omniORB 3.0.2 bugs

The following bugs in omniORB 3.0.2 have been fixed. Update from CVS to get the fixes.


Summary:	Incorrect connection closure on Windows (bug number 16)
Date:	Mon Feb 5 12:12:14 GMT 2001
Reported by:	Victor Joukov
Description:	In extremely rare cases, a recv() could be interrupted on Windows, causing a connection closure, rather than the proper retry.

Summary:	omniIDL/C++ backend interface inheritance errors (bug number 15)
Date:	Sun Jan 28 15:30:10 GMT 2001
Reported by:	Gnana
Description:	The C++ backend failed to map the names of some interfaces to unambiguous C++ class names in some circumstances. Example IDL: interface I{}; module M{ interface I : ::I{}; interface J : I{}; };

Summary:	Bug in omniidl C++ back-end with struct member within a union which in turn is within a namespace. Only causes compilation problem with MSVC++. (bug number 14)
Date:	Wed Jan 24 13:28:27 GMT 2001
Reported by:	Lars Immisch
Description:	IDL like the following would cause invalid code be generated for MSVC++: namespace A { union B switch(long) { case 0: struct C { string op; } ex; default: long digit; }; };

Summary:	Bug in omniidl C++ back-end with unusual inheritance (bug number 13)
Date:	Mon Jan 15 15:49:43 GMT 2001
Reported by:	Daniel von Tabouillot
Description:	IDL like the following would cause an internal error in the omniidl C++ back-end: interface A; typedef A B; interface A {}; interface C : B {};

Summary:	CORBA::Object::_narrow() does not increment the reference count. (bug number 12)
Date:	Mon Dec 11 09:51:35 GMT 2000
Reported by:	Kevin Bailey

Summary:	Invalid string in postinvoke() with long operation names (bug number 11)
Date:	Tue Dec 5 11:57:03 GMT 2000
Reported by:	Chris Newbold
Description:	If the operation name is longer than 32 characters, postinvoke() would be called with an invalid operation name string.

Summary:	Bug with multiply-recursive TypeCode (bug number 10)
Date:	Mon Nov 27 18:48:52 GMT 2000
Reported by:	Walter Stroebel
Description:	A recursive type in which more than one member recursed to the same type could cause an infinite recursion, leading to a stack overflow.

Summary:	New platform file for Tru64 5.0 (bug number 9)
Date:	Thu Nov 23 11:22:57 GMT 2000
Reported by:	dek_ml@konerding
Description:	Also fixed a problem with omkdepend when compiled with native cc.

Summary:	On Digital Unix/Tru64, gethostbyname_r() is not used properly (bug number 8)
Date:	Thu Nov 23 11:22:57 GMT 2000
Reported by:	Neeraj Bhatia
Description:	This bug causes intermittent problems and causes a crash in gethostbyname_r.

Summary:	Invalid object references handled incorrectly (bug number 7)
Date:	Tue Nov 21 10:46:54 GMT 2000
Reported by:	Nick Belshaw
Description:	An IOR containing no recognised profiles (e.g. from ORBit with default settings) would be treated as nil by string_to_object, and a MARSHAL exception by the unmarshalling code. The correct exception is INV_OBJREF.

Summary:	System exceptions not forwarded from incarnate() and preinvoke() (bug number 6)
Date:	Mon Nov 13 11:47:35 GMT 2000
Reported by:	Chris Newbold
Description:	System exceptions thrown from a ServantLocator's preinvoke() method, or a ServantActivator's incarnate() method are silently converted into an OBJ_ADAPTER exception. They should be passed on to the client without change.

Summary:	Examples fail to build with nmake on Windows 9x (bug number 5)
Date:	Tue Nov 7 15:31:24 GMT 2000
Description:	As explained in 3.0.1's bug 9, some Windows 98 machines (but not all) fail to cope with the pipe between the C pre-processor and omniidl. The work-around is to use the -T flag. The dir.mak files in the examples now use this flag.

Summary:	omniidl system dependency clean-up (bug number 4)
Date:	Tue Oct 24 10:44:08 BST 2000
Description:	Not really a bug, but some long-standing ugliness in omniidl's system dependencies has been cleaned up. It's listed here because any code which uses omniidl's C++ interface may need to be slightly modified.

Summary:	omniIDL/C++ backend tie template operation mapping error (bug number 3)
Date:	Mon Oct 16 18:56:53 BST 2000
Reported by:	Chris Newbold
Description:	The tie template code used the _objref argument mapping rather than the _impl mapping.

Summary:	Race condition in POA triggers assertion failure (bug number 2)
Date:	Fri Oct 13 09:56:03 BST 2000
Reported by:	Chris Newbold
Description:	There is a race condition in the POA between deactivating an object and it being etherealised, which can trigger an assertion failure (which in turn throws an omniORB::fatalException()).

Summary:	C++ back-end claims that it does not support long long (bug number 1)
Date:	Fri Oct 6 09:58:55 BST 2000
Reported by:	Chris Knight
Description:	A new check added to the C++ back-end incorrectly reported that long long is not supported, even though it is. Note that long long is not currently supported in Any.