make bench (tools/bench.sh, tools/bench_report.gd) measures the server's
p50/p99/max tick with 24 bots on three seeds over a whole match, and the
client's draw calls, primitives, objects, nodes and memory at six moments
of a match rendered in the capture container. It fails on anything over
the budgets now recorded in docs/ARCHITECTURE.md: p99 tick 12 ms, 2,000
draw calls and 2,000,000 primitives per frame (all passes), 1 GiB video
and 512 MiB static memory. Every capture now reports its frame's render
counts in a "render" field.
The first run put the barge, with 24 figures together, at 2,197 draw
calls. Each figure part had one surface per colour (about 43 per figure,
drawn again for shadows). FigureRig now merges a part's surfaces into one
per finish (coat, gilt, plain) and carries the Palette colours as vertex
colours, with shared materials. The barge is down to 1,207 draw calls,
with the same triangles and the same look.
Frame rates on a real GPU are not measured here; that stays with R6.
Closes#54
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Weather.at(tide, tick) derives the weather from the tide's schedule
alone, so every client sees the same sky with nothing new on the wire:
the storm builds from calm at the first tide to a gale at sudden death,
and the light runs from mid-afternoon to dusk at the time limit.
WeatherView applies it each sync to the sky, sun, fog, ambient light,
the sea's roughness and rain around the camera (CPUParticles3D, for
both renderers). Under water the view closes into SEA_MURK, the rain
stops and the surface is a rippled ceiling seen from below.
The sea shader darkens open water and adds chop and whitecaps with the
storm, and keeps a breathing foam band at the waterline with a fainter
line beyond. In the Compatibility renderer, whose depth texture the
shader can't read (the sea came out all foam), water depth comes from
a texture of the terrain's heights instead. Sea colours move into
Palette; the rain and wind beds follow the weather.
Tests: test_weather (7). make check passes (611 tests). Captures:
make sea-captures (every tide phase, calm beach, the last land in the
gale, under water, and three in the Compatibility renderer).
Closes#46
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The islander is built in headless Blender (tools/models/figure.py,
`make figure`, `make art`) as twelve rigid parts with their origins at
their joints: an oilskin coat, a sou'wester, sea boots, a bedding roll
and the canvas wing-sail. Debian's Blender can't write glTF, so the
figure is jointed rather than skinned.
FigureRig hangs the parts on pivots, paints them from Palette with the
coat in the player's own oilskin, and plays FigurePoses' hand-written
key poses through one AnimationTree: idle, walk and run by speed,
crouch and creep, air, skydive, glide, swim and aboard; long, short and
carried holds; reload and use layers; fire, boathook swing, keg throw
and building as one-shots; and a flinch when health drops. The chest
takes a third of the pitch and the held item all of it.
Figures animate from replicated state only: positions, look, health,
the held item and the FLAG_ bits (reloading and using now reach
mirrored players), with actions from FX. Every boathook swing is now a
sim "swing" event, sent as a SHOT with no item, so misses are seen and
heard; the new FX kind PLACE shows pieces placed. Protocol version 4.
Tests: test_figure_rig (15), plus swing and PLACE cases in test_harvest
and test_net_fx. make check passes (604 tests); make client-check
passes. Captures: figure_states, figure_actions, figure_close
(make figure-captures), client_glide and client_landed.
Closes#42
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/models/items.py builds every item in headless Blender from
primitives: the boathook, the eight weapons and powder keg, four
consumables and five kinds of ammo, each with its own outline. It
writes an OBJ per item (the Debian Blender's glTF exporter needs numpy),
a JSON with the grip and the left hand's support point, and a 128 px
Cycles icon rendered on the CPU. Materials are named after Palette
colours, and TRIM2_/TRIM3_ surfaces are the tier treatment: rope
whipping from tier 2, brass bands from tier 3; tier 0 is faded.
ItemModels paints the models from Palette by tier. EntityView puts the
held item's grip in each figure's right hand, turned with the player's
pitch and put away aboard, falling, gliding or swimming, and draws
items lying in the world with their models, one MultiMesh per model
and tier. The hotbar shows each slot's icon.
make models rebuilds them; make item-captures renders the items held,
over the shoulder, on the ground and by tier.
