This was suggested recently by @TheDeibo. We don't want users running source-code installations unless they are developers, and developers should know how to boot a source-code installation anyway.
this process of fast-serialization, fast-deserialize, network-serialize is an order of magnitude slower than just doing the network encode directly on the main thread, and also copies more useless data.
For the main thread, the figures were something like 3x more expensive, and then an extra 7x for deserialization on the worker thread. This is a ridiculously large overhead.
this is perfectly fine to use, and preferable to getting a cyclic ref to the scheduler. TaskScheduler->cancelTask() does pretty much the exact same thing, and the scheduler internals are designed to deal with this anyway.
this should produce some reduction in spam at the source.
This could also be used to control the rate at which constantly-crashing servers restart to stop them spamming the disk as well, but the main concern here is eliminating crash archive involuntary DDoS by crashy servers.
This can happen when a light source is removed and later encountering another light source to fill the gap. A higher light level may get set and then not propagated. This bug is difficult to explain, but fairly easy to reproduce.
Grass can cause issues here by requesting blocks randomly offset away from itself, which can cause silent chunk loading on chunk ticking. It also causes crashes if chunk autoloading is taken away, which is obviously undesired.
It was also noticed that player chunkloaders cause chunks to start getting ticked as soon as they load their first chunk, which is before the entity is visible to everyone else on the server. This is probably undesired behaviour.