A common source of crashes on modded Minecraft servers comes from modders accidently calling client only code from the client, this PR is another large step towards elimitating that.
This PR has been months in the making and years in the planning, requiring major changes to Loom & Loader. In recent Minecraft versions Mojang has made it easier than ever to cleanly split the jar, going against the status-quo of merging the client and server into one jar.
From the start we have designed Fabric to have a very clear split between client and common (client & server) code. Fabric has always encoraged keeping client only code seprate from the server, this can be seen at a fundamental level with the entrypoints in Loader. Fabric API's have all been designed with this mind.
This PR provides a compile safety net around Fabric API using client only code on the server. Even though there are almost 400 changed files, minimal changes beyond moving the files were required to achieve this in Fabric API, thanks to the effort of all contributors in the past.
These changes should not affect modders or players in anyway, a single "universal" jar is still produced. Im happy to awnswer any questions.
* Update command registration API to reflect vanilla changes
* Allow module versions in impl+mixin packages
* Use v2 module, keeping v1 for server commands at least
* update, address review
Co-authored-by: modmuss50 <modmuss50@gmail.com>
* Networking api v1
Some final docs?
Licenses and testmod
Fix a bunch o imports and make things work for v1 (v0 is bork)
Make the testmod pass checkstyle and work
Docs for v1
* Deprecate v0 and implement using v1
* Drop files down one package due to package check error
* Fix issue with channel registration, add another testmod
* jaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaavadoc
* Make javadoc use `code`, move impl interface to package access
* this things
* Rename a few internal methods
* Mark all client side stuff client only, move client mixins
* Add null checks around the place, clarify some javadoc and method names
* Make FutureListeners uninstantiable
* Some internal nullable annotations
* An impl class I forgot to rename
* Some comments and clarify some client login handler javadoc
* Add a missing FunctionalInterface annotation
* Split play and login, move client stuff to right package
* No interface left behind
* Inline channel registries in api
* Login and play subpackages not needed
* Add helper method to create play custom packets
* hasGlobalChannel -> hasGlobalReceiver
* Just rename the collection method for now
* Inline PlayPacketSender into static methods
* Start on testmod idea for verifying dynamic registration
* Add client login events
* You don't say hello when talking to yourself.
Also more testmod stuff
* Make event names present tense
* Some javadoc and impl interface rename
* Change the test keybinding
* Begin working on dynamic reg
* Dynamic reg works, just need a lot of cleanup and reimpling global
* A few renames, readd global methods
* Try to reduce the amount of duplicate registration logic
* Reimplement dynamic accessors
* More impl
* Start reimplementing global receivers. Still very hacky solution.
* Reimplement some server global reciever stuff
* Add login init event for server login.
* Implement client login query start event
* Move event invocations into addon, don't dual register global recievers
* Finally reimplement global recievers for all networking phases
* A revelation: Send packets properly
This also finds the issue with screen getting the proper S2C channels, current on TODO list.
* Disconnect event does not need a packet sender
* Clarify, add methods to get channels net handler can recieve on client
* Unregister actually works now
* Bunch of null checks, add simpler login delay test for vanilla clients
* Add some debug logging entries, fix unregister on client's session reg
* Play channel event javadoc and rename login query handlers
* More channel -> channelName
* thisening
* Introduce the basics infrastructure for tracking global receivers
* Add more substantial javadoc to login connection events
* Javadoc, reimplement unreg methods on v0, 1 impl fix
* Implement tracking for global recievers
* Dont forget to start tracked sessions in 3/4 cases
* Global receiver docs and move methods in classes
* Complete null checks
* big boi javadoc part 1
* Finish the main javadoc, usage javadoc is left
* Set so has method is not needed
* Rename receiveable and sendable methods
* Add the two missing private ctors
* buildscript update to upstream
* Split out player finding stuff to networking player tracking API v1
Signed-off-by: liach <liach@users.noreply.github.com>
Forward v0 PlayerStream to new module, add entity track events
Rename module to player tracking
Well javadoc can make sense
Decide on tracking for the name
Update fabric-player-tracking-api-v1/src/main/java/net/fabricmc/fabric/api/networking/player/tracking/v1/package-info.java
Co-authored-by: Erlend Åmdal <erlend@aamdal.com>
Remove exceptions from javadoc that are not thrown
javadoc fix again
Handle a case where the player manager happens to be null
rename player tracking to player lookup
Yeet
* Cherrypick #1092
* Remove some redundant client networking methods, rename `(un)register` to `(un)registerReceiver`
* Simplify access to dynamic reg on client
* Param shifting, let users get sender.
* Warning about time and distance units
* Make sure these are client only
* Fix control flow in ClientPlayNetworking#send
* Correct example code javadoc
* javadoc correction in server login
* Put login delay tests behind system property
Also remove unnecessary junk added by old module that was merged together.
* Fix ordering so channel registrations during `PHASE`_INIT work
* Fix prod bug and an oversight
* Fix login when connecting to dedicated server
* Update registry sync to v0 to prevent issue with reg sync hanging client
* this is done