Closes#44
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every sound is made by scripts in tools/audio/ and committed as a WAV in
assets/audio/ with a manifest of hashes and levels:
- dsp.gd: seeded noise, wavetable oscillators, filters, envelopes and
struck modes. recipes.gd: 50 effects (footsteps on five surfaces,
a report per weapon, impacts, carpentry building and breaking, the
tide's bell, foghorn, surge and lighthouse, player, ambience, menus).
- score.gd: an original eight-bar tune and chords in D Dorian, 6/8,
rendered as four looping layers (drone, frame drum, concertina,
fiddle).
- loudness.gd and check.gd: per-category loudness targets, a -1 dBFS
peak ceiling and the mix order (the tide loudest). make audio and
make audio-check.
In the client, SoundDirector turns FX records, footsteps, building,
the tide and the player's own hurts into sounds, placed on 3D voices
at the event so gunfire and footsteps come from their direction.
MusicDirector fades the score's layers with the tide on an
AudioStreamSynchronized, full in menus. MenuSounds knocks for buttons.
Closes#48
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
make soak runs a real server and 24 headless clients in three processes for
two whole matches in real time. Each client reaches the server through its own
UdpRelay, which adds 60 ms, +/-20 ms jitter and 2% loss each way, so ENet runs
unchanged over a bad link. SoakPlayer drives each client from what a client
knows: it drops, glides, heads for high ground, harvests and tries to build.
tools/soak_report.gd checks world hashes against the server's, errors,
unreadable packets, every client playing every match with its connection
intact, memory between matches, and the p99 tick with the network's share.
make soak-selfcheck proves a drifted client world is caught.
Server changes: --matches=N, --world-hash-every=N, tick timing that includes
after_tick, and memory in match and progress lines.
What the soak found and fixed:
- Snapshot building for 24 clients took the p99 tick to 13.8 ms, over the
12 ms budget, against a 2.8 ms simulation. Encoded records are now shared
between clients for the tick, unchanged records skip the field diff, and
clients are spread over the snapshot interval by handle. p99 is 7.3 ms.
- ENet peers had a timeout limit of 0, which gives up on a peer as soon as any
reliable message has waited the minimum timeout. Now ENet's default, 32.
Not fixed: one or two clients still lose their connection about 15 s into the
second match's drop in each run (see the issue), so the soak still fails and
N8 stays open.
Refs #39
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A hostile test client crafts INPUT packets byte by byte and sends them
through NetGame.take_input() into a real match: extreme axes, the same
tick claimed repeatedly, the whole lead window at once, past, far-future
and maximum ticks, fire toggled every tick with slot swaps, builds aimed
anywhere with junk kinds and materials and no materials, and
pickup/drop/select tricks. Speed stays within sprint speed per tick,
shots within the weapon's rate, pieces within build reach and paid for,
and every item is conserved.
- Fix: a command tick beyond the simulation's lead window no longer moves
NetGame's dedupe marker, so one bogus far-future tick can't lock out a
player's real commands.
- Connections are rate-limited per address (32 per 10 s, 8 handshakes in
progress) with a new REASON_BUSY and a message for it.
- ParserFuzz mutates real messages from a bot match and random bytes and
feeds them to every network decoder, the snapshot receiver and the list
server. 20,000 per make check; make fuzz runs 1,000,000 (about 40 s):
no errors, 2.9 KB of memory growth.
- ARCHITECTURE.md: "Trusting nothing from clients".
Closes#40
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- LAN discovery: a listening server answers QUERY packets on its game port
plus one (ServerBeacon, off with --no-lan). ServerBrowser scans eight game
ports from 24650 by broadcast.
- A self-hostable list server: --list-server [--port=N] (ListServer,
default UDP 24660). Game servers register with --list=HOST[:PORT] and
--name=NAME; they are listed under their source address only after
echoing a CHALLENGE token. Entries expire after 45 s, 8 per address,
256 in all; every source is rate-limited (SourceLimiter).
- NetDiscovery messages: every QUERY is padded to 512 bytes and no answer
is bigger, so neither service amplifies spoofed requests.
- tidewall --browse [--list=...] [--lan-target=IP] [--lan-port=N] prints
the servers found as JSON until menus (A10) show them.
- People who connect mid-match or during results watch a living player
(Match.watch) and take places ahead of bots in the next lobby, instead
of being told the server is full. Bots fill only the places left when
the match starts.
- Tests: codec round trips, size limits and a 40,000-message fuzz; list
server with fake servers (challenge, stolen and wrong tokens, renewal,
expiry, caps, flood); LAN scan over loopback; watchers over UDP; lobby
backfill. make discovery-check runs it all as real processes.
