fix(multiplayer): adversarial review fixes for Phase 2

An Opus subagent's adversarial review of Phase 2 found real bugs the
smoke tests couldn't catch, since constant-velocity dead reckoning still
moves a ship far enough to pass a "moved > 1.0" check:

- The interpolator never actually interpolated. NetInterpolator.to_tick()
  assumes physics_frame * TICK_MS == Time.get_ticks_msec() on the server,
  which is off by a steady ~45-55ms in practice (real startup work before
  the first physics step, widened by any dropped tick). Every sample_at()
  call took the extrapolation branch, 100% of the time, defeating the
  interpolation buffer entirely. Fixed with a shared, min-filtered rolling
  bias estimate in networked_match.gd, applied before every to_tick() call.

- Goals caused a ~27m visual slide: _reset_gen was bumped before the
  queued teleport actually landed, so the client's buffer-clear kept
  exactly the stale in-goal sample and lerped a slide to the next, real
  one. Fixed by tracking the tick the goal was detected on and only
  bumping the generation once strictly later ticks confirm the teleport
  has landed - a naive "next _physics_process" boolean flag doesn't
  work, since a goal Area's body_entered fires before that same tick's
  _physics_process runs, not on the next one.

- _local_input_sampler (a Node, never added to the tree) was never freed
  - this was the unexplained "3 resources still in use at exit" warning
  on every Phase 2 test run.

- Ball angular velocity decoded 8x too small (rescale_avel was never
  called); get_server_time_estimate_ms() was used before the clock had
  synced; net_sim.gd's delayed-send timer stopped ticking while the tree
  was paused and didn't check connection status before firing;
  _broadcast_snapshot's ball index could silently break if a ship were
  ever despawned; declared-but-unemitted HUD lifecycle signals showed a
  permanently frozen timer widget.

Also confirmed, empirically, several things the review checked and found
fine: a hostile client sending malformed input cannot crash the server,
skipping GameMode's super() drops nothing load-bearing, deterministic
slot assignment is correct with 2 real simultaneous clients, and RPC
authority enforcement genuinely rejects a forging client.

