Re: <Rhamphoryncus> gib lart id * gib cats /dev/urandom into id's ear

[Adam O. <rh...@d2...>]

On Wed, Oct 17, 2001 at 09:16:16AM +0000, Adam Olsen wrote:
> There's a bug in Netchan_OutOfBandPrint, in which it uses strlen
> instead of strlen + 1, causing the terminating nul to be stripped.
> The client doesn't notice because it never bothers to check for short
> reads with connectionless packets.  And I checked, it's in the
> original q1source.  So the net result is that my modified
> MSG_ReadString, which returns "" incase of error instead of a copied
> string with termination added, is not able to parse connection
> packets.
> 
> It's not the end of the world though.  I think I should be able to
> leave CL_ConnectionlessPacket using a (renamed) version of the old
> MSG_ReadString, and make the rest of the engine use my modified
> MSG_ReadString.
> 
> Hmm, or I could make it append a \0 to the buffer incase it's not
> terminated, since I know that'd work in this case.  I dunno how well
> that'd work in general though.  And it's kinda hacky.

Well, I did the "append \0" method, because it means simpler code (no
duplicate functions), and I don't have to pick and choose between two
said functions.  It seemed to work well initially, but after a bit of
spectating on prozac I reconnected and got this:

Fatal error: SZ_GetSpace: overflow without allowoverflow set (0)

I realize now that calling MSG_WriteByte won't work in overflow
situations, since I want it to tollerate overflows, but what concerns
me more that maxsize is 0.  I've no idea how it could get that, while
still getting data in the firstplace:

(gdb) p *buf
$2 = {allowoverflow = false, overflowed = false, 
  data = 0xbfffe7ac "ÿÿÿÿgetchallenge\n", maxsize = 0, cursize = 17}

Any ideas?

-- 
Adam Olsen, aka Rhamphoryncus