EnchantmentsComponent (used by both regular and stored enchantments)
was directly casting the registry VarInt ID to the Enchantments enum
via (Enchantments)id. However, the Enchantments enum is ordered
alphabetically (AquaAffinity=0, BaneOfArthropods=1, ..., Sharpness=32)
while the 1.20.6 registry uses a completely different order
(protection=0, fire_protection=1, ..., sharpness=13). This caused all
enchantment names to display incorrectly (e.g. Sharpness V shown as
"Unknown Enchantment with ID: 32").
Changes:
- Parse now uses EnchantmentMapping.GetEnchantmentByRegistryId1206()
to properly map registry IDs to enum values via the existing
1.20.6+ mapping table
- Serialize now uses EnchantmentMapping.GetRegistryId1206ByEnchantment()
to convert enum values back to registry IDs (reverse lookup)
- Fixed translation key prefix: "Enchantments.minecraft." (wrong) ->
"enchantment.minecraft." (matches en_us.json resource keys)
- Fixed 3 long-standing typos in the Enchantments enum that prevented
translation lookup from matching resource keys:
- DepthStrieder -> DepthStrider (depth_strieder vs depth_strider)
- Efficency -> Efficiency (efficency vs efficiency)
- Loyality -> Loyalty (loyality vs loyalty)
Verified on vanilla 1.20.6 server: items with sharpness, efficiency,
unbreaking, fortune, mending, and bane_of_arthropods all display
correct localized names (锋利, 效率, 耐久, 时运, 经验修补, 节肢杀手).
Made-with: Cursor
Both components had incorrect Parse/Serialize implementations that would
cause packet deserialization misalignment when encountered in-game.
PotionContentsComponent (3 bugs):
- Serialize unconditionally wrote VarInt(PotionId) and Int(CustomColor)
even when HasPotionId/HasCustomColor was false. The official format
(PotionContents.STREAM_CODEC) uses Optional encoding: Bool(hasValue)
followed by the value only when true. The extra bytes caused all
subsequent fields in the packet to be read at wrong offsets.
- Serialize omitted the VarInt(count) prefix for the custom effects list.
The official codec uses ByteBufCodecs.list() which always writes a
VarInt count header before the list elements.
- Also fixed typo: PotiononId -> PotionId.
InstrumentComponent (3 bugs):
- The official Instrument.STREAM_CODEC uses ByteBufCodecs.holder() which
encodes as VarInt(holderId): 0 = inline data, N>0 = registry ref (N-1).
The SoundEvent field inside uses the same holder pattern. The old code
unconditionally read SoundName (ResourceLocation) and HasFixedRange/
FixedRange even when SoundEventHolderId != 0 (registry reference case
has no inline data).
- UseDuration was read/written as Float, but the official codec uses
ByteBufCodecs.VAR_INT. This caused a 4-byte vs variable-length
mismatch that would shift all subsequent data.
- HasFixedRange was read unconditionally when SoundEventHolderId == 0,
but FixedRange was also read unconditionally. The official SoundEvent
DIRECT_STREAM_CODEC uses Optional<Float> encoding: Bool(hasValue)
followed by Float only when true.
These components are used for potion items and goat horns respectively.
Verified against official 1.20.6 decompiled source:
- net.minecraft.world.item.alchemy.PotionContents (STREAM_CODEC)
- net.minecraft.world.item.Instrument (STREAM_CODEC/DIRECT_STREAM_CODEC)
- net.minecraft.sounds.SoundEvent (STREAM_CODEC/DIRECT_STREAM_CODEC)
- net.minecraft.network.codec.ByteBufCodecs (holder/optional/list)
Made-with: Cursor
Audited all 58 StructuredComponent subclasses against the official Minecraft
1.20.6 decompiled source to verify Parse()/Serialize() symmetry. Found and
fixed four bugs across four components:
1. ContainerComponent: Parse() skipped null item slots (empty slots in a
container) but Serialize() looped NumberOfItems times using Items[i],
causing IndexOutOfRangeException when any slot was empty. The official
ItemContainerContents uses OPTIONAL_STREAM_CODEC which serializes empty
slots as VarInt(0). Fixed: Parse now stores all slots including nulls,
Serialize uses Items.Count and iterates all entries. GetItemSlot(null)
correctly writes VarInt(0) for empty slots.