Closes#37
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The client program now turns a connection into a match someone can play.
- ClientSession: a ClientGame, the keyboard and mouse as commands
(ClientInput), and a MatchView of the client's own copy of the world.
- ClientMirror writes what the client is showing into that world every
frame: other players from the recent past, its own player where it was
predicted, items, the barge, the clock and the sea. The existing views
draw it unchanged.
- LocalServer: offline play is the same ServerLoop and NetGame in the same
process, with packets handed across in memory, so there is one path.
- main.gd: --connect=HOST[:PORT] joins a server, otherwise an offline
match; --autoplay and --client-seconds=N for scripts and checks.
- The camera aboard the barge follows the player's own place on the deck.
Tests drive a scripted client through a whole match on a real server over
loopback UDP, lobby to results, and through an offline match. `make
client-check` does it with two real processes, and `make client-captures`
renders what the client sees.
Closes#77
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Built pieces, props' health, opened chests and the tide's schedule now reach
clients as WORLD messages on the reliable channel, and each player is told
what they carry as OWN messages.
- NetWorld: the WORLD and OWN wire formats, and world_hash(), a digest of
everything replicated this way.
- WorldSender: one message after each tick in which something changed, the
same for everyone, and the whole state for someone joining or a new match.
StructureSet and PropSet keep the change lists it reads.
- ClientWorld: applies them to the client's world, so pieces and broken props
are solid for prediction. A piece's health follows from its building and
damage, so building costs nothing on the wire.
- Placing ahead of the server: a provisional piece goes up with the command
and is replaced by the server's, or taken back once a snapshot shows the
server had its chance. Snapshots carry the newest WORLD number for that.
- Protocol version 3.
A test plays a whole scripted match on a real island with a ClientGame
attached and compares the two worlds' hashes throughout.
Closes#36
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A network client draws other players in the past and its commands take time
to arrive, so by the time the server fires a shot the target has moved on:
8 ticks at a 50 ms round trip, 12 at 200 ms.
Each PlayerCommand now carries view_lag: how far back its player was seeing
everyone else, in quarter ticks. The simulation keeps where every player's
body has been for the last 15 ticks (HitHistory), and a hitscan shot is
traced against other players where they were at the moment the shooter saw,
and against the world as it is now. Projectiles aren't rewound.
The client claims only when it was looking and where it aimed; the server
works out the hit from its own record. The moment is never more than
Tuning.LAG_COMPENSATION_MAX_TICKS back, never in the future, and never
earlier than a moment the same player has already claimed.
Commands are 15 bytes, so the command log's version is 2. The lag is in the
log, and replays judge every shot as the match did.
Measured in test/net/test_lag_compensation_link.gd over a simulated link:
a client firing at a walking target as drawn on its own screen hits 41 of
41 shots at 50, 100 and 200 ms round trips, 0 of 41 with the lag ignored,
and 0 of 41 at 500 ms, past the cap.
Closes#35
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The client's side of a networked match, in src/net/:
- ClientGame takes the server's notices and snapshots, generates the island
from the seed and answers with commands. It works on packets, so it runs
over a NetClient or a test's simulated link.
- ClientClock decides which tick the next command is for, from the server's
echo of the newest command tick it has received, holding three commands
waiting at the server.
- ClientPrediction runs each command on a copy of the client's own player
with the server's movement code, and compares every snapshot with what it
predicted. A difference is replayed from the server's state and eased in.
- SnapshotInterpolator draws everyone else four ticks in the past, between
snapshots, on a clock steered gently by arrivals.
Snapshots now carry the client's own block: exact position and velocity,
the water around the player, stamina and state flags, so a prediction can
match the server to the bit. The command echo is now the newest tick
received rather than taken. Protocol version 2.
Snapshots go out every second tick. The interval was three ticks, which is
10 a second, not the 20 the documents claimed; 15 a second is the nearest
even spacing and what interpolation is tuned for.
Drop.ride() holds the barge ride for the server and the client alike, and
NetGame's joining, input and snapshot steps can be called without a socket.
Measured in test/net/test_prediction_link.gd over a simulated link with
100 ms each way, 20 ms jitter and 2% loss: no corrections to own movement
in about 1,400 snapshots per client, and the other player drawn 0.4 mm on
average from where a perfect link would draw them.
