Re: [PATCH] Avoid polluting cygwin namespace.

[Bruno Haible]

Hi Eric,

>    cygwin*)
>      FAULT_CONTEXT='CONTEXT'
> -    FAULT_CONTEXT_INCLUDE='#include <windows.h>'
> +    # CONTEXT is defined in <windows.h>, but cygwin programs should not
> +    # normally inspect the contents of CONTEXT, and the outright inclusion
> +    # of <windows.h> pollutes the namespace.
> +    FAULT_CONTEXT_INCLUDE='typedef struct _CONTEXT CONTEXT;'

This patch will break in two ways. I'm therefore opposed to it.

1) If the user does a #include <windows.h> before or after <sigsegv.h>,
   he'll get a compilation error about a redefinition of a type.

   error: redefinition of typedef ‘CONTEXT’

   In C++, duplicated redundant typedefs are harmless, but not in C.

2) The users of libsigsegv need to access the field of the FAULT_CONTEXT.
   That's what it is provided for, in the first place. For example, GNU clisp
   does in src/spvw_sigsegv.d:

     stackoverflow_context_t scp = (stackoverflow_context_t) arg1;
     ...
     #if defined(WIN32_NATIVE) || defined(UNIX_CYGWIN32)
      #ifdef I80386
       if (scp) { setSTACK(STACK = (gcv_object_t*)(scp->Ebx)); }
      #endif
     #endif

   This would be impossible if stackoverflow_context_t was an opaque type.

Instead, let's look more closely at the original problem: <windows.h>
defines WCHAR to a type, whereas grep/src/dfa.h uses it as an enum item.

The proposed workaround would not completely fix this problem: It would
solve it for Cygwin only, but not for mingw.

I would instead propose to add to dfa.h two lines:

   /* <windows.h> defines WCHAR as a type.  */
   #undef WCHAR

When porting GNU clisp to Windows, for some time I avoided collisions
through #undefs like this, but after some amount of workarounds I ended
up simply choosing different identifiers. For example, WIDECHAR instead of
WCHAR.

Bruno