2. ChargedProjectilesComponent: Used Items.OfType<Item>() in Serialize()
which silently dropped null entries, causing the serialized count to
differ from the written VarInt header. The official ChargedProjectiles
uses STREAM_CODEC (non-optional, no empty slots allowed). Fixed: Items
list is now List<Item> (non-nullable), Parse defensively skips nulls,
Serialize writes Items.Count matching the actual list.
3. BundleContentsComponent: Same issue as ChargedProjectilesComponent.
Applied the same fix pattern.
4. FoodComponentComponent: Two type mismatches vs the official
FoodProperties.DIRECT_STREAM_CODEC:
- Saturation was declared as bool and read with ReadNextBool (1 byte),
but the protocol sends it as float (4 bytes). This caused all
subsequent fields in the component to be read at wrong offsets,
corrupting CanAlwaysEat, SecondsToEat, and the effects list.
- NumberOfEffects was serialized with GetFloat() instead of GetVarInt(),
writing 4 bytes of IEEE 754 float instead of a variable-length integer.
Fixed both Parse and Serialize to use correct types.
Also removed redundant NumberOfItems/NumberOfEffects fields from components
where the count is derivable from the list length, and replaced
ArgumentNullException with cleaner patterns.
Tested end-to-end on vanilla 1.20.6 server: item receiving (diamond_sword,
golden_apple, diamond_pickaxe), inventory slot movement (click to pick up
and place), and inventory listing all work correctly with no server-side
protocol errors.
Made-with: Cursor
After the previous commit (99ac3d0) moved attribute lookup from a hardcoded
dictionary to the dynamic RegistryData, MCC would crash immediately upon
joining a vanilla 1.20.6 server with:
System.ArgumentException: An item with the same key has already been added.
Key: unknown
Root cause: When KnownDataPacks negotiation tells the server that MCC already
has the "minecraft" data pack, the server skips sending RegistryData for
registries it considers "known" — including minecraft:attribute. This left
the dynamic attribute map empty, so every VarInt attribute ID resolved to
"unknown". The EntityProperties packet often contains multiple attributes
(e.g. armor, max_health, movement_speed), and `keys.Add("unknown", ...)` on
the second "unknown" attribute threw ArgumentException.
Two fixes applied:
1. World.GetAttributeNameById(): When the dynamic attribute map is empty
(server didn't send the registry), automatically load the vanilla 1.20.6
default attribute order (22 entries matching Attributes.java registration
order). This mirrors the pattern used for dimensions where defaults are
loaded when RegistryData is not sent. If a modded server sends a custom
attribute registry, the dynamic map takes precedence.
2. Protocol18.cs EntityProperties handler: Change `keys.Add(propertyKey,
propertyValue2)` to `keys[propertyKey] = propertyValue2` to tolerate
duplicate keys defensively, in case an unknown attribute ID still appears.
Tested: MCC now connects to a vanilla 1.20.6 offline-mode server, stays
online for 6+ minutes with no crashes or disconnections. Verified: chat
messages received, inventory listing (item names/counts correct), entity
detection, TPS query, and health query all work correctly.
Made-with: Cursor
In 1.20.6+, EntityProperties packets reference attributes by VarInt registry
IDs instead of string names. Previously, a hardcoded dictionary of 22 attribute
entries (matching the vanilla 1.20.6 registry) was used to map these IDs back
to names. This works for vanilla servers but would fail silently for modded
servers that add custom attributes — any unknown ID would be reported as
"unknown".
This commit replaces the hardcoded attribute dictionary with dynamic registry
parsing, following the same pattern already used for dimension_type and
chat_type registries:
- World.cs: Add static `attributeIdMap` field, `SetAttributeIdMap()` and
`GetAttributeNameById()` methods for storing/querying attribute names by
their VarInt registry IDs.
