native: divergence at tic 143, a damaged thing keeps its old frame

[MarcusKainth]

ROM sha256

9a6a47d01119f67580e48e9875207186c25efd56ff93019df331eb307cfaa5d9

WAD sha256

1d7d43be501e67d927e415e0b8f3e29c3bf33075e859721816f652a526cac771

ClickHouse version

26.7.5.10

First divergent tic (gametic)

143

First divergent frame, if a frame differs

No response

What disagrees

A mobj field

The divergence report

clickdoom: error: tic 143 mobj slot 118 m_x: 8272000 against the probe's 8245976

native diff stops on the first field, so the report names m_x. Comparing
every mobj column at that tic gives the whole set:

col       slots       ours                 probe
m_frame   [118,258]   [2,2]                [7,7]
m_momx    [118,258]   [-26024,47238]       [-23585,42809]
m_momy    [118,258]   [-77677,-113436]     [-70395,-102802]
m_momz    [265..271]  [131072 x7]          [65536 x7]
m_x       [118,258]   [8272000,13992912]   [8245976,14040150]
m_y       [118,258]   [-4310912,4297488]   [-4388589,4184052]
m_z       [265..271]  [2384896,...]        [2515968,...]

m_health, m_flags, m_state and m_tics agree on every slot, so the
damage itself lands. Three separate things are missing.

m_frame on slots 118 and 258 is this report's subject. P_DamageMobj
calls P_SetMobjState, which writes the entered state's sprite, frame and tics
along with the state. The damage writes m_state (449 to 455) and m_tics,
and slot 118's state matches the probe, but m_frame stays at 2 where the
probe has 7. states gives state 455 a frame of 7, so the frame the thing
shows is the frame of the state it left.

m_sprite agrees only because states 449 and 455 share sprite 0.

The other two are separately tracked work: m_x, m_y, m_momx and m_momy
on the same two slots are P_XYMovement for a thing that is not the player,
and m_z and m_momz on slots 265 to 271 are P_ZMovement for the blood the
shots spawned.

Random draws at that tic

No response

Reproduction

# On the branch that fires the shotgun's seven shots.
make native-load
clickdoom native diff 150 \
  --probe refemu/reference_traces/demo3/probe.9a6a47d01119.tsv \
  --host localhost --port 8123 --password clickdoom --database clickdoom_native

# The whole column set at the tic, rather than the first field:
#   SELECT arrayFilter(k -> n.m_frame[k] != p.m_frame[k], arrayEnumerate(n.m_frame))
#   FROM clickdoom_native.native_state n, clickdoom_native.probe_state p
#   WHERE n.tic = 143 AND p.gametic = 143

[MarcusKainth]

Tic 143 agrees on every field this report names, on main at 5404848 with
ClickHouse 26.8.2.7. native diff reports no divergent field at all and stops
at the first refused tic instead:

$ clickdoom native diff 181 --probe refemu/reference_traces/demo3/probe.9a6a47d01119.tsv
refemu/reference_traces/demo3/probe.9a6a47d01119.tsv: 2172 rows over 2133 tics into clickdoom_diff.probe_state
# native diff elapsed=70.8s tics=169 tics/s=2.4
clickdoom: error: tic 181 unresolved: CHASE_STUCK

Slot 118 across the hit, both sides:

side   tic  state  sprite  frame  health
ours   142  449    0       2      60
ours   143  455    0       7      50
ours   144  456    0       7      50
probe  142  449    0       2      60
probe  143  455    0       7      50
probe  144  456    0       7      50

The frame follows the state now. fire_shots' own writeback reads the picture
off the state the shots left, in native/src/sql/sim/player.rs:

// `P_SetMobjState` writes the entered state's picture along with the
// state, so a thing the damage put into its pain or its death frames
// shows that state's picture.

The momentum and position fields the report listed as separately tracked agree
at 143 too, since the run reaches 180 with nothing to report.

The same mechanism is still open on the damage paths that writeback does not
reach: a claw, a missile's own impact and a thrown missile's first tic all set
m_state in the thinker stage and leave m_sprite and m_frame behind.
That is #528.

Recommending this one be closed. Leaving that to a maintainer rather than
doing it myself.

[MarcusKainth]

Resolved on main: clickdoom native diff 181 at 5404848 on 26.8.2.7 reports no divergent field through tic 181, and slot 118 at tic 143 matches the probe on state, sprite, frame and health (the gunshot writeback reads the picture off the state the shots leave). The thinker-stage damage paths that still leave the picture behind are tracked in #528.