Client-side memory leaks in 1.12

[jchung01]

EDIT: I'll update this list as I find more.

I've been doing some memory profiling on my client on a large modpack, and I figured I would list some mods in 1.12 that seem to be causing memory leaks by holding onto WorldClient and/or EntityPlayerSP, noticeable when changing dimensions or reconnecting (as that creates a new client/player):

• CB Multipart: known leak, documented here.
• MrTJPCore: pretty much the exact same leak as CB Multipart, documented here
• New Tardis Mod: leak in ClientProxy due to adding players to a map that is unused (because this mod was in alpha for 1.12)
• Cosmetic Armor Reworked: possible leak in PlayerRenderHandler. It uses a timed eviction cache which could be the issue (more recent version use weak keys)
• JourneyMap (ARR): Possible leak in journeymap.client.task.multi.RenderSpec, I don't have much to back this up though besides what JProfiler says
• Bibliocraft: Not really a memory leak in the strict sense, but does hold the first WorldClient (for whatever reason) of the session on world load/connect due to VersionCheck

EDIT:

• Simply Jetpacks: Leak in SyncHandler maps client side - Server side, these maps are cleaned up properly here and here, but client side also updates these maps and those events are never called client-side, so the client-side maps are never cleaned up.
• OpenComputers: Network leak for all(?) packets due to closing the ByteBufInputStream of the data, but not releasing the actual data.
• CodeChicken Lib: Network leak for PacketCustom due to not releasing its data.
• Iron Chests: Retains first WorldClient each on first time rendering crystal chest or crystal shulker box due to using a reusable EntityItem
• Compact Machines: Retains client's current world when each miniaturization recipe is first viewed
• Ender IO: Retains first WorldClient for each obelisk renderer due to using a reusable EntityItem

The first 3 (bolded) were the most consistent in holding the WorldClient, and although Tardis should be easy enough to fix (by just disabling adding to the map), I have tried a few ways to fix CB Multipart and MrTJPCore, but they are in Scala, so even if a simple fix is to use a WeakHashMap, I don't know how to apply it externally.
Some screenshots from a heap dump:

[jchung01]

Other memory leaks that recently got fixes by RLMixins that can be added to UT as well as I'm still not sure if it's compatible with MixinBooter 8:

• BetterNether: see this
• BetterQuesting: see this
• MoBends: various fixes, see MoBends files

[ACGaming]

Highly doubt that mixins for Scala imports would be easy, so I included them in separate forks.

[jchung01]

I think there may be a way to properly cleanup the maps using packets, so I'll look into it while looking into SimplyJetpacks.

[ACGaming]

Other approaches might be possible, yes. But don‘t want to mess with Scala more than needed!

[jchung01]

Added fix for TARDIS and Simply Jetpacks in above PR. It would indeed require messing with a lot more files in Scala with MrTJPCore and CBMultipart and their dependents, so looks like the forks are the way to go.

[ACGaming]

Maybe we could add this to the list as well?
xJon/Tekkit-2#43

[jchung01]

@xJon Should be addressed in #496. Hopefully there's no bugs with my fix.

[xJon]

@jchung01 Thank you 💯 (for some reason GitHub did not send a notification)