This is the mail archive of the cygwin@sourceware.cygnus.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

I: gcc ... -U_WIN32 ... may cause problems


Hello All,
I am currently working on a 'source-portable' application, intended to run 
natively under Win32, various flavours of Unix as well as OpenVMS.  Usually I 
develop using Microsoft Developer Studio (VC++ 5.0, Intel), re-build and test 
under Cygwin32 B19.3 (EGCS 1.0.2), re-build  and test on Linux (gcc 2.7.2, 
Intel), re-build and test on OpenVMS/VAX 7.0 (DECC 5.5).  This arrangements 
allows me to conveniently test and resolve 'gross' portability issues. When I 
am sufficiently far advanced with the project I will build the application on 
DEC Unix, OpenVMS/AXP and, possibly, other platforms.
I use the same GNUmakefile for Cygwin32 and Linux (will use for other Unix(es) 
as well).  Since gcc under Cygwin32 does not pre-define "__unix__" or "__unix" 
or "unix", as is the case under Linux, and in the Cygwin32/EGCS/gcc environment 
_WIN32 is defined the same as in the MS VC++ environment I had to code
#if defined(_WIN32) && !defined(__CYGWIN32__) in a number of places.  This was 
required to ensure that  Windows NT-specific service code is not compiled under 
Cygwin32 but unix-specific daemon code is compiled instead, for example.  To 
avoid coding #if defined construct as shown I decided to undefine _WIN32 when 
compiling in Cygwin32 environment by using a command line of a form "gcc ... 
-U_WIN32 ...".  Before I did that the application worked fine however as soon 
as I did undefined _WIN32 it started to misbehave.
It turned out, after some investigations, that the 'stat' system call, which 
worked fine with _WIN32 defined, no longer works and causes stack corruption 
with the resulting SEGV.  It turned out, after further investigation, that the 
'missing' _WIN32 affects unistd.h header file.  If _WIN32 is defined just prior 
to #include <unistd.h> and undefined just after all is well otherwise the 
problem repeats.  It turned out, after further investigation, that 
'__MS_types__' is defined in unistd.h and that this is what makes all the 
difference.
Now I code that as:
#if defined(__CYGWIN32__) && !defined(_WIN32)
#   define __MS_types__	/* this breaks stat function if not defined  */
#   define _MS_types_DEFINED	/* at the time unistd.h is included */
#endif
#include <unistd.h>
#if defined(__CYGWIN32__) && defined(_MS_types_DEFINED)
#   undef __MS_types__
#   undef _MS_types_DEFINED
#endif
This way I avoid defining _WIN32, can safely (?) undefine _WIN32 and code on 
the basis of 'if _WIN32 is not defined than it must be Unix or VMS' :-).
Since I found that workaround I stopped poking around to see what exactly 
breaks the 'stat'.  If anyone knows or finds out please let me know :-).

This message may, perhaps, save somebody a great deal of hassle in the future 
:-)

Thanks Cygnus, Sergey and all other contributors for Cygwin32 and all related 
bits.
------------------
Cheers ...
Michael Czapski


-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]