- Protocol18.cs (RegistryData handler): When the server sends a
`minecraft:attribute` registry during the Configuration phase, parse all
entries and store the ID→name mapping. The `minecraft:` prefix is stripped
from entry names to match the format used in EntityProperties packets
(e.g. "minecraft:generic.armor" → "generic.armor").
- Protocol18.cs (EntityProperties handler): Remove the hardcoded 22-entry
`attributeDictionary` and use `World.GetAttributeNameById()` instead.
Unknown IDs still fall back to "unknown" for safety.
Also closes issue #4 (Disconnect packet extra boolean) — verified that both
Play and Configuration phase Disconnect handlers already use `ReadNextChat()`
(NBT format since 1.20.4+), matching the 1.20.6 protocol spec. No code
changes needed; updated tracking document to mark as closed.
Made-with: Cursor
In 1.20.6+, items use structured components instead of NBT for metadata.
Previously, ReadNextItemSlot parsed the components but never stored them
on the Item instance, leaving DisplayName/Lores/Damage/Enchantments all
empty. GetItemSlot also still used the pre-1.20.6 format (bool + VarInt +
byte + NBT), causing the server to reject any item operation packets.
Changes:
Item.cs:
- Add List<StructuredComponent>? Components field to hold the raw
component list for round-trip serialization
- DisplayName property: read from CustomNameComponent (with
ItemNameComponent as fallback) when Components is present
- Lores property: read from LoreNameComponent1206 when Components is
present
- Damage property: read from DamageComponent when Components is present
- Add EnchantmentList property: read from EnchantmentsComponent (covers
both normal and StoredEnchantmentsComponent for enchanted books)
- ToFullString(): use EnchantmentList with EnchantmentMapping for display
when available, fall back to NBT path for older versions
- Add CloneWithCount() method that preserves both NBT and Components
DataTypes.cs - ReadNextItemSlot:
- Assign parsed strcturedComponentsToAdd to item.Components
DataTypes.cs - GetItemSlot:
- Add 1.20.6+ branch: write VarInt(count) + VarInt(itemId) + component
counts + serialized components (using each component's TypeId and
Serialize() method)
- Empty slot sends VarInt(0) per the 1.20.6 protocol spec
StructuredComponent.cs:
- Add int TypeId property (default -1) to store the registry type ID
assigned during parsing, enabling round-trip serialization
StructuredComponentRegistry.cs:
- Set component.TypeId = id after instantiation in ParseComponent()
McClient.cs:
- Replace manual Item constructor calls (new Item(type, count, nbt))
with Item.CloneWithCount() to preserve Components during inventory
operations like slot moves, stack splits, and right-click placement
Made-with: Cursor
The JoinGame and Respawn packet handlers for 1.20.6+ used hardcoded
switch expressions to map dimension type VarInt IDs to names:
0 => overworld, 1 => overworld_caves, 2 => the_end, 3 => the_nether
This only works for vanilla servers with exactly 4 default dimensions.
Modded servers (Forge/Fabric/NeoForge) or servers with custom
datapacks can register additional dimensions with IDs beyond 0-3,
causing the switch to fall through to the default "overworld" for
any non-vanilla dimension. This means players in modded dimensions
would have incorrect world parameters (height, lighting, etc.).
Fix: Replace both hardcoded switch expressions with
World.GetDimensionNameById(), which looks up the VarInt ID in
the dimension ID map populated during the RegistryData phase.
Also fixes two pre-existing issues in the SetDimension dispatch:
- JoinGame (pre-1.20.2 path): The `case < MC_1_20_6_Version` guard
was technically correct within its enclosing `if` block, but
changed to `default` for clarity and future-proofing.
- Respawn: The `case <= MC_1_20_6_Version` guard excluded protocol
versions above 766 (e.g. 1.21 / protocol 767), meaning
SetDimension was never called for those versions. Changed to
`default` so all versions >= 1.19 properly update the dimension.
