Field ports batch 2 with the porting tools (15-30 routines)

Gatchan commented

2026-11-04 10:19:06 +00:00

Owner

First batch made with draftc.py / draft_check.py / liveness / groundtruth / portstyle. Next field routines in docs/coverage.md order that the tools can draft (skip frame-crossing ones, #37). Measure: routines per session and tokens per routine (agent usage / routines).

Gatchan added this to the Phase 2 - field ports milestone

2026-11-04 10:19:06 +00:00

Gatchan added the

task

port

labels

2026-11-04 10:19:06 +00:00

Gatchan referenced this issue from a commit

2026-11-04 13:54:33 +00:00

tools: portbridge.py generates a port's bridge and registry row from its drafts; port_carry, port_overflow, port_decimal_mode, port_set_db (refs #54)

Gatchan referenced this issue from a commit

2026-11-04 13:54:33 +00:00

state: RAM map for field batch 2 (sprite MSB tables, HDMA effect data, mosaic, shake, letter buffers, palette animation fields, timer event state) (refs #54)

Gatchan referenced this issue from a commit

2026-11-04 13:54:33 +00:00

port: field batch 2, 15 routines (SetTopSpriteMSB, SetBtmSpriteMSB, FixPlayerSpritePriority, UpdateObjLayerPriorityAfter, UpdateMosaic, UpdateFixedColor, UpdatePalAnimPulse, UpdateShake, Rand, ShiftPrevLetter, DrawLetter, CopyDlgTextToBuf, ClearTextGfxBuf, CloseMapTitleWindow, CheckTimer); bridges and rows generated by portbridge.py (shadow 5 routes 0/745371 US, 0/668195 JP, 0 fallback; fuzz 0 mismatch; 683/683 original instructions US/JP) (refs #54)

Gatchan referenced this issue from a pull request that will close it,

2026-11-04 13:54:35 +00:00

Field batch 2: 15 routines with the porting tools (fixes #54) #55

Gatchan closed this issue

2026-11-04 13:54:41 +00:00

Gatchan referenced this issue from a commit

2026-11-04 13:54:44 +00:00

Merge pull request 'Field batch 2: 15 routines with the porting tools (fixes #54)' (#55) from field-batch2 into main

Gatchan commented

2026-11-04 13:54:52 +00:00

Author

Owner

Merged in PR #55. Field batch 2: 15 field routines ported with the tools (draftc -> draft_check -> port -> generated bridge -> groundtruth + shadow).

Ported: SetTopSpriteMSB, SetBtmSpriteMSB, FixPlayerSpritePriority, UpdateObjLayerPriorityAfter, UpdateMosaic, UpdateFixedColor, UpdatePalAnimPulse, UpdateShake, Rand (field), ShiftPrevLetter, DrawLetter, CopyDlgTextToBuf, ClearTextGfxBuf, CloseMapTitleWindow, CheckTimer.

New tool: tools/portbridge.py generates each bridge and registry row from the routine's drafts (entry guards, live outputs from liveness, don't-care set, hardware writes); the porter states how each live output the routine may change is left (--set/--same), the tool refuses otherwise. Documented in docs/drafts.md.

Validation

Drafts proven first (draft_check, 28 drafts, title_to_mines_save + field_menu_tour US/JP): 436 407 calls, 2 mismatches, both SetWindowSelect (PPU BG1 hscroll changed by HDMA during the original, not written by it) -> SetWindowSelect skipped.
Shadow, 5 routes, all ports: US 745 371 calls of the 15 ports, JP 668 195, 0 mismatch, 0 fallback (UpdateMosaic guard fixed after 7 fallbacks/version on narshe_to_world_map, rerun 0/0); every movie matches its end hashes.
Fuzz (field_menu_tour state, 120 iterations/version + callees alone 60): 0 mismatch, 0 fallback.
Original instructions executed (routes + fuzz coverage): 683/683 US, 683/683 JP.
groundtruth: 16/30 same; the differences are tool limits (buffer accessors, scratch carried by the bridge, overlapping HDMA element names) and UpdateShake reading the offsets to keep their low byte.
portstyle 0 findings; debug and clang builds 0 warnings, ctest -LE slow 27/27 each.

Skipped: SetWindowSelect (PPU compare noise, needs a window video service), TfrBGAnimGfx/TfrBG3AnimGfx/TfrObjGfxSub (VRAM DMA with partial register setups: need a vid VRAM upload service; TfrObjGfxSub keeps everything live), InitBG1/2VScroll, _c0249a, UpdateBG1, UpdateBG3HScroll, UpdateScrollHDMA, InitMaskHDMAData, _c0714a, MoveObjs, DrawOverlaySprites, LoadLetterGfx, UpdateDlgWindow (larger, scratch-heavy: next batch). Frame-crossing routines not attempted (#37).

Merged in PR #55. Field batch 2: 15 field routines ported with the tools (draftc -> draft_check -> port -> generated bridge -> groundtruth + shadow). Ported: SetTopSpriteMSB, SetBtmSpriteMSB, FixPlayerSpritePriority, UpdateObjLayerPriorityAfter, UpdateMosaic, UpdateFixedColor, UpdatePalAnimPulse, UpdateShake, Rand (field), ShiftPrevLetter, DrawLetter, CopyDlgTextToBuf, ClearTextGfxBuf, CloseMapTitleWindow, CheckTimer. New tool: `tools/portbridge.py` generates each bridge and registry row from the routine's drafts (entry guards, live outputs from liveness, don't-care set, hardware writes); the porter states how each live output the routine may change is left (`--set`/`--same`), the tool refuses otherwise. Documented in docs/drafts.md. Validation - Drafts proven first (draft_check, 28 drafts, title_to_mines_save + field_menu_tour US/JP): 436 407 calls, 2 mismatches, both SetWindowSelect (PPU BG1 hscroll changed by HDMA during the original, not written by it) -> SetWindowSelect skipped. - Shadow, 5 routes, all ports: US 745 371 calls of the 15 ports, JP 668 195, 0 mismatch, 0 fallback (UpdateMosaic guard fixed after 7 fallbacks/version on narshe_to_world_map, rerun 0/0); every movie matches its end hashes. - Fuzz (field_menu_tour state, 120 iterations/version + callees alone 60): 0 mismatch, 0 fallback. - Original instructions executed (routes + fuzz coverage): 683/683 US, 683/683 JP. - groundtruth: 16/30 same; the differences are tool limits (buffer accessors, scratch carried by the bridge, overlapping HDMA element names) and UpdateShake reading the offsets to keep their low byte. - portstyle 0 findings; debug and clang builds 0 warnings, `ctest -LE slow` 27/27 each. Skipped: SetWindowSelect (PPU compare noise, needs a window video service), TfrBGAnimGfx/TfrBG3AnimGfx/TfrObjGfxSub (VRAM DMA with partial register setups: need a vid VRAM upload service; TfrObjGfxSub keeps everything live), InitBG1/2VScroll, _c0249a, UpdateBG1, UpdateBG3HScroll, UpdateScrollHDMA, InitMaskHDMAData, _c0714a, MoveObjs, DrawOverlaySprites, LoadLetterGfx, UpdateDlgWindow (larger, scratch-heavy: next batch). Frame-crossing routines not attempted (#37).

Gatchan referenced this issue

2026-11-04 13:55:29 +00:00

Field ports batch 3: scroll/HDMA/dialogue routines deferred from batch 2, VRAM upload operation #56

Rows
Columns

Field ports batch 2 with the porting tools (15-30 routines) #54