Re: in-kernel ORBit2 works

[Michael Meeks <michael ximian com>]

Hi Dou Wen,

On Sat, 2002-01-05 at 17:55, dou wen wrote:
> i have ported the ORBit2 2.3.100 to linux kernel 2.4.17 , the port is very coarse, 
> i just palce it in kernel in module, with echoserver in whole.

	Wow - quick work.

> i test it with user-space echo-client and it seems work fine.

	Nice.

> but there is a strange problem, in function giop_recv_buffer_demarshal_request_1_2 of
> giop-recv-buffer.c , before return , i must do this :

	That is very odd indeed.

> .....
>  buf->msg.u.request_1_2.service_context._buffer = NULL;
>   if(giop_IOP_ServiceContextList_demarshal(buf, &buf->msg.u.request_1_2.service_context))
>     return TRUE;
>   buf->cur = ALIGN_ADDRESS(buf->cur, 8);
>   buf->cur +=4;/*<---------------* here, i dont know why must plus 4/
>   return FALSE;
> }
>  
> if buf->cur dont plus 4, the params decode error value,use echoString for example,

	It looks like an offset problem somewhere; are you sure that on the
marshaling side that you are allocating memory aligned to 8 byte
boundaries ? it's possible that your malloc is returning it aligned to 4
byte boundaries, since alignof (double) is only 4 on intel; but ... the
ORB assumes the memory returned by malloc will be 8 byte aligned [ quite
possibly - this is something we should work on / check. ].

> it is a strange problem.

	Indeed.

> i intend to place it to sourceforge,maybe nextweek, because there are
> lot of trival thing to do and i am busy these days.

	Great.

		Michael.
 
--   mmeeks@gnu.org  <><, Pseudo Engineer, itinerant idiot