Re: ORBit2 bug (I think)

[Michael Meeks <michael ximian com>]

Hi Bill,

On Tue, 2002-05-28 at 19:19, Bill Haneman wrote:
> I'm running into trouble in the java-access-bridge, seeing the following
> messages on the client side:

	:-)

> WARNING: Dropping a fragmented packet on the floor!
> WARNING: Hyper unusual code path of little testing
> 
> (then segv, usually, sometimes something more graceful)

	Yes - as expected pretty much.

> So it looks like a message body is getting fragmented.  Is there a limit
> on the length of CORBA_String that can be passed?  [I sure hope not, as
> that would require at-spi API change to return some sort of TOO_LONG
> error if you requested too long a buffer].

	Well - no, you shouldn't change the API, but we should really implement
support for fragmented packet collation; something that we just havn't
got round to (as yet) - I don't believe it'd be the most difficult thing
in the world to implement, but moderate pain.

> To see this fun, try running Stylepad with the java-access-bridge and
> at-poke, and poke the text element.

	:-)

> The getText call on the servant side is trying to return what looks to
> be a valid, null-terminated string (albeit a long one, 1694 bytes long),
> but it never makes it to the client.

	Sure; the Java ORB probably helpfully fragments it into a number of
smaller packets for some reason, perhaps they use fixed size internal
buffers smaller than that order or somesuch. ORBit2 allocates its
buffers dynamically, and always sends a full message - depending on
memory constraints - which on reflection is perhaps not the best way to
do things, but we have some strange backwards re-writing stuff.

	It might be interesting for us to move to writing fragments as well in
certain circumstances.

> I will try turning on some debug spew in ORBit2, I expect that might
> help trace this.

	Oh; sadly not really - you'll not de-marshal enough to get useful data
- it's a known issue - we need to support fragmented packets.

	I'm currently busy to the hilt on nautilus / other critical crasher bug
closage, but Mark might be able to do it for you.

	Regards,

		Michael.

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