Entry 117

"Skeith's pictures: the effects, the cross's trail, the reversed layer and the arena's fireflies"

2026-09-25battle render world

Documented in Bosses - ccBoss and Skeith, The story maps - EVENTAREA01 (area 13), EVENTAREA02 (area 15), EVENTAREA03 (area 43), EVENTAREA07 (area 16), EVENTAREAB0 (the boss arenas) and EVENTAREAB8 (Kyvia's disc).

Skeith fought in its arena (113, 115), but the picture was thin. The rules made effects that nothing drew. The cross left no trail, the magic's inverted screen never came, and the arena's 54 fireflies were stand-ins. The boss also stood frozen in one pose. All of it is drawn now (bosses, the story maps).

The effects

piney_effect::boss is ccBossEffManager and the six effects Skeith makes: WaveShock, MagicSquare (with its ccBossEffLight), ForceGenerator (ccBossEffBrightMagicSquare's photons), AutoSamonRing, IceBreak and Dead. Each has its Create, its constructor and its Draw. Their generators go to the particle system.

A photon's generator follows the photon (syncPos). The port publishes the photon's place as an anchor (VecRef::Anchor) that the generator reads each step.
The light is a real omni light. Its blue rises and falls over 80 frames, turned with the camera's heading (Host::camera_rot). The field lights the characters with it beside the arena's lights.
The ice rocks are effIceRock in the effect control, ids 15-18.
ccIBossEff::cmn_bossCam: none of the six reads it, so BossRun.cam is not needed. Beside their parameter tables, the globals their Draws read are camID, cc3d and scFadeDef.
The order. The manager's task runs before the boss's. The field calls the manager's pass before ccBoss01::Main, and makes the boss's Out::Effects right after it; in the game the Creates run inside Main. Of the Creates, IceBreak's constructor draws one ccRand(), which the port draws when it makes the effect.
Sounds and flashes reach the area mode. A note with no position (ccSeOnNote) became SeNote.

tools/test_boss_effect_rs.py (new) runs the game's Creates and the manager's Draw loop in eemu on the effect machine, against boss_probe:

each of the six kinds, two cases each;
OnMagicAtk's own order of Creates, a WaveShock and the dead effect alongside, 420 frames.

It compares every draw, sound, flash, generator start and kill, the light, m_bEnabled, the ice rocks' slots, rand and the ccRand count, frame by frame. It found 0 mismatches.

Making the harness honest took three fixes:

the force fields a generator is made with are heap ccParticleForceField objects, so the harness reads each one's parameter row through it;
ccCoord's stubbed constructor left every ccAnm's transparency 0, and it now sets the defaults;
the probe's empty answers had shifted the case numbering.

The frozen boss

The actor drew CMP_trall at play.posed, which nothing set, so every frame showed the clip's first pose. piney_battle::boss::Anm now keeps posed, the frame _AnimateForward posed (as forward_clip's pose_at). The actor, the after-images and the wave draw at it.

The cross's trail

ccBoss01::DrawCross runs at the end of Main. In act 3 it lays a new row of ccLattice(2, 7) from the sword's two dummies (DMY_xdummy_w01, _w02) under OBJ_ex11swd's last world matrix, and every frame it calls Disp. piney_world::lattice is ccLattice:

16 rows as a ring;
life 6;
the alpha fades under life 5;
nothing is sent until 16 rows have been made;
ADC for the first pair and for vertices behind the eye or off the guard band.

The field sends it on effLayer, sorted by mean z, as triangles, depth GREATER with no write. In skeith_act_shots (act 3) it shows as a magenta ribbon behind the swing.

test_skeiths_trail (in test_boss_effect_rs.py) drives the game's own ccLattice(2, 7) in eemu the way DrawCross does: 605 frames of cross runs and idle runs, 443 of them sending a strip. It compares the lattice after every call and, after every Disp, the strip SendPacket sent: each vertex's RGBA and ADC as MakePacket wrote it, and the sort key. They match. Getting there turned up three things:

The stand-in projection was off-centre. MakePacket's clip measures from (0x7000, 0x7200), not (0x9000, 0x8e00) as a first reading of its addiu immediates suggested. The port already had 0x7000 and 0x7200 (XYOFFSET); only the harness's stand-in was wrong.
The key was an ulp out. The port had summed and divided in IEEE singles. It now sums each vertex's z with the EE's add, as MakePacket does, and divides by 2.0 * rows with the EE's divide, as SendPacket does. The key matches to the bit.
SendPacket's "two or more" counts rows, not vertices. The port tested out.len() < 2 on vertices. That was the same in effect, since a head apart from the tail always gives two rows, and it is now literal.

The reversed layer

ccBufferReverce::MakePacket is one white sprite over the view, blended (Cs - Cd) * 128/128. That is 255 less the frame: the picture inverted. Its alpha test fails into the frame buffer only, with no depth. It goes on the boss's own layer, priority 1, while the rules report Out::Reverse. In the magic's last part the arena turns negative (green sky to pink) under the ice crystal, and the HUD stays as it was.

The fireflies

FIREFLY2 is ported for type 0, the arena's kind (piney_world::firefly). The arena constructor makes the 54, each over DMY_marker(k % 18 + 1). They draw from the arena's fieldrand (six draws each, 324 in all), so Arena::new_seeded takes the seed before they are made. EVENTAREAB0::Draw calls each one's Draw, then its Move, between model 5 and the bobbing models. What the reading showed:

Out of reach, a firefly is frozen. Move returns at once while its position relative to the player is more than 1000 across. Draw skips it too.
Kite at the centre sees few or none. The markers are 1200 or more out. skeith_act_shots gained PINEY_WALK to walk Kite off. At (1742, 3302) up to 15 sprites are drawn, fireflies and sparks, as small pale dots.
The rebase is relative. At life 0 a firefly calls SetBasePosition with its position relative to the player, so it starts again near the world's origin. For the arena that is the centre.
They rise. The constructor's velocity is (0, 0, 1). The turns set only x and y, so a firefly climbs a unit a frame for as long as it lives.
Transparency. Draw sets a fading firefly's transparency to 0.1 times its life, up to 9.9. Move then sets it to life / 100, which the next Draw overwrites. The sprite's alpha saturates.
Sparks. A firefly makes a spark with probability 39/100 a frame while it has under three: a drifting one (life 60-89, scale 0.75) or a quick one (life 5-9, speed 2, scale 1.2).
Render state. A spark or firefly keeps ccEff::Init(chunk, 1)'s state: fog and the depth test. StorySprite now carries the pattern, scale and transparency, and whether it is a flare, whose constructor turns both off.

ArenaAgainstGame drops the stand-ins and runs the game's own FIREFLY2 code in eemu. It makes 900 draws, the player walking among the markers and the origin, and compares every firefly and spark ccEff::Draw (position, pattern, transparency, scale, layer) and fieldrand's seed after each frame. They match.

Still unknown

ccBossBlur, the cinema and hold 7 are not ported. - Type-1 fireflies, the fields' (SetBasePosition2, the camera-based rebase in Move), are not ported. - Only the eye has checked the picture. The trail's projection and its off-screen ADC, the reversed sprite, and the light's effect on the characters' shading have not been set beside the game's frames. - The rules and the effects are not checked together. In the game the boss's choices and the effects' generators share rand. The field port shares it too, but no test runs the rules and the effects together against the game. Nor was it checked whether anything later in Main draws ccRand after IceBreak's Create, which the port's later Create would reorder.