SMBC reads level objects — ProcessAreaData and the decoder

ProcessAreaData is translated, with DecodeAreaData and every area object handler — pipes, blocks, holes, staircases, the flagpole, castles, the loop command, the lot. The level data stream now actually builds columns: the parser core's per-column call into the decoder is live, and with it the last stub in the area parser.

The details worth keeping

The decoder is a three-slot pipeline walking slots 2 down to 0, so a single pass can buffer several objects as the renderer catches up. Scratch registers $00, $05, $06 and $07 became locals and nullable out-params — the asm's register juggling doesn't survive structured C, but the flow does, labels and all.

Two faithful quirks worth naming. Castle and jumpspring objects don't check FindEmptyEnemySlot's carry, so a full enemy buffer makes them write to slot 5 — one past the end of the asm's own enemy arrays, and the port keeps it. And the castle's start row comes from the second byte's low nybble via GetLrgObjAttrib, not the row field.

The sanitizers earned their keep: a NULL-deref in ChkLrgObjLength (the length out-param can be NULL now that scratch registers are parameters) crashed at -O0 while -O2 happened to survive. Two test corrections too — IncAreaObjOffset resets the page select behind a page-control object, and the loop command's real encoding is $4b, since d7 of every object's second byte is the page-select toggle. The port was right; my test data wasn't.

What's next

The enemy data interpreter (ProcessEnemyData and friends) is the other half of level loading, then the backloading/area-initialisation glue that ties the parser into InitializeArea.