Fix several Windows build warnings

[topolarity]

Should fix:

• [-Wstrict-prototypes]
• [-Wattributes]
• [-Wc++-compat]
• [-Wunused-but-set-variable]
• [-Wdiscarded-qualifiers]
• [-Wundef]

This leaves:

• [-Warray-bounds]
• [-Winfinite-recursion]

see JuliaCI/julia-buildkite#484

[vtjnash]

What is this check? The C standard states that environ is required to be defined by the user (thus must not be a macro)

[topolarity]

I'm attempting to fix:

  | C:/workdir/src/sys.c:500:1: warning: function declaration isn't a prototype [-Wstrict-prototypes]
  | 500 \| extern char **environ;
  | \| ^~~~~~
  | In file included from C:/workdir/src/sys.c:8:
  | C:/workdir/src/sys.c:500:15: warning: '__p__environ' redeclared without dllimport attribute: previous dllimport ignored [-Wattributes]
  | 500 \| extern char **environ;
  | \|               ^~~~~~~

but this doesn't reproduce for me locally (MinGW 11.5.0) so I don't know where this definition is coming from yet.

Presumably it's something like https://github.com/mirror/mingw-w64/blob/93f3505a758fe70e56678f00e753af3bc4f640bb/mingw-w64-headers/crt/stdlib.h#L595 ?

[topolarity]

Reproduces with:

// #define _POSIX_ // if uncommented, everything works
#define _UCRT
#include <stdlib.h>
extern char **environ;

$ x86_64-w64-mingw32-gcc -Wstrict-prototypes -c test.c
test.c:4:1: warning: function declaration isn’t a prototype [-Wstrict-prototypes]
    4 | extern char **environ;
      | ^~~~~~
In file included from test.c:3:
test.c:4:15: warning: ‘__p__environ’ redeclared without dllimport attribute: previous dllimport ignored [-Wattributes]
    4 | extern char **environ;
      |

maybe the right guard is defined( _POSIX_C_SOURCE )

[giordano]

What's the problem with _WIN32_WINNT_WIN8? It wasn't defined at some point? If so, maybe the offending file wasn't included a header file?

[topolarity]

Yes, it's all over our old logs: https://buildkite.com/julialang/julia-master/builds/50748#019972b4-b648-4997-9624-70e22d7230dc/864-2038

I'm not sure why that happens, but until recently (cbea8cf) we always just used the hex code

[giordano]

Make sure sdkddkver.h is included in the file which needed the macro?

[topolarity]

It reproduces for me with:

$ x86_64-w64-mingw32-g++ -D_WIN32_WINNT=_WIN32_WINNT_WIN8 -Wundef -c -x c++ - <<<"#include <mutex>"
<command-line>: warning: "_WIN32_WINNT_WIN8" is not defined, evaluates to 0 [-Wundef]

seems like a MinGW issue

[giordano]

julia> BinaryBuilderBase.runshell(Platform("x86_64", "windows"); preferred_gcc_version=v"10");
sandbox:${WORKSPACE} # echo "" | g++ -dM -E -std=c++17 -x c++ - | grep _WIN32_WINNT_WIN8
sandbox:${WORKSPACE} # echo "#include <mutex>" | g++ -dM -E -std=c++17 -x c++ - | grep _WIN32_WINNT_WIN8
sandbox:${WORKSPACE} # echo "#include <sdkddkver.h>" | g++ -dM -E -std=c++17 -x c++ - | grep _WIN32_WINNT_WIN8
#define _WIN32_WINNT_WIN8 0x0602

[topolarity]

What should I take away from that?

[giordano]

That sdkddkver.h is necessary for the macro to be defined, which is what I've been trying to say from https://github.com/JuliaLang/julia/pull/59634#discussion_r2372580441🙂

[topolarity]

But why should you have to include <sdkddkver.h> everywhere you use the C++ standard library?

[giordano]

Ah, ok, that's because the macro is defined in the command line arguments passed everywhere.

BTW, nowadays Windows 10 is the minimum supported version.