All fixes verified with real two-process runs (including forcing an
actual goal and reading the server's own broadcast stream) and temporary
instrumentation, removed once each fix was confirmed. Full Phase 1 +
Phase 2 regression suite, including the net-sim-latency milestone gate,
re-run clean after every fix.
This commit is contained in:
Josh Creek
2026-08-20 12:43:33 +01:00
parent 7b150ef72e
commit 14698d4ccb
3 changed files with 169 additions and 12 deletions
+144 -10
View File
@@ -15,11 +15,14 @@ extends GameMode
# rather than relying on GameMode's default (arena-required-synchronously)
# flow.
signal timer_updated(minutes: int, seconds: int)
# Only score_changed is actually emitted in Phase 2 — Phase 5 owns the match
# lifecycle state machine (timer, kickoff countdown, overtime, results), so
# those signals get declared there, alongside real emission. Declaring one
# here without emitting it isn't harmless: HUDController gates the timer
# widget's visibility purely on has_signal("timer_updated"), so a declared-
# but-dead signal shows a permanently frozen timer rather than correctly
# hiding it the way free_play.gd's total absence of the signal does.
signal score_changed(score: Dictionary)
signal match_ended(winning_team: int, score: Dictionary)
signal kickoff_countdown(count: int)
signal overtime_started
const NetCodec = preload("res://scripts/net_codec.gd")
const NetBodyState = preload("res://scripts/net_body_state.gd")
@@ -35,6 +38,33 @@ const INTERP_DELAY_MIN_MS := 25.0
const INTERP_DELAY_MAX_MS := 200.0
const SNAPSHOT_INTERVAL_MS := 1000.0 / 60.0
# NetInterpolator.to_tick() assumes Time.get_ticks_msec() == physics_frame *
# TICK_MS on the SERVER, i.e. that physics frame 0 happened at process-start
# wall time. It doesn't: real startup work (autoloads, asset loading) elapses
# before the first physics step, and any dropped tick widens the gap further
# — it only ever grows. An adversarial review found this was NOT a rounding
# error: it measured a steady +45-50ms bias on a real run, meaning EVERY
# to_tick(get_server_time_estimate_ms()) call landed 3+ ticks past the
# newest buffered sample, so sample_at() took the extrapolation branch 100%
# of the time — zero real interpolation ever happened, on LAN or under
# simulated latency alike, silently defeating the entire interpolation
# buffer this phase was built around.
#
# Fix: this bias is a property of the server's clock, not of any one body,
# so track ONE shared estimate here (not per-interpolator) from every
# snapshot's own server_tick versus this client's server-time estimate at
# receipt. Take the MINIMUM over a rolling window — same rationale as
# NetworkManager's own min-RTT filtering (network_manager.gd): the sample
# with the least one-way transit delay best isolates the constant epoch
# bias from per-packet network noise, and a rolling (not all-time) window
# lets a real increase in the bias — the server dropping more ticks later
# in the match — still get picked up rather than staying pinned to a
# now-stale historical minimum.
const TICK_BIAS_WINDOW_SEC := 5.0
var _tick_bias_samples: Array[Dictionary] = [] # [{t_ms:int, bias_ms:float}], client only
var _tick_bias_ms := 0.0 # best current estimate; 0.0 until the first snapshot
class SlotInfo:
var peer_id: int
@@ -51,6 +81,31 @@ var _ball_interpolator := NetInterpolator.new() # client only
var _local_input_sampler := PlayerShipController.new() # client only: reads local input each tick to forward; never added to a Ship, never in the tree — get_action() only touches the global Input singleton
var _input_seq := 0 # client only
var _reset_gen := 0 # server only: bumped on every kickoff/goal reset so the client hard-snaps instead of interpolating across the teleport
# Server only. _on_goal_scored's reset_ball()/reset_ships() only QUEUE
# teleports (task 0.15's queue_teleport — applied on each body's next
# _integrate_forces), but _broadcast_snapshot runs later in the SAME frame
# _on_goal_scored fires in, before that teleport lands. Bumping _reset_gen
# immediately would tag the still-pre-teleport snapshot with the new
# generation: the client clears its buffer expecting a hard snap, then
# keeps exactly that stale in-goal sample and lerps a full-arena slide to
# the next, genuinely-post-teleport sample — an adversarial review measured
# a 26.8m ball slide from this.
#
# A plain "bump on the next _physics_process" boolean flag turned out NOT
# to fix it: the goal Area's body_entered signal (and so _on_goal_scored)
# fires as part of physics tick N's OWN step processing, before tick N's
# _physics_process callback — so a flag set there is already true by the
# time that SAME tick's _physics_process checks it, consuming on tick N
# instead of N+1 as intended (empirically confirmed: with a boolean flag,
# gen still bumped on the same tick the stale position was broadcast).
# The queued teleport, by contrast, isn't applied until tick N+1's
# _integrate_forces. So the two must be compared by TICK NUMBER, not by
# "next callback": only bump once the current tick is strictly later than
# the tick the goal was detected on, which guarantees at least one full
# _integrate_forces has run — and therefore the queued teleport has
# landed — since the flag was set.
var _pending_reset_gen_bump := false
var _pending_reset_gen_bump_tick := -1
func _ready() -> void:
@@ -86,6 +141,20 @@ func _owns_world_simulation() -> bool:
return multiplayer.is_server()
# _local_input_sampler is a plain Node (PlayerShipController extends
# ShipController extends Node) that's deliberately never added to the tree
# — dropping the last reference to it does not free it. An adversarial
# review traced the "3 resources still in use at exit" warning on every
# Phase 2 test run directly to this: --verbose named the leaked script
# chain (player_ship_controller.gd, ship_controller.gd, ship_action.gd)
# exactly, and adding this cleanup made the warning disappear. Runs
# unconditionally (not just client-side) since the field is initialized
# unconditionally too, despite its "client only" comment.
func _exit_tree() -> void:
