Compatibility Report

after further debugging, we've determined the issue to be with Garbage Collection. For some unknown reason, Unity's garbage collector (BoehmGC) is unable to collect garbage fast enough. Unity's memory debugging tools show all of the leaked memory coming from "Reserved Managed Memory", which it describes as being memory that is allocated, but not used by any Unity objects

we've been able to reproduce this issue on a fresh vanilla wine prefix. Wine 8 seems to be the most recent version that is unaffected by this issue

Just to add to this issue. I have strong reason to believe that the same problem is happening with Warudo as well. With Proton 8, I observe no memory leak, but with Experimental (or anything newer than 8), there is a memory leak completely identical to this. I'm not sure about VNyan, but on Warudo, if you add PROTON_USE_WINED3D=1 to the launch options, it forces the memory leak to show up faster and makes the memory usage increase faster than without it.

The issue has been traced back pretty definitively for VNyan, and it's suspected to affect nearly every Unity program. The bug is related to syscall handling. VNyan triggers sleep syscalls ~5000x per second, as uOSC triggers sleeps ~1000x per second from busy waiting, and VNyan runs uOSC on 5 ports. Warudo also uses uOSC, which would explain it having the same issue

This is the relevant wine bug: https://bugs.winehq.org/show_bug.cgi?id=59333

We have a temporary fix for modern wine versions by patching Unity's garbage collector, and a proper fix in wine is being discussed actively

We suggest changing the title of this to something like "Leak in Unity Garbage Collection causes Full System to Hang".

While VNyan and Warudo are definitely affected by this, the issue spans to all of Unity in general. We've seen the same behavior out of a few games in Unity after playing them for extended time.

We wouldn't be surprised if this bug is causing issues for a significant portion of Unity Games running beyond Proton 8, and is the cause of countless unidentified crashes.

Fix patches backported to Proton are here. This is submitted upstream here, but hasn't been reviewed yet.

The bug dates back to the dawn of time, including Proton 8 and older versions. However, Proton/Wine 9 regressed it in a way that makes it much more likely to happen.

All Unity applications that use managed threads are affected. However, how likely it is to happen depends heavily on how many threads there are, how busy those threads are, and what specifically they are doing. The race likelihood is proportional to the number of Wine syscalls made per second across all managed Unity threads. VNyan happens to score the highest here (due to reliance on a very poorly designed polling thread loop in uOSC, across multiple instances). However, the bug has been definitively observed with at least Warudo and VTube Studio too.

In Wine/Proton 8 and older, only certain specific syscalls/flows could potentially trigger the race. Since Wine/Proton 9, it's every single syscall. I also suspect there is a different presentation, where the same general issue causes the GC to prematurely free an in-use object, which would cause application crashes instead of memory leaks. However, that one hasn't been identified definitively as happening in the wild, it's just theoretically possible.

I surmise that this has also hit people across huge numbers of Unity apps, it's just rare enough that likely most people never saw it consistently enough to identify it as a Proton issue and not, say, a random crash or app bug.