as seen in pmmp/BedrockBlockUpgradeSchema@ebd768e5b2, this enables use of newFlattenedName in more places (by allowing the flattened values to be transformed before building the new ID), as well as reducing the number of remappedStates in general by compacting stuff which was partially transformed like color silver -> light_gray.
ErrorToExceptionHandler currently prevents these from appearing by turning them into exceptions, but this won't always be the case.
For example, in the future we may not want to turn ALL types of E_* errors into exceptions (e.g. E_DEPRECATED).
we don't actually care about the specific values, only whether all the blocks and their states have been correctly registered.
I'd prefer to track all of the state data permutations, but the APIs for that are private, so tracking the number of permutations will have to suffice (this should be good enough to detect bugs anyway, and also takes way less space).
Because ext-phar sucks, tmp gets spammed by cache files for every thread when loading files from the phar on the fly.
Instead, we convert the `.phar` into a decompressed `.tar` in the tmp directory and require files from inside it. Surprisingly, this works because `ext-phar` supports `tar` and `zip` natively. No stream wrapper is required, as the `PocketMine.php` bootstrap loads files relative to its location, so the cache is automatically used for everything.
To be honest I would rather get rid of phars entirely, but they are still the easiest way to have PhpStorm load PocketMine-MP API information for now, and the alternatives are more complicated and inconvenient.
### Caveats
Everywhere that previously used `new Phar(Phar::running(false))` in the core code needs to be updated to use `PharData` for this to work correctly. Plugins don't need to do anything.
### Why not just use `Phar::decompressFiles()`?
This requires setting `phar.readonly` to `0`, which is a security issue. Technically, we could have used a subprocess to do this, but it just didn't seem right.
### WTF? `phar://` can be used on `tar` files???
Yup. I was just as surprised to find out that `require` works in such contexts.
### Relevant issues
- Closes#6214
## Changes
### API changes
None.
### Behavioural changes
Server startup will be slightly slower, as the phar has to decompress and convert itself into a `.tar`. However, testing showed that this generally takes less than 200 ms, so it should be barely noticeable.
## Backwards compatibility
No BC issues.
## Tests
Locally tested and the CI will also verify
while this would be more user-friendly as a config option, configs are a pain because they aren't initialized until after the server log has already been set up.
In any case, I foresee that people will likely want to bake this into Dockerfiles directly anyway.
`server.log` is not rotated by default and grows unmanageably large, to the point where it's so huge that it's not possible to read it with any standard text editor anymore.
This PR implements automatic log rotation.
- When the `server.log` reaches 32MB in size, it's moved to the `log_archive/` folder of the server's data folder.
- The archive's file name will look something like this: `server.2024-03-15T15.26.24.0.log`
- The file's name contains the date and time when the file was archived. This may be useful if you're trying to find logs from a particular time frame.
This has several benefits:
- Much more easily find logs from a particular time frame without scrolling through GBs of logs
- Free up space without stopping the server - Archived log files in `log_archive/` can be safely deleted and/or modified while the server is running
If you want to automatically compress or clean up the log files, I suggest an external cron job or disk watcher.
Closes#4029.
again, very annoying that we have to account for this (it makes it easier for attackers to waste CPU time and memory), but we don't really have much of a choice.
some persona skins are insanely bloated and get split into hundreds of parts.
it's quite annoying that we have to accommodate this, but we can't keep allowing players to experience login timeouts without an obvious indication what's wrong.
Prior to this commit, upgrade schemas would be applied to blockstates with the same version, as there wasn't any way to know if they should be applied or not given Mojang's tendency to forget to bump the version.
However, it occurred to me that we only need to do this if there are multiple schemas present for the same version ID, which is rarely the case.
This allows skipping costly logic for blockstates on the newest version (the common case), reducing the time taken to process the blockstate for upgrading by over 30%.
Overall, this translates into less than 10% real performance improvement for chunk loading, but it's still a worthwhile improvement.
