Being able to receive direct callbacks about UdonBehaviour exceptions with the relevant information included telling us which behaviour has crashed and what the exception message was would help us inform the players in the world immediately that something has catastrophically failed and they should rejoin the world or create a new instance. Not only that, we could also format useful error messages for users to copy from an InputField to include in their bug report without having to look through thousands of lines of logs.

When an exception occurs during Udon execution (the UdonBehaviour "crashes"), the component will get disabled and nothing in the Udon program will execute afterwards. As of now there is no way to access information within the world via Udon about what has happened apart from repeatedly checking the enabled state, which only tells us that something has gone wrong, but even then some programs utilize the enabled state by design for optimization and thus it becomes an extremely unreliable method of checking for halted UdonBehaviours.

Large worlds in VRChat can easily have over a thousand UdonBehaviours and repeatedly checking their enabled state to make sure everything is running as intended is not optimal and in most cases the only problem is the rare edge cases where a client/SDK update results in a regression or a player uses an avatar with an unsupported/unstable configuration which affects the world's stability.

...Yes, you could just keep adding more null checks like you're normally supposed to, but Udon suffers from severe performance limitations and in large frameworks every microsecond counts and iterating through behaviours is not a suitable solution.

Many players either never turn on their client logging for performance reasons or they don't even know about the existence of such, so not only would this help alleviate the process of handling user bug reports as they've already been clearly informed by the world about what has happened, it would also improve the general user experience as the features in the world don't just suddenly go unresponsive without any form of feedback.

One of the most frustrating experiences as a world creator is seeing someone encounter an UdonVMException in the middle of an 80 player event, resulting in a central world feature becoming unusable, and nobody involved had their logging enabled due to performance reasons or they didn't know how to enable it.

As a final note to clarify, I'm not asking for the implementation of C# exceptions into Udon, just an Udon callback informing about the VM encountering one. I'm completely okay with UdonBehaviours becoming permanently halted, but as VRChat is becoming more complex by the day with all of the new features, more opportunities for edge cases open up and being able to catch them in realtime without the end user having to even understand how to access logs on the platform would significantly improve not only the players' experience in worlds, but also the world creator's experience with creating content for the platform.

I've also attached examples of error messages targeted at end users that I've implemented across my products at the additional cost of Udon performance as a result of continuous scanning to ensure that UdonBehaviours critical to world's operation are still enabled.