Closes#34
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
NetSnapshot defines the game traffic after the handshake: delta snapshots of
quantized entity records, the client's input packet and a reliable match
notice. SnapshotSender builds each client's snapshot against the newest one
it acknowledged, tells it only about entities near the player it is watching,
and keeps every snapshot under 1,200 bytes, nearest first, with a staleness
bonus so nothing starves. SnapshotReceiver rebuilds the picture, ignores old
and repeated snapshots and asks for a fresh start if it is ever sent a delta
on something it doesn't hold.
NetGame joins the transport to the match: people who connect become
participants, their inputs go through ServerLoop.submit() once per tick, and
snapshots go out 20 times a second. `--server --listen` makes the headless
server take network players and wait in the lobby for the first of them.
Measured by test on a 24-player match start with every player a client, a
tenth of packets lost and late acknowledgements: 2.4 KB/s mean and 2.5 KB/s
for the worst client, packet headers included, against a 40 KB/s limit.
Closes#33
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ENet over UDP, used directly through ENetConnection, with a handshake before a
connection counts as a player.
- NetProtocol: HELLO, WELCOME and REJECT layouts, the protocol version, name
cleaning, the reasons a connection ends and a sentence for the player about
each. Decoders return nothing for malformed input and never raise an error.
- NetServer: listens, takes each guest through the handshake, and reports
joined, left and refused. It refuses a different protocol version, a full
server, a malformed HELLO and a guest that stays silent, each with its own
reason. Handles are never reused; duplicate names get a number.
- NetClient: connects, introduces itself, and ends up joined or closed with a
reason. Leaving says goodbye so the server knows at once; a client that
vanishes is dropped by ENet's timeout.
- Tests run a real server and clients over loopback UDP, with bare ENet peers
sending malformed and random packets.
The headless server doesn't listen yet; N2 connects this to the match.
Closes#32
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`make match` plays one whole match from the lobby to the results and back to the
lobby: one scripted player and 23 bots on a local server, with no script errors.
- CommandLog records the seed, the rules and every command the server is given.
MatchReplay plays a log again with nobody thinking and ends on the same state
hash; `make match` saves the log, reads it back and checks that.
- ServerLoop.submit() is now the one way commands reach the simulation, so a
recording server sees them all.
- ScriptedPlayer plays a human seat: it drops, loots, fights and flees as a bot
does, and also cuts driftwood and builds a breakwater and a wall.
- Bots draw randomness from their own per-bot generators, not the simulation's
streams. A bot thinking no longer changes the simulation's state, which a
replay without bots needs.
- MatchView and EntityView draw a match from one Simulation: terrain, sea,
points of interest, props, pieces, players, the barge, items and chests, with
stand-in shapes for the art pass to replace.
- `make match-captures` renders eight stages of the recorded match.
Closes#31
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
BotBrain plays a match with the commands a player sends: it picks a landing
marker near the barge's route, loots points of interest (following their own
walking routes into huts, lofts and piers), fights opponents it can see,
bandages itself, and gets to ground the coming tide will leave dry. BotWorld
is the map every bot shares: loot markers, routes, and each tide phase's safe
ground, which is the dry land joined to the island's highest walkable ground.
Three difficulties (Tuning.BOT_DIFFICULTIES) set reaction time, sight range
and aim error. The server mixes them unless --bot-difficulty says otherwise.
NavGrid labels connected regions when it is built, so a goal that is cut off
is turned down without a search of the whole island (about 70 ms each).
make bot-matches plays 20 bot-only matches and checks winners, drownings and
difficulty. Seeds 101 to 120: 20 of 20 with a winner, 7 of 460 deaths were
drownings (1.5%), easy bots placed worst (13.7 against 11.8 and 12.0) and
hard bots took the most eliminations (186 against 137 and 132).
Closes#28
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
NavGrid is the map bots plan on: a point per 2 m terrain sample, solid on cliffs and
where solids stand between step and head height, and costlier where the water is
deeper than wading. Path search uses Godot's AStarGrid2D. The grid syncs every tick:
new solids (walls, pieces) block points at once, removed ones are found by a rolling
scan, and points turn wet as the tide and flood reach them. Each change is stamped
with its tick.
NavBot follows a path with ordinary PlayerCommands: it steers at waypoints, sprints
on long dry stretches, jumps when stalled, and marks a point blocked after repeated
stalls. Every half second it checks its remaining path against the change stamps and
plans again if water or a wall now crosses it. One path is planned per server tick
across all bots.
