KrepAPI (Mod) 1.0.1 для Minecraft 1.21.11 (Fabric)

KrepAPI 1.0.1-Mod

Issues that have been fixed

ByteBuffer with Fixed-Size Allocation — No Bounds Check

In ProtocolMessages, ByteBuffer.allocate(65536) is used for bindings and 512 for Hello/ClientInfo. If a payload grows larger (e.g. many bindings with long strings), there's a silent BufferOverflowException crash. No dynamic growth, no try-catch.

c2s_key_action Handler is Empty

In the Fabric server code, the receiver for KrepapiKeyActionC2SPayload is completely empty — registerGlobalReceiver(..., (payload, context) -> {}). Key presses on the Fabric server are therefore silently ignored. This isn't a bug if you're using Paper, but it's entirely undocumented for Fabric server users.

No Input Validation in ProtocolBuf.readUtf()

Without a length limit on UTF string reads, a malicious client can send arbitrarily long strings in actionId or modVersion. This leads to uncontrolled heap growth during decoding. Standard practice in Minecraft protocol code: a fixed max-byte limit per string.

Dispatch Doesn't Cancel on First Match

In KrepapiKeyPipeline.dispatch(), consume = true is set when a listener returns true, but all subsequent listeners still run. This is a deliberate non-early-exit, but it means: a low-priority listener cannot set consume back to false afterwards — the variable is never reset. As a result, all downstream listeners always run to completion, even if the event was already consumed long before.

volatile Fields Instead of AtomicReference / Final Config

requireClientOnDedicatedServer, minimumModVersion, and handshakeTimeoutTicks are all volatile. For simple primitives this is correct, but for Strings (like minimumModVersion), volatile only protects the reference, not the contents — though this is sufficient here since Strings are immutable. Still stylistically questionable compared to a clean config class.

Lexicographic Version Comparison

With minimum-mod-version: "1.9" and then "1.10" — lexicographically "1.9" > "1.10" (since '9' > '1'). The project documents this as a known limitation, but it's a real footgun for server admins. No real SemVer parser.

HANDSHAKE_TICKS Map Doesn't Clean Up Reliably

In the tick event, HANDSHAKE_TICKS.remove(id) is only called on e.answered or timeout. If a player disconnects before the timeout fires, the entry is removed in the DISCONNECT event — but only from HANDSHAKE, not explicitly from HANDSHAKE_TICKS (which has its own separate remove). Fine in isolation, but with many rapid joins from unmodded clients the map could grow.

No Unit Tests

None. The :protocol module would be ideal for encode/decode roundtrip tests — it's independent of Minecraft, pure Java. Tests would be especially important for the ByteBuffer handling and the version policy logic.

Release Phase Not Wired Up

KeyAction.PHASE_RELEASE is defined in the code and transmitted in the protocol, but the client-side KeyMapping never fires a release event. Documented, but still an incomplete feature.

Notes

• Install a Fabric API build that matches the Minecraft version.

KrepAPI (for developers)

• Build version (Fabric mod release / c2s_client_info.modVersion, gradle.properties → mod_version): 1.0.1
• Wire protocol version (KrepapiProtocolVersion.CURRENT, protocolVersion in s2c_hello / c2s_client_info): 1 — increment only when the binary packet layout or semantics change