Convex Hulls cause a memory leak

[ProbablePrime]

Describe the bug?

Using convex hulls causes a memory leak

To Reproduce

1. Create a new world
2. Spawn out a convex hull brush
3. Draw 5 or so Convex Hulls
4. Watch your memory use sky rocket

Expected behavior

Convex Hulls should perform normally.

Screenshots

No response

Resonite Version Number

2024.2.2.906

What Platforms does this occur on?

Windows, Linux

What headset if any do you use?

Index

Log Files

ALFRED - 2024.2.2.1015 - 2024-02-02 23_39_36.log

Additional Context

Worlds that use a lot of convex hulls are currently death traps. To avoid some trolling I will not list any but have provided a list in the internal team discord.

Panda also experienced this in just deleting convex hulls.

This might be the same issue as: #1290 or 1290 might be separate Charlie was participating in helping me reproduce but the issue looked different so I kept them as two items.

A video from Panda is here: https://cdn.discordapp.com/attachments/850883244501368855/1203246881379979274/Funny_Leak_Cause_Convex_Hulls.mp4

Reporters

So many people jumped in to verify this. Medra was the one who DM'ed me, but the others listed below helped reproduce, diagnose or just experienced the issue in the Resonite Down Under server.

Medra, ProbablePrime, Panda, Bredo, Sveken, Charlie-san, Beef, Unintentional, Canadian_RedZ, ocelot342

[ProbablePrime]

This is likely some sort of weird side effect of this change:

From the 2024.2.2.906 update.

[LeCloutPanda]

Here is my log file incase you need more logs in the future: GEOFF - 2024.2.2.1015 - 2024-02-03 18_49_01.log

[Canadian-RedZ]

Here is my log files also
Windows -> REDZ-LENOVO - 2024.2.2.1015 - 2024-02-03 02_31_29.log
Debian/Linux -> RedZ-Lenovo_-2024.2.2.1015-_2024-02-03_03_13_10.log

[Frooxius]

We'll have to report this to author of BepuPhysicsV2, it's probably the code not clearing up some memory it allocates, so it accumulates on repeated runs.

If anybody wants to have a look at the backported code before I get to it, it's here: Yellow-Dog-Man/bepuphysics2@64795ac

[Banane9]

This is likely some sort of weird side effect of this change:

I'm not sure about it being completely new - this could explain what happened about two weeks ago, where I got up to some very spicy RAM usage that also went up pretty fast 🤔

Here's a log in any case (puush because GitHub only allows 25MB and this was at the end of a long session I was hosting, where convex hull brushes came out at some point to show them off to new people):
https://puu.sh/K0ae6/346bdf5626.log

[Frooxius]

Is there a way to confirm that the leak before was also due to convex hulls specifically?

They can happen for a number of reasons, so I don't want us going down a wrong rabbit hole.

[Banane9]

Is there a way to confirm that the leak before was also due to convex hulls specifically?

Not specifically - it just reminded me of that occasion and I wanted to provide the log just in case it does have anything relevant.

[Frooxius]

Would anyone be able to jump on an older build (I think one in the pre-release might work) and see if this can be replicated with old BepuPhysicsV2 too?

That way we'd know that we have to look for source of this issue outside of the changed code.

[LeCloutPanda]

Is there a way to confirm that the leak before was also due to convex hulls specifically?

They can happen for a number of reasons, so I don't want us going down a wrong rabbit hole.

If you look at the video I provided to Prime it shows the leak is due to convex hulls because I used the bruh to cause the leak.

[Banane9]

If you look at the video I provided to Prime it shows the leak is due to convex hulls because I used the bruh to cause the leak.

Pretty sure Froox was just replying to my comment about it potentially happening before the update as well.

[LeCloutPanda]

If you look at the video I provided to Prime it shows the leak is due to convex hulls because I used the bruh to cause the leak.

Pretty sure Froox was just replying to my comment about it potentially happening before the update as well.

My bad you are most likely correct, I have an archived build from pre-release when I got access to the game as a mentor so I'll do more testing and note my findings when I do some testing.

[Canadian-RedZ]

Would anyone be able to jump on an older build (I think one in the pre-release might work) and see if this can be replicated with old BepuPhysicsV2 too?

That way we'd know that we have to look for source of this issue outside of the changed code.

Here is a test done with Steam's pre-release build 2024.1.21.825
-> https://github.com/Yellow-Dog-Man/Resonite-Issues/assets/20827894/c85f2280-75e0-40b6-bf70-2942f0191ffb
Here is the log from the same test -> REDZ-LENOVO - 2024.1.21.825 - 2024-02-03 11_14_03.log

[Nutcake]

We'll have to report this to author of BepuPhysicsV2, it's probably the code not clearing up some memory it allocates, so it accumulates on repeated runs.

If anybody wants to have a look at the backported code before I get to it, it's here: Yellow-Dog-Man/bepuphysics2@64795ac

It looks like when merging in the changes from upstream you accidentally left in a debug struct called DebugStep, which balloons up with huge amounts of data but has zero actual use beyond troubleshooting.
I've created a PR on your fork of BEPU that reverts that part back to upstream, fixing the issue: Yellow-Dog-Man/bepuphysics2#3

[bredo228]

Did a comparison of Build 2024.1.19.687 and Build 2024.2.2.1015

2024.1.19.687:

DYSNOMIA - 2024.1.19.687 - 2024-02-04 10_15_14.log

convex.hull.build.2024.1.19.687.mp4

2024.2.2.1015:

DYSNOMIA - 2024.2.2.1015 - 2024-02-04 10_18_39.log

2024.2.2.1015.convex.hull.mp4

[ProbablePrime]

Thank you all for the work!

It does look like: Yellow-Dog-Man/bepuphysics2#3

Will fix us up.

[Frooxius]

This has been merged in 2024.2.5.848, thank you!