Release Notes for snapshot 971224                  - by Sai-Lai Lo
---------------------------------

A new snapshot is now available for download at:

   http://www.uk.research.att.com/omniORB/snapshots/snapshot_971224.tar.gz


IMPORTANT NOTE:
   The naming service now uses the prefix pragma "omg.org". This is
   introduced to bring our environment in line with the clarifications
   in the CORBA specification. Unfortunately, this results in on-the-wire
   incompatibility with clients that are linked with the old library, i.e.
   any programs linked before this version is released. Please read the
   section on omniNames below for details.


Summary of changes since snapshot_971120:
----------------------------------------

1. The Naming service now uses the prefix pragma "omg.org" in its IDL.

2. Support for egcs-1.0 with Redhat-5.0.

   Both x86 Redhat-5.0 and alpha Redhat-5.0 are supported.
   Notice that the unpatched egcs-1.0 does not have thread-safe exception
   handling. You *MUST* apply the thread-safe EH patch for it to work
   properly with omniORB2.

3. New hook for plugging in access control policy modules to determine 
   whether to accept new client connections.
   Unix platforms use host-based access control based on tcp_wrapper_7.6.
   Win32 remains unchanged, i.e. all incomings are accepted.

4. The port to HPUX with aC++ (B3910B A.01.04) is now completed (thanks to 
   Hugo Vandeputte and Marcel Ruff).

5. Working LifeCycle service support.

6. Fixes to bugs uncovered since snapshot_971120.


You should also consult the release notes of snapshot_971112 and 
snapshot_971120 for more instructions on building this snapshot.

---------------------------------------------------------------------------

omniNames
---------

The naming service now uses the prefix pragma "omg.org" in its IDL. 

Unfortunately, this is not compatible on-the-wire with clients that are
linked with the old library, i.e. any programs linked before this version
is released.

The implications of this change are as follows:

a) omniNames built in this snapshot *CANNOT* read the data file of the old
   omniNames. A new instance must be started from scratch. If you start
   omniNames and tell it to use the old data file, it will certainly crash.

   Possible migration path:
     In our lab, we are running the old and the new omniNames in parallel.
     The new instance is setup to run on a different port. Also the
     new tree is built to read by default a different configuration file:
     (/project/omni/var/omniORB_NEW.cfg instead of 
     /project/omni/var/omniORB.cfg) so that a client will pick up the
     IOR of the old or the new omniNames depending on whether it is compiled
     with the old or the new library. When all our services have been
     recompiled, the old omniNames will be withdrawn.


b) If you want to continue to use the old omniNames, you can undo this
   change with the following steps:
     1. Edit <top>/src/lib/omniORB2/Naming.idl and remove the 
        #pragma prefix "omg.org" line.
     2. Do the same to <top>/idl/Naming.idl.
     3. Remove <top>/include/omniORB2/Naming.hh.
     4. Rebuild everything. You should do a gnumake/make veryclean at the
        top of <top>/src to make sure that the naming service stubs are
        regenerated.


---------------------------------------------------------------------------

egcs-1.0
--------

Egcs-1.0 must be patched with the thread-safe exception handling patch by
Teemu Torma for it to generate the correct code for running omniORB2.

With the exception of the seq_tests and seq1.idl, the build passes all the
tests in the omniORB2_testsuite. There is a compiler bug in egcs-1.0 that
make it unable to compile some of the sequence templates in omniORB2. The
bug might be fixed in the latest snapshot of egcs but I'm not certain.

Teemu Torma's patch is reported to be working on Solaris, HPUX and AIX as
well. Andrey Slepuhin has run the omniORB2_testsuite with egcs on AIX. Let
me know if you have success in running the testsuite on other platforms.


I've left my binaries for x86 Redhat linux 5.0 and alpha Redhat linux 5.0
in:
      ftp.uk.research.att.com:/pub/linux/egcs-1.0.alphabin.tgz  (4M) and
      ftp.uk.research.att.com:/pub/linux/egcs-1.0.x86bin.tgz    (4M)

The diff against the unpatched egcs-1.0 is in:

      ftp.uk.research.att.com:/pub/linux/egcs-1.0-thread-safe-eh.diff (60K)

The diff is based on Teemu Torma's patch + patch for RedHat 5.0 by H.J. Lu
+ my bug fix to get the thread-safe-eh stuff to work with linux-threads (or
any draft 10 or later posix thread implementations).

For x86 Redhat linux 5.0, you must update binutils to 2.8.1.0.17.

If you are building the compiler yourself, you have to apply this patch to
the resulting spec file:

23c23
< %{pg:-lgmon} %{pg:-lc_p} %{!pg:-lc} %{pthread:-lpthread}
---
> %{pg:-lgmon} %{pg:-lc_p} %{!pg:-lc} %{pthread:-rpath /usr/local/lib/pthread -lpthread}

This is to ensure that the thread-safe-eh version of the stdc++ library is
picked up when the executable is run.


---------------------------------------------------------------------------

Host-based access control
-------------------------

A new hook is provided for plugging in access control policy modules to
determine whether to accept new client connections.

On unix platforms, an access control module based on Wietse Venema's
tcp_wrapper_7.6 is implemented. Based on the client's IP address, the server
can decide whether to accept or reject the connection. The access control
policy is defined in 2 files: hosts.allow and hosts.deny. The default
location of the files are in /etc. Alternatively, the pathname of these
files can be specified in the GATEKEEPER_ALLOWFILE and GATEKEEPER_DENYFILE
options in the omniORB configuration file (omniORB.cfg). E.g.

