SDK Bug & Feature Requests

In the new SDK 3.7.6, it's now possible to build a world on Linux, but it is not possible to launch a local test with proton. Pls add proton support, similarly to: https://github.com/BefuddledLabs/LinuxVRChatSDKPatch

Following up https://feedback.vrchat.com/sdk-bug-reports/p/freeze-to-world-on-constraints-will-always-bake-offsets-if-there-are-0-sources A parent constraint with "Freeze to world" enabled before the object containing it is turned off will not properly snap back when "Freeze to world" is disabled and the object is turned back on (on the same frame), if "Freeze to world" is re-enabled a frame later. Repro: Two objects, X (i.e. the avatar's armature) and Y (droppable) Parent constraint on object Y has a single source to X, activated via "Zero". A toggle is linked to two animations: * False = Object Y is turned off (2 keyframes, both are the same, animation duration is 0.017). * True = Object Y is turned on, "Freeze to world" is disabled on keyframe 0, enabled on keyframe 1 (animation duration is also 0.017) Assumption would be that because the parent constraint "freeze to world" is turned off, it would make object Y snap back to object X. This works in the editor , but does NOT work in game. A further observation was then made - adding a 3rd keyframe ("keyframe 2", animation duration 0.033) that is a copy of "keyframe 1" (object on, freeze on) made it work in game, but removing "keyframe 1" (so the animation duration is simply longer) did not solve it, it appears that for whatever reason the duplicated keyframe is necessary for the expected unfreeze/freeze snap to work in game.

When you grab a physbone, IsGrabbed becomes true. Then when you pose it, IsGrabbed becomes false and IsPosed becomes true. However if you grab it while posed, IsGrabbed becomes true but IsPosed also remains true at the same time, until you let go of the bone. This seems unintentional, and adds and extra layer of complexity to designing logic that needs to check for IsPosed, since you must also check to see if IsGrabbed and IsPosed are true at the same time, which means it is grabbed. IsGrabbed = grabbed IsPosed && IsGrabbed = grabbed IsPosed && !IsGrabbed = posed IsPosed = maybe grabbed, maybe posed

what is fame

Make a stereo audio file with very obvious differences between the Left and Right channel Assign it to an AudioSource component Enter play mode (with ClientSim) You hear both channels Add a VRCSpatialAudioSource component to the same gameobject Enter play mode You hear only the left channel Do a Build & Test In the actual client, you hear both channels

Performing multi-platform builds with overrides set have several issues: They can fail if an avatar that's part of the overrides is disabled - block the build or re-enable targets temporarily. They can fail if the override does not pass upload limits for the target platforms (i.e. custom shaders on android) - the stats breakdown should display breakdown per platform (separate tabs/foldout), taking avatar override into account. There's apparently a way to cause an infinite upload loop (not sure how, but this was reported by someone who set the overrides incorrectly) Platform swaps occasionally fail (mostly commonly when it's the first time on a project)

With SDK version 3.7.5 several settings were migrated from the static class VRCSettings , existing within the VRCSDKBase-Editor.dll library provided alongside the SDK to VRCPackageSettings , the change also included a refactor of the class, introducing a static property Instance , used to get or create an instance of the class, an instance is attempted to be created by the PerceptualPostProcessor as DPID settings were part of the migration, however this fails because of a null exception: VRCPackageSettings.Instance.dpidMipmaps within PerceptualPostProcessor executes VRCPackageSettings.Instance_get VRCPackageSettings._instance is null so Create() is called, which calls Load() Load() and calls within it, including EnsurePathExists() expect GetPath() to not be null, causing a null exception when it is because GetPath() calls GetPathFromType(GetType()) which calls UnityEditor.PackageManager.PackageInfo.FindForAssembly(t.Assembly) and returns null when executed by a worker process, GetPathFromType then returns null (a comment assumes this would only occur when the SDK is not located within the Packages folder) but this execution path was seemingly never tested and so methods that expect valid strings such as Directory.CreateDirectory within EnsurePathExists() throw null exceptions. Simply resolving the null exceptions is not enough as the current implementation would cause Load() to not load any data when it's meant to, leading to VRCPackageSettings.Instance to contain default values when executed by worker processes ( AssetDatabase.IsAssetImportWorkerProcess() ), you cannot simply abort execution because that invalidates the use of Parallel Import and would cause textures marked as "dirty" to not actually be updated to have mipmaps re-generated, a proper solution leading to the same settings file must be implemented, the whole dependency on the assembly's package info seems extreme. This can be easily reproduced on any 3.7.5+ project with Parallel Import enabled (Project Settings -> Editor, mine is configured with 8 desired workers and 2 on standby) by enabling/disabling "Override Kaiser mipmapping" in the SDK settings.

I've noticed that the VRChat SDK occasionally causes the m_Automatic field in ProjectSettings.asset for iOS Graphics API to switch between 0 and 1, seemingly at random. This tends to happen on Unity launch or when entering/exiting play mode. It's not a serious issue since I can just reset the changes each time, but it's a bit annoying when working with version control, as it creates noise in diffs even when I haven't changed any settings myself. I'd appreciate it if the SDK could stick to one consistent value for this field to avoid unnecessary diffs. Either 0 or 1 is fine — I just want it to stop flipping back and forth. Unity 2022.3.22f1 on Windows with iOS build module VRChat SDK 3.8.1 (though this has been happening for a long time)

On most PC shaders, when Audio Link is set to control the emission, the emission is set to off when Audio Link is not in a world. for Toon Standard, the emission is set to full when audio link is not in a world. The emission should be overridden to off when audio link is not present in a world and audio link has been set to use the emission mask. This could also be an option in the shader, like a check box, so as to not break current logic that already gets around this issue.

In the attached video, I try to to set Primary Style as Object and Secondary Style as Furry for a new Avatar Dynamics Robot Avatar PC sample avatar upload in a new project with nothing but SDK - Avatars 3.10.5, but after the upload finishes both styles end up being Not Specified instead of the chosen information. Editing primary and secondary styles from the SDK works for existing uploads, but for new uploads. Affected SDK versions tested: 3.9.0 (minimum for new uploads), 3.10.3, 3.10.4, 3.10.5 (latest). Affects all avatars, and all combinations of primary/secondary styles (e.g., Human/Anime, Anime/Human, Object/Furry). I've run into this issue for some six months now, every week. Behavior in SDK 3.9.0 today is slightly different, but still buggy: In the SDK it shows the Primary Style is set to the chosen Secondary Style, and the Secondary Style as Not Specified, but after checking the VRChat Home website or tabbing in/out of Builder tab it goes back to Not Specified like in SDK 3.10.5.