if is_instance_valid(_local_input_sampler):
_local_input_sampler.free()
# ============================================================
# Server
# ============================================================
@@ -141,17 +210,24 @@ func _on_goal_registered(conceding_team: int) -> void:
func _on_goal_scored(_conceding_team: int) -> void:
_reset_gen = (_reset_gen + 1) % 256
reset_ball()
reset_ships()
_pending_reset_gen_bump = true
_pending_reset_gen_bump_tick = Engine.get_physics_frames()
func _broadcast_snapshot() -> void:
var server_tick := Engine.get_physics_frames()
var bodies: Array[NetBodyState] = []
# Always one entry per slot, even for a momentarily-invalid ship
# (placeholder zero state), so the ball always lands at the fixed index
# _slots.size() the client assumes in _on_snapshot_received — skipping
# invalid ships entirely would shift every later index. "No ship is ever
# despawned" (§6.4) means this is unreachable today, but it's a silent
# total-garbage failure mode the moment that stops being true, and the
# fix costs nothing.
for slot in _slots:
if is_instance_valid(slot.ship):
bodies.append(_ship_to_net_body_state(slot.ship))
bodies.append(_ship_to_net_body_state(slot.ship) if is_instance_valid(slot.ship) else NetBodyState.new())
if is_instance_valid(ball):
bodies.append(_ball_to_net_body_state(ball))
var segment := NetCodec.pack_snapshot_body_segment(server_tick, 0, _reset_gen, bodies)
@@ -273,11 +349,55 @@ func _on_snapshot_received(decoded: Dictionary) -> void:
var server_tick: int = decoded["server_tick"]
var reset_gen: int = decoded["reset_gen"]
var bodies: Array = decoded["bodies"]
_update_tick_bias(server_tick)
for i in _slots.size():
if i < bodies.size():
_slots[i].interpolator.add_sample(server_tick, bodies[i], reset_gen)
if bodies.size() > _slots.size():
_ball_interpolator.add_sample(server_tick, bodies[_slots.size()], reset_gen)
var ball_state: NetBodyState = bodies[_slots.size()]
# unpack_snapshot() decodes every body's angular_velocity assuming
# SHIP_AVEL_RANGE; the ball was quantised at BALL_AVEL_RANGE
# (_ball_to_net_body_state), so it decodes 8x too small without this
# — dormant today (nothing reads decoded angular_velocity yet) but
# silently wrong the moment ball-spin VFX or Phase 4 prediction does.
NetCodec.rescale_avel(ball_state, NetCodec.BALL_AVEL_RANGE)
_ball_interpolator.add_sample(server_tick, ball_state, reset_gen)
# See the class-level comment above _tick_bias_samples for why this exists.
# bias_ms is how much further ahead to_tick(server_time_est) lands than the
# server_tick this snapshot actually carries — mostly the server's own
# physics-frame/wall-clock startup skew, plus a little real one-way transit
# noise that the rolling minimum below filters back out.
func _update_tick_bias(server_tick: int) -> void:
# get_server_time_estimate_ms() is meaningless before the first pong
# lands (clock_offset_ms == 0.0 until then, per network_manager.gd's own
# doc comment) — recording a bias sample from it during that window
# produced a garbage value (~-1.1s, the client's own raw pre-sync
# uptime standing in for a server-synced estimate) that the rolling-min
# window then locked onto for the rest of a short test, since 5 real
# seconds never fully elapsed before the test ended. Skip entirely
# until the clock is actually synced.
if NetworkManager.rtt_ms < 0.0:
return
var server_time_est := NetworkManager.get_server_time_estimate_ms()
var bias_ms := server_time_est - float(server_tick) * NetInterpolator.TICK_MS
var now_ms := Time.get_ticks_msec()
_tick_bias_samples.append({"t_ms": now_ms, "bias_ms": bias_ms})
var cutoff := now_ms - int(TICK_BIAS_WINDOW_SEC * 1000.0)
_tick_bias_samples = _tick_bias_samples.filter(func(s: Dictionary) -> bool: return s["t_ms"] >= cutoff)
var best: float = _tick_bias_samples[0]["bias_ms"]
for sample: Dictionary in _tick_bias_samples:
var sample_bias: float = sample["bias_ms"]
if sample_bias < best:
best = sample_bias
_tick_bias_ms = best
# Bias-corrected replacement for NetInterpolator.to_tick(server_time_est) —
# use this instead of calling to_tick() directly on a server-time estimate.
func _estimated_tick(server_time_ms: float) -> float:
return NetInterpolator.to_tick(server_time_ms - _tick_bias_ms)
func _current_interp_delay_ms() -> float:
@@ -298,12 +418,24 @@ func _physics_process(_delta: float) -> void:
if _owns_world_simulation():
_respawn_escaped_bodies()
if multiplayer.is_server():
if _pending_reset_gen_bump and Engine.get_physics_frames() > _pending_reset_gen_bump_tick:
_reset_gen = (_reset_gen + 1) % 256
_pending_reset_gen_bump = false
_broadcast_snapshot()
return
_send_local_input()
# get_server_time_estimate_ms() is meaningless before the first pong
# lands (network_manager.gd's own doc comment says so explicitly) — an
# adversarial review found this was used unguarded here, which against
# a long-running dedicated server (clock_offset_ms == 0.0, so this
# process's own short uptime is compared against the server's enormous
# tick count) freezes every remote body at the oldest buffered pose for
# the whole first second of every match.
if NetworkManager.rtt_ms < 0.0:
return
var server_time_est := NetworkManager.get_server_time_estimate_ms()
var collider_tick := NetInterpolator.to_tick(server_time_est)
var collider_tick := _estimated_tick(server_time_est)
for slot in _slots:
if is_instance_valid(slot.ship) and slot.interpolator.has_samples():
_apply_collider_state(slot.ship, slot.interpolator.sample_at(collider_tick))
@@ -330,8 +462,10 @@ func _process(_delta: float) -> void:
NetworkManager.poll()
if multiplayer.is_server() or _slots.is_empty():
return
if NetworkManager.rtt_ms < 0.0:
return
var server_time_est := NetworkManager.get_server_time_estimate_ms()
var visual_tick := NetInterpolator.to_tick(server_time_est - _current_interp_delay_ms())
var visual_tick := _estimated_tick(server_time_est - _current_interp_delay_ms())
for slot in _slots:
if is_instance_valid(slot.ship) and slot.interpolator.has_samples():
_apply_ship_visual_state(slot.ship, slot.interpolator.sample_at(visual_tick))