This is the mail archive of the mailing list for the libstdc++ project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Anyone got irix5? I have questions about your ctype implementation

 > From: Mark Mitchell <>
 > > So for that reason and also for being able to specifically target
 > > pre-irix6.5, IMHO I prefer to solve it the way I originally did with
 > > libstdc++-v3 os_defines.h.
 > We're really just debating which evil is more evil.  There are lots of 
 > evils
 > in both directions.
 > Your points are valid, but not strong enough to cause us to do something
 > different on IRIX, with respect to everything else.  And having programs
 > behave different due to whether or not a V3 header is included first is
 > a user-visible problem; maintenance issues are visible only to us.

Yes you're right.  If there's a burden to be had both ways, it should
fall on maintainers, not users.

 > So, let's go with CPLUSPLUS_CPP_SPEC.

Here's a patch I'm currently testing on irix6.2.  (I can't test it on
6.5)  Assuming bootstrap and tests pass, ok for the trunk?


PS: I'll be away from email next week, so if its approved past Monday
morning, installation will have to wait a few days.

2002-05-12  Kaveh R. Ghazi  <>

	* config/os/irix/irix5.2/bits/os_defines.h: Revert 2002-05-06

	* mips/iris6.h (CPLUSPLUS_CPP_SPEC): Define.

diff -rup orig/egcc-CVS20020511/libstdc++-v3/config/os/irix/irix5.2/bits/os_defines.h egcc-CVS20020511/libstdc++-v3/config/os/irix/irix5.2/bits/os_defines.h
--- orig/egcc-CVS20020511/libstdc++-v3/config/os/irix/irix5.2/bits/os_defines.h	2002-05-06 10:01:27.000000000 -0400
+++ egcc-CVS20020511/libstdc++-v3/config/os/irix/irix5.2/bits/os_defines.h	2002-05-12 16:23:22.594661495 -0400
@@ -44,22 +44,6 @@
 // the command-line.
-/* On at least irix6.2 (which uses this file,) the header files are
-   very fragile and depend heavily on _SGI_SOURCE and __EXTENSIONS__
-   being defined.  If they're not, then features disappear that
-   configure was relying on which then breaks the libstdc++ headers.
-   These macros are normally implicitly defined by the command-line,
-   however they are not defined when -ansi is used, and -ansi is
-   passed to the testsuite.  The g++ test results (where it uses
-   libstdc++ headers) get dozens of extra failures without these
-   macros, so ensure they always are defined here.  */
-#ifndef _SGI_SOURCE
-# define _SGI_SOURCE
-#ifndef __EXTENSIONS__
-# define __EXTENSIONS__
 #define __off_t off_t
 #define __off64_t off64_t
 #define __ssize_t ssize_t
diff -rup orig/egcc-CVS20020511/gcc/config/mips/iris6.h egcc-CVS20020511/gcc/config/mips/iris6.h
--- orig/egcc-CVS20020511/gcc/config/mips/iris6.h	2001-12-17 11:33:33.000000000 -0500
+++ egcc-CVS20020511/gcc/config/mips/iris6.h	2002-05-12 16:23:12.345807810 -0400
@@ -124,6 +124,18 @@ Boston, MA 02111-1307, USA.  */
 %{mabi=64: -D__mips64} \
 %{!mabi*: -D__mips64}"
+/* The GNU C++ standard library requires that __EXTENSIONS__ and
+   _SGI_SOURCE be defined on at least irix6.2 and probably all irix6
+   prior to 6.5.  They normally get defined in SUBTARGET_CPP_SPEC if
+   !ansi, for g++ we want them regardless.  We don't need this on
+   irix6.5 itself, but it shouldn't hurt other than the namespace
+   polution.  */
+%{ansi:-D__EXTENSIONS__ -D_SGI_SOURCE} %(cpp) \
 /* Irix 6 uses DWARF-2.  */

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