[Freeciv-Dev] Re: [PATCH] bug in vnotify_conn_ex() (PR#1185)

[Jason Short <vze2zq63@xxxxxxxxxxx>]

Ross W. Wetmore wrote:

> You still don't understand that ENOEVENT (-1,-1) is a perfectly valid
> case, whether or not your topology allows (-1,-1) as a real tile. It 
> is required that this be passed on. There are other cases as well. The
> value is used as a signal, independent of topology hardwired into lots 
> of code.
> Thus you don't fix the original bug, but just kill the server at a
> different point. The original problem was ENOEVENT (-1,-1) in a torus
> world if you read the comments. They also explain ENOEVENT syntax.
>
> (-1,-1) *is* a special case that *must* just be passed through. All
> the asserts that specifically target this are dead wrong. This "fix"
> is not only not required but breaks valid behaviour. In fact (-1,-1) 
> is the only unnormalized coordinate you absolutely *must* not touch.
> Both Raimar and your zeal to stamp out unnormalized coordinates is 
> exemplary, but misplaced in this case. Leave this code out of your 
> targets! 
I think you are misreading the code.  (-1,-1) _is_ treated as a 
special-case for a nonexistent tile.  It fails the check and so is 
passed through as (-1,-1) directly to the client.
The problem is that if (-1,-1) is passed in as a legitimate set of 
coordinates, it will be treated as the special-case of no coordinates 
associated with the event.  Thus it is very important that we stamp out 
non-normal coordinates being passed to this function!  Of course, we 
must allow (-1,-1) to be passed but if any _other_ non-normal 
coordinates are passed in then we're in danger.  This is a bug in the 
calling code that can only be caught here.
jason