[Opentnl-general] Two issues related with GhostConnection.

[Norbert H. <nh...@gm...>]

Hi.

First of all, thank you GarageGames, for putting TNL under GPL. I have
spent the last two days working around with it. This is really a fine
piece of code.

I tried to compile TNL & Zap with Metrowerks CodeWarrior 8. I had to
make a couple adjustments (maybe more to that in another e-mail), but
in the end, Zap seemed to run fine.

However the TNLTest didn't work. I tracked down two issues. I am not
completely sure if the description below is 100% correct, so it would
be great if somebody could look into to verify it:


Bug #1:

When TNLTest is started as a client, GhostConnection::mGhostZeroUpdateIndex
is never properly initialised. The VC++ DEBUG build initialises
it to a non-zero default value of 0xcdcdcdcd for example.

Now, somewhere down the road, the method NetConnection::checkPacketSend()
uses the isDataToTransmit() to check if a package should be sent. This is
however always the case, since GhostConnection::isDataToTransmit() checks
if mGhostZeroUpdateIndex != 0.

This leads to empty packages (besides header) being sent all the time.
The weird thing is, that this covers a second bug, which makes the app
just run fine when built with VC++.

I noticed this, because CodeWarrior initialises mGhostZeroUpdateIndex
to 0 instead of 0xcdcdcdcd. So it 'unfortunately' (read below, why)
behaved correctly.


Bug #2:

Now, let's say mGhostZeroUpdateIndex is correctly initialised to 0 and
no further empty packages are being sent.

However, the only time the TNLTest client sends a package to the
server is when the user sets a new position of his player,
(rpcSetPlayerPos() and rpcPlayerWillMove() is called). So if the user
doesn't do anything, there are no packages sent at all, and no ACK's
are ever sent back to the server.

When I turn on the LogConnectionProtocol logging, I see that no
Notifies are received by the server. I think this causes the server to
stop sending packages after a short time.

This was not a problem up to now, because the empty packages carried
the ACKs in their header. The behaviour can also be observed when
making a release build of TNLTest with VC++ (where
mGhostZeroUpdateIndex = 0).

I guess the solution would be to also check on pending Notfies in
isDataToTransmit(). I am not sure though. I have barely understood the
underlaying network mechanisms.


Thanks,
Norbert.