The server builds the grid while the lobby waits and its bots now roam between
random reachable places, replacing the placeholder wander bot.
Tests: 150 of 150 random reachable targets reached on 50 islands in the sweep; the
test runs 2 per island and needs 95%. New walls and rising water both cause a new
path around them; a broken wall opens the grid again; with no dry way round, the bot
swims. 23 roaming bots on an island: p99 tick 1.6 ms against the 12 ms budget.
Closes#27
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
TerrainView draws the island as 16 x 16 chunks of 64 m, each at one of
four detail levels (2, 4, 8 and 16 m spacing) chosen by distance to the
camera. TerrainChunkBuilder builds chunk geometry on WorkerThreadPool
threads; the main thread only uploads finished meshes, within a 2 ms
per-frame budget, nearest chunks first.
- Seams: every chunk hangs skirts deeper than twice the most any level
strays from the full-resolution edge, so neighbours at any two levels
meet without a crack. Normals come from the full-resolution heightfield
at every level, and back faces light like the ground above them.
- Splat material: a procedural shader blends sand, marsh, grass and rock
with weights from the island's bands, averaged over 4 m so borders
blend. Steep ground turns to rock and summits weather pale.
- Collision stays SimGround's heightfield, with no renderer involved.
- Capture scenarios terrain, terrain_lod (tinted by level) and
terrain_cracks (magenta sky). tools/crack_check.gd counts sky pixels
seen through the ground; make terrain-capture requires cracks in a
skirtless control and none in the real terrain.
Tests: chunk vertices sit on the heightfield; skirts cover every pairing
of levels on sampled shared edges; detail falls with distance; flying
across the island at 30 m/s for 2,037 frames streams 661 chunk rebuilds
with the worst main-thread frame at 0.8 to 1.4 ms (headless, CPU only);
24 players over 900 ticks on two generated islands never sink into the
ground, and a player sprints up towards a summit unharmed.
make terrain-capture: control 437 crack pixels, terrain 0.
Closes#10
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Simulation (src/sim/):
- SimGround holds the terrain heightfield (bilinear heights, normals)
and axis-aligned box solids, with support, body-overlap, point and
ray queries. The island generator and building will fill it.
- Movement steps a player per tick: walk, sprint (forward only),
crouch (lower body, slower, no jump), jump with exact constant-gravity
integration, air control, step-up onto ledges up to STEP_HEIGHT,
ground snapping on downhill walks, slopes over MAX_SLOPE_DEGREES that
block walking and jumping uphill and make players slide (speed capped),
ceiling bumps, sliding along walls, and fall damage from impact speed.
- Gravity now applies on ticks with no command. on_ground and crouching
are part of the state hash. New movement numbers are in Tuning.
- Fixed the strafe direction: right is forward x up in Godot's frame,
(-cos yaw, 0, sin yaw). Documented the look convention.
Client (src/client/):
- CameraRig places an over-the-shoulder camera with an eased
aim-down-sights blend (closer, narrower FOV), casting against the same
SimGround: pivot to shoulder, shoulder back, then a straight line from
the pivot, with a final clearance pull-in so the near plane never
enters geometry.
- InputSettings: mouse sensitivity, aim sensitivity scale and invert-Y,
clamped and saved to user://settings.cfg. PlayerCamera applies both.
Tests: 46 new across movement, ground queries, camera and settings,
including a 24,480-view scripted camera sweep through five arenas (wall,
corridor, low roof, corner, steep bank) that checks the eye and every
near-plane corner stay out of geometry with a clear line to the pivot.
The hash test now flips a high stick bit, because a 1/127 nudge to an
airborne player moves them less than the hash's millimetre.
Capture scenarios camera, camera_wall and camera_aim.
Closes#8
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add the fixed-tick Simulation, the EntityRegistry with never-reused IDs, SimRng's named
seeded streams, and the 14-byte PlayerCommand used by humans, bots and the network.
state_hash is a SHA-256 over a canonical encoding of the state. Movement is a flat
placeholder until the v0.2 character controller. Tests replay a 24-player command log
over 10,000 ticks and scan src/sim for unseeded randomness and presentation code.
Closes#5
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add docs/ARCHITECTURE.md covering the tick rate, simulation and presentation split,
command stream, entity IDs, snapshot format, seeded RNG rules, module map and
performance budgets. A test scans src/sim for presentation dependencies and unseeded
randomness, so the separation is enforced.
Closes#3
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>