Re: Stub returns non-NULL return value even when a CORBA server isn't available.

[Jules Colding <colding omesc com>]

Hi Dmitry,


On Tue, 2007-09-25 at 18:07 +1200, Dmitry Maksyoma wrote:
> Hello,
> 
>   When I call a method of a CORBA server, but that server is down, I still
>   get a return value that is non-NULL. 

You should get an exception if the server is down. A TIMEOUT or a
TRANSIENT is the most normal in this case depending on how the server is
down. Section 1.2.2 in the CORBA spec (both 2.3 and 3.2) says in
excerpts:

"The return result and the values stored into the output and
input-output parameters are undefined if an exception is returned"

So ORBit2 can do whatever it wants to the return value and it is still
standards compliant.

HTH,
  jules




> Stub code declares return value as a
>   pointer and doesn't set it to NULL. Later, the code that is called from
>   stub doesn't change that ret variable, as a result, whatever uninitialized
>   value was there in the beginning, is returned to client code.
> 
> Here's stub code:
> /*
>  * This file was generated by orbit-idl-2 - DO NOT EDIT!
>  */
> 
> #include <string.h>
> #include <stdio.h>
> #define ORBIT2_STUBS_API
> #include "cmd.h"
> 
> Cmd_OctetSeq* Cmd_run_cmd(Cmd _obj, const CORBA_char * cmd, const Cmd_StringSeq* argv, const Cmd_OctetSeq* input, CORBA_Environment *ev){
> Cmd_OctetSeq* _ORBIT_retval;
> gpointer _args[3];
> _args[0] = (gpointer)&cmd;
> _args[1] = (gpointer)argv;
> _args[2] = (gpointer)input;
> 
> printf("stub start: %p, %p\n", _ORBIT_retval, &_ORBIT_retval);
> ORBit_c_stub_invoke (_obj, &Cmd__iinterface.methods, 0, &_ORBIT_retval, _args, NULL, ev, Cmd__classid, G_STRUCT_OFFSET (POA_Cmd__epv, run_cmd),
> (ORBitSmallSkeleton) _ORBIT_skel_small_Cmd_run_cmd);
> printf("after ORBit_c_stub_invoke: %p\n", _ORBIT_retval);
> 
> return _ORBIT_retval;
> }
> 
> This behaviour was introduced somewhere after 2.12.2, the version I used for
> the above testing was 2.14.3.
> 
> Because of that behaviour, it's now necessary to check whether exception has
> happened, before releasing memory. In 2.12.2 it was sufficient to check
> whether pointer is non-NULL.
>