Made-with: Cursor
Two critical issues in the 1.20.6 configuration phase that could cause
connection instability and packet desync:
1. RegistryData: The handler used an early `break` when it encountered
a registryId other than "minecraft:dimension_type" or
"minecraft:chat_type". This skipped reading the remaining entries
for that registry, leaving unconsumed data in the packet buffer.
Subsequent packet reads would start at the wrong offset, causing
cascading parse failures and eventual disconnection.
Fix: Always read all entries (entryId + hasData + optional NBT)
for every registry, regardless of whether we process it. For
dimension_type entries, if the server sends inline NBT data (i.e.
non-vanilla dimensions from mods/datapacks), parse and store
the dimension directly via World.StoreOneDimension(). Only fall
back to hardcoded defaults when no dimension data was received.
2. KnownDataPacks: The client echoed back ALL packs the server
listed, including non-vanilla ones. This told the server "I have
these packs cached" when the client actually did not, so the
server would skip sending full registry data for those packs.
The result: incomplete registries for modded/datapack content.
Fix: Filter the response to only include packs with the
"minecraft" namespace. Non-vanilla packs are omitted, forcing
the server to send their full registry data inline.
Also adds supporting methods to World.cs:
- SetDimensionIdMap(): Store VarInt ID -> dimension name mapping
from RegistryData entries (needed by JoinGame/Respawn)
- GetDimensionNameById(): Look up dimension name by numeric ID
- HasAnyDimension(): Check if any dimensions were loaded from
server-provided data
Made-with: Cursor
The 1.20.6 EntityProperties packet sends attribute IDs as VarInts
instead of strings. The existing mapping dictionary had three issues:
1. IDs 5/6/7 used the wrong prefix "generic." but the official
1.20.6 registry uses "player." for these attributes:
- 5: player.block_break_speed (was generic.block_break_speed)
- 6: player.block_interaction_range (was generic.block_interaction_range)
- 7: player.entity_interaction_range (was generic.entity_interaction_range)
2. IDs 22-24 (submerged_mining_speed, sweeping_damage_ratio,
water_movement_efficiency) do not exist in the 1.20.6 attribute
registry — they were introduced in 1.21. Their presence could
cause incorrect attribute resolution.
3. Direct dictionary indexing (attributeDictionary[id]) throws
KeyNotFoundException if the server sends an unknown attribute ID,
crashing the packet handler. Replaced with TryGetValue and a
safe fallback to "unknown".
Made-with: Cursor
* add miscellaneous fixes
* Fixed connecting to server when compression threshold is set to 0
The client assumes that 0 means disabled, when on a notchian (vanilla) server, it is possible to set the compression threshold to 0 (compress all packets).
* Try to capture all exceptions through Sentry
No exceptions are being logged through Sentry, so be more aggressive when sending exceptions
(cherry picked from commit eb1c2f5e771760fb3be32ffea79f8292adca92f1)
* Call OnSpawnPlayer packet when a player is spawned using the SpawnEntity packet
references #2721
(cherry picked from commit ef28ae09ac89e8988dd612de61f2849a9f0e528c)
* Add Sentry Error Tracking
* Omit personally identifiable information and add additional sentry context
* Remove debug message
* Make sentry opt-out and add related notices and strings
Also add Minecraft Version to error context
* Update build to send release info to sentry
* Adjust sentry error tracking
- Send the user-friendly Minecraft Version in the error logs
- Capture exceptions in more parts of the application
We now capture exceptions from the following locations:
- Protocol18 (1.8+) Packet errors
- Errors during client initialization phase (When client is about to start, session keys are NEVER sent to sentry)
* Make Sentry DSN configurable and repository-specific
The Sentry DSN will automatically be filled out on the main repository through the Github Actions build.
* Update build-and-release.yml
Update sed command
* style: change variable name
nitpick, just to make it a little bit more descriptive
* Add Sentry branding in README.
* remove old code (merge conflict)
The fix is to remove the ParseText call from the OnConnectionLost call, as the ReadNextChat function already calls ParseText. Calling ParseText on an unparsable string returns an empty string, therefore the disconnect message never gets propagated to the user.
