> FILE/stdio version?
>> (This was also not present in previous version 4.5.3-3).
> 
> Yes, this fixes internal libstdc++ usage of printf/scanf routines.
> libstdc++ internally assumes that printf/scanf function have POSIX
> semantics. So it makes no sense to make our POSIX printf/scanf
> functions optional, and so I turned them on by default.  To get still
> MS printf routines in C++, it is recommented to use
> underscored-function-name instead.
> 

Yes, this was added deliberately in the latest version.