Unity.Mathematics.Random not consistent per seed when called within IEnumerator?

Unity Engine
, , ,
1

Hello all.

I believe I have discovered a bug, whether that be in my own code or in unity’s, or I’m using it improperly.
I have some code that I am using to generate a level. I am generating the level in a Coroutine. I want to be able to slow this generation down to see it in-action to explain to people what’s happening easier. For this I implemented a way to control when my coroutine yields: DebugInfoSettings.GenerationSpeed. When this is 0, I skip a lot of yields and only hit a few yield return nulls. When it is not 0, I use this value at several points with yield return new WaitForSeconds(DebugInfoSettings.GenerationSpeed). When this value is not 0, Unity.Mathematics.Random starts returning different numbers than when it is 0. It seems as if anything above 0 returns the same end result, but 0 returns a different end result. I believe this to mean that somehow yield return new WaitForSeconds is the culprit.

Here is the code snipet that it breaks in:

// remove rooms based on rules
for (int i = 0; i < generationInfo.Side; i++)
{
   for (int j = 0; j < generationInfo.Side; j++)
   {
   byte gnr = GetNeighborCount(i, j);

   if ((gnr <= removeRules.LimitLTET || gnr > removeRules.LimitGT) && !(i == generationInfo.Side / 2 && j == generationInfo.Side / 2) && random.NextFloat() <= removeRules.ActionChance && rooms[i, j] != null)
   {
     Object.Destroy(rooms[i, j].gameObject);
     Log($"Destroyed By RemoveRooms: {rooms[i, j]}");

     if (debugInfoSettings.GenerationSpeed != 0)
       yield return new WaitForSeconds(debugInfoSettings.GenerationSpeed);
     }

     dd.Add(random.state);
   }
}

Here is a hastebin link to the entire file: Hastebin

I have debugged this by writing to a file whenever I generate a level, writing the random.state whenever it is possibly used. I generated a level while GenerationSpeed was 0, and while it was not 0, then iterated over every number until i found when random was not in the same state. Here is the code I used to save and test the files in case you guys want to see it: Hastebin

Below I have attached two JSON files (converted to .txt files) created with the debugging class I wrote, file1 is with GenerationSpeed at 0, file2 is with GenerationSpeed at 0.001. Both were generated with a seed of 4. For this instance it broke at 4,439 uses of my random instance. If I run this test again with a seed of 3, it breaks at 5,052 uses instead.

I am using Unity version 2022.3.7 and Unity.Mathematics version 1.2.6. Any help is appreciated!

9260538–1295622–file1.txt (567 KB)
9260538–1295625–file2.txt (567 KB)

2

This could be something as simple as missing array element assignment to null after destruction on L149 and L273.

1 Like
3

You use Object.Destroy inside your remove rooms loop in like 149 and your loop relies on a null check on array elements. As you hopefully know, Object.Destroy is delayed until the end of the current frame. So when you introduce wait times in between, of course the behaviour will change.

I just realised Spy-Master just beat me to it… :slight_smile: Yes, manually setting the array elements to null when you destroy the objects should fix that issue. It’s a common Unity gotcha.

4

oh my god I can’t believe this is what it was. I must have accidentally undone it because I had it there before, but at one point I undid a lot. Thank you.

1 Like