PseudoKnight
commented
10 months ago I replaced player injection in the player_spawn event with a new workaround that also addresses an additional issue with world changes during a quit event: https://github.com/EngineHub/CommandHelper/commit/b39b8f0b8e6784977e4414f8b90329cac8773e8c. Using this in the abstraction implementation for Spigot is a more appropriate location to workaround a bug in Spigot. Unfortunately this doesn't just affect getting players by name, but also all_players(), so we might want to write a workaround for that as well.
Some events pass a player object for players that are missing from the player list during the event. This causes a CREPlayerOfflineException when getting these players from Static.GetPlayer() or Static.GetCommandSender(), as well as made it tricky to pass the player to the bind environment. This was originally solved by hijacking the player injection system used for capture_runas(), but it wasn't designed for this.
Now, after some changes, the player is added to the environment from MCPlayerEvent instead, but injection is still used for getting players by name (and as a fallback for environment adding). But this happens in AbstractEvent, and this is undesirable as it mixes MC and MethodScript stuff. This will need to be changed some day in the future, but extensions would need to be considered.
Two events use injection still: player_login, where the player is not yet added to the player list at all; and player_spawn where the player is missing from the byName player map but not the byUUID player map (byUUID determines whether the player is "online" or not), which means they're technically online but Static.GetPlayer() can't find them without injection. I consider the player_spawn situation a bug, so it might be better to move the workaround for this to the bukkit abstraction layer.
As for player_login, the player is technically offline, so I'm not sure what the expectation should be here. Due to this, currently you can run functions from player context but not by name, except functions that take a CommandSender instead. (e.g.
has_permission(<playerName>, <permissionNode>)) We could change Static.GetPlayer() to take an environment object for situations like this (is player_login the only one?), but that'll affect a lot of functions and some will malfunction in unknown ways for truly offline players. I haven't done a comprehensive test.