gcc -U_WIN32 pitfalls!

[Mumit Khan]

Just for kicks, I tried to build a small analysis program with -U_WIN32;
this tool is essentially a filter and doesn't use any win32 specific 
features, so it should be pretty safe, right?

Looks like certain typedefs are *WRONG* when you undefine _WIN32 and it
causes amazingly subtle bugs! Any code using stat for example will be
subtly but surely broken.

I'm attaching a small trivial program that should illustrate this. Here's
the header comments from the file. 

This is with stock B20.1. Compiler version is no relevant here.

//
// A simple program to show the pitfall of using -U_WIN32 in your code
// under Cygwin b20.1!
//
// To build:
//  $ c++ -U_WIN32 -o cyg-stat-bug cyg-stat.cc
//  $ c++ -o cyg-stat-ok cyg-stat.cc
//
// Now run it giving a filename on command line:
//
//  $ ./cyg-stat-bug ./cyg-stat.cc
//  $ ./cyg-stat-ok ./cyg-stat.cc
//
// To find the cause of the problem (after you notice it in the output
// of course!), you need to look at the preprocessed files.
//
//  $ c++ -save-temps -U_WIN32 -o cyg-stat-bug cyg-stat.cc
//  $ mv cyg-stat.ii cyg-stat-bug.ii
//  $ c++ -save-temps -o cyg-stat-ok cyg-stat.cc
//  $ mv cyg-stat.ii cyg-stat-ok.ii
//  $ diff -u3p cyg-stat-bug.ii cyg-stat-ok.ii
//
//  Look for ino_t.
// 

Regards,
Mumit