Re: gcc-3.0 bug in unixware 7.1.1 (serious)

[Matt Schalit <mschalit at pacbell dot net>]

Franciszek Czekala wrote:
> 
> The file byteorder.h
> (gcc-dir/lib/gcc-lib/i586-sco-sysv5uw7.1.1/3.0/include/sys/byteorder.h)
> defines htons and ntohs functions as having the argument unsigned int
> instead of unsigned short (short is what these functions are about,
> aren't they?) thus conflicting with /usr/include/arpa/inet.h
> and preventing network applications from compiling.
> 
> This can be fixed by manually editing appropriate declarations
> and definitions in byteorder.h but it would be better if the correct
> file were generated (it is meant to fix some problems with default
> system include files so it is funny that it introduces new ones).
> 
> Best regards
> Franciszek Czekala, Poland


It's fixed in CVS.

I reported it and Bruce Korb fixed it.  Here's what the
patch is, essentially:  

   Apply it to the file  -->  gcc-3.0/gcc/fixinc/inclhack.def


----------------------------------------------------------------
--- inclhack.def.orig   Thu Jun 14 21:16:00 2001
+++ inclhack.def        Thu Jun 28 13:10:57 2001
@@ -3043,7 +3043,7 @@
     hackname = uw7_byteorder_fix;
     files    = arpa/inet.h;
     select   = "in_port_t";
-    test     = "-f $DESTDIR/sys/byteorder.h";
+    test     = "-f sys/byteorder.h";
 #ifndef SVR5
        mach = "*-*-sysv4*";
        mach = "i?86-*-sysv5*";
-----------------------------------------------------------------------------



You can see that it was searching for byteorder.h in the
wrong place, considering the fact that, when you get this running,
the uw7_byteorder_fix is the first fixinclude that gets applied.

Thus there won't be any $DESTDIR/sys/byteorder.h at that time,
unless you already installed gcc-3.0 once and didn't rename
gcc-libs when you were building it again.

I included this all in the gcc-mini-HOWTO I posted last night to
comp.unix.unixware.misc.

Cheers,
Matthew