Re: _WIN32 and __CYGWIN__

[Bruno Haible]

Hi Ken,

> There's a lot of code in gnulib that contains 'defined _WIN32 && ! 
> defined __CYGWIN__'.

Yes.

> The '! defined __CYGWIN__' part is redundant, 
> because _WIN32 is never defined on Cygwin.

No. _WIN32 is not defined _by_default_ on Cygwin. But users can use
"gcc -D_WIN32", when they want to access native Windows APIs (for
whatever reason).

gnulib doesn't control what kind of -D options are given when a package
that contains some gnulib source files is compiled. Yet, the expectation
is that it produces the same code with "gcc -D_WIN32" as with plain "gcc".

> I think the presence of a redundant '! defined __CYGWIN__' is confusing to 
> people who read the code.

I disagree. Given that Cygwin always runs on Windows, people who read the
code might be confused about what the macro '_WIN32' means. So the
'! defined __CYGWIN__' actually clarifies things.

It might be clearer if we had a file that explains the correspondence
between platforms, config.guess triples, and predefined macros (sort of
a summary of predef.sourceforge.net). But we don't have such a file yet.

Bruno