Entry 205

"The stream effects on every disc: Setup's switch by register, the effects' literal tables carried, the carry's sequence pass"

2026-09-26volumes

Filestools/main_data.pycrates/piney-data/src/main_datacrates/piney-effect/src/particle/tables.rscrates/piney-effect/src/drawelm.rscrates/piney-effect/src/summoned.rscrates/piney-effect/src/upheaval.rscrates/piney-effect/tests/volumes.rs

After 0204 the three later discs reached their desktops, but every stream's effects failed to build ("the stream's effects: ..."). A new test, piney-effect/tests/volumes.rs, builds the tables on each disc part by part. The failures were in three places.

ccParticle::Setup's switch

The port reads the switch on a texture id out of the function's code: its bias, its bound, its jump table and each case's row and CLUT. The old reader matched Infection's exact words.

Outbreak and Quarantine recompiled Setup:

• the switch sits one instruction later;
• the id is in s2 rather than s0;
• the row is in s1 rather than s2;
• the CLUT is in s0 rather than s3.

The reader now finds the addi a0, id, -first / sltiu at, a0, n pair in the function's first 0x200 bytes. It reads each case's two assignments (li, move r, zero, move r, id) by register: the row's register and the CLUT's are the ones the first case assigns. All four volumes decode to the same map: ids 91-244, where 91-100 are the default and 101 draws row 0 with CLUT 101 (the_setup_switch_on_every_volume).

Literal tables behind helpers

The upheaval, summoned-creature and element-draw tables read literal addresses through local helpers (names, ints, arr), which did not carry. The 0202 sweep had wrapped only literal reads made directly. The helpers now carry. read_f32s and read_i16s already did.

The carry's sequence pass

Mutation's ConvergenceSystem grew by 224 bytes, so no alignment reached its @3328. The name pass had put @3329 where @3328 is: each compile numbers its labels anew.

by_sequence takes each pair of same-named functions and lists the data addresses each builds, in instruction order, leaving out bare luis and high halves alone. It uses a pair only when:

• the two lists are equally long;
• every named global the carry already holds sits at its carried place, with at least one there.

A global still uncarried then goes to the address in the same place, on unanimous votes. A compiler label (@123) the name pass carried doesn't count as an anchor, and the order may move it ("the code wins"). A global with no pointers in it moves only where the volume's bytes are Infection's. That keeps out places where the tracker paired a lui with the wrong low half (@1871 built at 0x0034fff0, zeros).

On Mutation it moved 11 labels. Among them are @1308, @7305 and their runs, where the names had landed on the neighbouring strings (CMP_sr1dat1_1 for _2, ANM_sd5ae0_a for sd1). It carried 286 globals besides: vtables, .sdata statics and @3328-@3330.

The misses the source still names: Mutation 6, Outbreak 29, Quarantine 28.

All three later discs now go through New Game into the desktop with their streams, effects included, and no error printed.