GATEKEEPER_ALLOWFILE   /project/omni/var/hosts.allow
GATEKEEPER_DENYFILE    /project/omni/var/hosts.deny

The syntax of the files are described in the man page in
<top>/src/lib/omniORB2/gatekeepers/tcpwrapper/hosts_access.5

---------------------------------------------------------------------------

LifeCycle service support
-------------------------

Support for lifecycle service is now included (thanks to Duncan
Crisby). The code in <top>/src/examples/lifecycle demonstrates how this is
used.  The documentation <top>/src/examples/lifecycle/lchowto.ps contains
further details.

---------------------------------------------------------------------------

Bugs fixed
----------

The following bugs are fixed (except bug 12 which is MS VC++ specific and
has no workaround):


(1)

// If a forward declared interface is used in a constructed type, and
// that constructed type is used in a Union, Digital C++ complains about the 
// non-existance of CORBA::release() for that type.
// (This is used in the destructor of the interfaceName_member template -
// Digital C++, unlike other compilers, uses CORBA::release() instead of 
// T_Helper::release() - a previous bug with templates in Digital C++ ? )
// --- fixed
 
interface intf1;

struct aStruct {
  intf1 an_intf1;
};


union unionb switch(long)
{
 case 1:
   aStruct a;
};


(2)

// If typedef to forward declared interface, code is generated that tries to
// create aliases for _proxy_, _sk_, and _nil_, which don't exist.
// -- fixed

typedef intf1 alias_intf1;


(3)

// If a forward declared interface is used in a Union, the member function
// to set that interface uses interfaceName::_duplicate(), which may not
// exist if the interface is defined in another IDL file. Should use 
// interfaceName_helper::duplicate() instead (have fixed this).
// --- fixed

union aUnion switch(long) {
 case 1:
   intf1 an_intf1;
};


(4)

// Incorrect code generated for anonymous arrays and sequences in exceptions.
// --- fixed

exception anException {
  long arrayLong[10];
  sequence<long> seqLong;
};


(5)

// Incorrect marshalling code generated for marshalling multi-dimensional
// arrays of object references and for calculating size of the message using
// NP_alignedSize. Generates a[_i0]._ptr[_i1]._ptr instead of a[_i0][i1]._ptr
// (Have fixed the marshalling code for this, as it affects Any insertion
// operator)
// Digital C++ also complains about ambiguous type-conversions when 
// marshalling a multi-dimensional array _slice of object references 
// (both with and without the fix).
// --- fixed

interface intf4 {
  void anOp();
};

typedef intf4 multiIntf4[10][20];

interface intf5 {
  void anOp(in multiIntf4 a, out multiIntf4 b);
};

(7)

// Typedef to type Object yields typedefs with only one argument for
// _proxy_ , _sk_, and _nil_ .
// --- fixed

typedef Object aliasObject;


(8)

// Union incorrectly places an array of fixed length Unions into 
// an anonymous Union (whether the array is anonymous or not).
// SunPro C++ complains:  
//  "Error: A union member cannot require initialization or destruction."
// --- fixed

union Union1 switch(long) {
 case 1:
   long a;
};

typedef Union1 arrayUnion1[10];

union Union2 switch(long) {
 case 1:
   arrayUnion1 b;
};

union Union3 switch(long) {
 case 1:
   Union1 b[10];
};

(9)

// A sequence of arrays of sequences produces an incorrect declaration in the
// stubs header (.hh) file.
// --- fixed

typedef sequence<long> arraySeqLong[10];
typedef sequence<arraySeqLong> seqArraySeqLong;

#if !defined(TEST_EGCS) || (EGCS_MAJOR != 1 || EGCS_MINOR != 0)
// The stub code of the following cannot be compiled by egcs-1.0 because of
// a compiler bug.
interface intf7 {
  seqArraySeqLong op(in seqArraySeqLong a1,
		     inout seqArraySeqLong a2,
		     out seqArraySeqLong a3);
};
#endif


(10)

// Defining a typedef to an array will result in _alloc(), _free(), and
// _dup() functions being declared for the alias. These functions are declared
// as extern and implemented in the stub header file. This results in 
// multiply-defined clashes at link-time between the stubs skeleton (SK.cc) 
// file and any file including the stubs header (.hh) file. Fix is to declare 
// the functions as inline or friend inline.
// --- Fixed

typedef long arrayLong[10];
typedef arrayLong aliasArrayLong;

(11)

// An anonymous array in a union called d will result in the generation of 
// an alias called _d, which clashes with the _d() used to get the discriminant
// value of the union
// --- fixed

union sampleUnion switch(long) {
case 1:
long d[10];
};


(12)

// MSVC++ specific
// If a multi-dimensional array of object references is used as the in
// argument of an operation, MSVC++ 4.2 can't convert the type as necessary 
// when the upcall to the implementation is made in the dispatch function.
//  --- Not fixed

interface intf8 {
void op();
};

typedef intf8 multiIntf8[2][4][6];

#if !defined(TEST_MSVC)
interface intf2 {

void multiOp(in multiIntf8 inArg);
};

#endif

(13)

_CORBA_Sequence template copy ctor should really allocate buffer according
to the max size of the original.

(14)

Can now use OMNIORB_CONFIG environment variable on Win32 platforms to tell
the ORB to read in the nameing service IOR.