System.ArgumentException at ORTS.Viewer3D.RenderFrame.Sort (X2162 crash switching fast from car to car)

[Carlo Santucci]

While being in passenger camera and switching fast from car to car I got a crash. Logfile attached.

[Carlo Santucci (carlosanit1)]

The problem occurs at least since version x2146, but up to today never occurred, while today it always occurs.

[Carlo Santucci (carlosanit1)]

The problem has arisen passing from x2130 to x2131 (a modification where I have a noticeable responsibility...) and happens only with some wagons and does not seem to depend from speed in passing from one wagon to the other. Will have to investigate further...

[Carlo Santucci (carlosanit1)]

After further analysis I found out that the crash does not arise in x2164 loaded from SVN, while it arises in x2164 loaded from James'site... Now I'm clueless.

[James Ross (twpol)]

Chances are, we have a threading hazard somewhere. Therefore, even the machine you run OR on might be relevant to how likely this is to occur.

[Carlo Santucci (carlosanit1)]

This patch solves the problem. I think it is not a threading hazard.
In my opinion the arrangement that was inserted to discriminate between near objects not only leads to crash; it neither solves the problem of discriminating between near objects. For this reason I think that the problem of discriminating between near objects does not need to be addressed until a proof against this arises.

[James Ross (twpol)]

The problem with that patch is that it throws away and ignores the data specifically included to ensure closely-positioned items are drawn in the correct order. That isn't acceptable for obvious reasons.

[Carlo Santucci (carlosanit1)]

Ok, second trial.
This patch maintains the original comparison algorithm and only compacts it. The crash has disappeared.

[James Ross (twpol)]

Having just translated the error message for the first time, this crash makes no sense - with any version of the code. I can only assume floating point rounding is rearing its ugly head somehow.

I also can't get a crash here at all. :(

[James Ross (twpol)]

This is a version of RunActivity/RunActivityLAA with the changes in the patch but also Math.Sign() replaces by a typecast to (int). Please let me know how it works out before I commit it.

[Carlo Santucci (carlosanit1)]

You can e.g. download:
- this high speed trainset: http://www.trainsimhobby.net/infusions/pro_download_panel/download.php?did=985 :one of its passenger views crashes to me
- or download this narrow gauge trainset: http://rhbformsts.spoorpunt.nl/EWIV.php?m=2 where the passenger view of trainset RhB_BD2475.wag crashes to me.

[cjakeman (cjakeman)]

May be a long shot, but this stack trace looks very similar to the inner exception part of the one reported in Bug 1300942:

Error: System.InvalidOperationException: Failed to compare two elements in the array. ---> System.ArithmeticException: Function does not accept floating point Not-a-Number values.
   at System.Math.Sign(Single value)
   at ORTS.Viewer3D.RenderItem.Comparer.Compare(RenderItem x, RenderItem y)
   at System.Collections.Generic.ArraySortHelper`1.QuickSort(T[] keys, Int32 left, Int32 right, IComparer`1 comparer)
   at System.Collections.Generic.ArraySortHelper`1.Sort(T[] keys, Int32 index, Int32 length, IComparer`1 comparer)
   --- End of inner exception stack trace ---
   at System.Collections.Generic.ArraySortHelper`1.Sort(T[] keys, Int32 index, Int32 length, IComparer`1 comparer)
   at System.Array.Sort[T](T[] array, Int32 index, Int32 length, IComparer`1 comparer)
   at System.Collections.Generic.List`1.Sort(Int32 index, Int32 count, IComparer`1 comparer)
   at System.Collections.Generic.List`1.Sort(IComparer`1 comparer)
   at ORTS.Viewer3D.RenderFrame.Sort()
   at ORTS.Processes.UpdaterProcess.Update()
   at ORTS.Processes.UpdaterProcess.DoUpdate()
   at ORTS.Processes.UpdaterProcess.UpdaterThread()
   at System.Threading.ThreadHelper.ThreadStart_Context(Object state)
   at System.Threading.ExecutionContext.Run(ExecutionContext executionContext, ContextCallback callback, Object state)
   at System.Threading.ThreadHelper.ThreadStart()
for which we may have a fix,

Chris

[James Ross (twpol)]

Chris,

I don't believe they are related, although the crash is very close-by. In this bug we're returning non-zero from RenderItem+Comparer.Compare when x == y which seems to be fixed by rearranging the order of operations. In bug 1300942 we're coming in to that function with NaN values which blow up differently.

[James Ross (twpol)]

This bug should be fixed in X2178. Thanks Carlo!