Bug#1061655: wine: FTBFS with error: unknown conversion type character ‘I’ in format [-Werror=format=]

[Petter Reinholdtsen <pere@hungry.com>]

I got more input on #debian-arm:

  <iam_tj[m]1> <pere> "iam_tj: i386 seem to work. :/" <- That's because
    the compiler options for the 32-bit architectures include `-Wformat`
    (equivalent to `-Wformat=1`) but the 64-bit architectures have
    `-Wno-format` (equivalent to `-Wformat=0`)
  
  <iam_tj[m]1> pere: see with:

    getbuildlog wine last 
    for log in wine_9.0~repack-2_*.log; do echo "Searching $log"; grep -n -- '-c -o programs/winedbg.*/info\.o' $log | tee $log.match; done
    grep -El 'Wno-format( |$)' wine_9.0~repack-2_*.log.match
    grep -El 'Wformat( |$)' wine_9.0~repack-2_*.log.match

  that gives:

    $  grep -El 'Wno-format( |$)' wine_9.0~repack-2_*.log.match
    wine_9.0~repack-2_amd64.log.match
    wine_9.0~repack-2_arm64.log.match
    $ grep -El 'Wformat( |$)' wine_9.0~repack-2_*.log.match
    wine_9.0~repack-2_armel.log.match
    wine_9.0~repack-2_armhf.log.match
    wine_9.0~repack-2_i386.log.match

  <pere> is this explaining why i386 build but armel do not?  Both seem
    to have -Wformat.
  
  <iam_tj[m]1> The i386 build uses i686-w64-mingw32-gcc but the
    armhf/armel builds use gcc
  
  <iam_tj[m]1> So there's two differences combine to obscure things; 1)
    64-bits don't check the format string and 2) different compilers
    between x86 and ARM arches for 32-bit

  <iam_tj[m]1> compiler options are quite different too:
      $ wdiff -3 wine_9.0~repack-2_{i386,armhf}.log.match 

      ======================================================================
      [-8455:i686-w64-mingw32-gcc-]{+8702:gcc+}
      ======================================================================
      [-programs/winedbg/i386-windows/info.o-] {+programs/winedbg/info.o+}
      ======================================================================
      [--D__WINE_PE_BUILD-]
      ======================================================================
      {+-pipe -fcf-protection=none -fvisibility=hidden -fno-stack-protector+}
      ======================================================================
      [--Wabsolute-value -Wenum-conversion -fno-omit-frame-pointer -mpreferred-stack-boundary=2-]
      ======================================================================
      {+-fPIC -fasynchronous-unwind-tables -D_WIN32 -fno-builtin -fshort-wchar+}
      ======================================================================
      {+-fstack-clash-protection+}
      ======================================================================
      {+-Wno-unused-function -mthumb -mfloat-abi=hard+}
      ======================================================================

  <iam_tj[m]1> pere: the format specifier itself is definitely incorrect
    as I said earlier - additionally it looks like the width value
    `ADDRWIDTH` is incorrect since it will be 64 on 32-bit systems, but
    the value should be the width in characters, NOT bits! My guess is
    that ought to be ` ADDRWIDTH / 8`; a minimal test-case to show the
    format specifier error:
  
      $ cat bad-format-specifier.c; gcc -Werror=format -Wformat=1 -o
        bad-format-specifier bad-format-specifier.c
      
      #include <stdio.h>
      int main(int argc, char **argv, char **env) {
        int ADDRWIDTH = 64;
        long long base = 0x11223344;

        printf ( "0x%0I*lld is not a valid module address\n", ADDRWIDTH, base);
        printf ( "0x%0*Illd is not a valid module address\n", ADDRWIDTH, base);

        return 0;
      }
      bad-format-specifier.c: In function ‘main’:
      bad-format-specifier.c:7:24: error: unknown conversion type character ‘I’ in format [-Werror=format=]
          7 |         printf ( "0x%0*Illd is not a valid module address\n", ADDRWIDTH, base);
            |                        ^
      bad-format-specifier.c:7:18: error: too many arguments for format [-Werror=format-extra-args]
          7 |         printf ( "0x%0*Illd is not a valid module address\n", ADDRWIDTH, base);
            |                  ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      cc1: some warnings being treated as errors

    This is on amd64, which also shows that it would have failed on the
    64-bit architectures if they used `-Wformat -Werror=format`

  <iam_tj[m]1> In the code above the 1st printf corrects the format
    specifier and the 2nd copies the one in wine (indirectly from
    debian/patches/arm/format-strings.patch )

-- 
Happy hacking
Petter Reinholdtsen