erik/acdream
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 2

fix(chargen): Campaign CC CC5 review fix round — F1-F14

Browse source 
Opus dual-lens review of 34e3a534+a975efd1 returned architectural
PASS-with-items / retail-fidelity FAIL. Every finding fixed:

- F1 (BLOCKER): deleted CharacterCreationSummaryPage's dead
  _suppressNextFieldEvent latch. UiField.SetText never raises
  OnFocusLost/OnSubmit, so the latch never had anything genuine to
  suppress — it stayed armed until the player's own next real commit
  and silently ate their typed name.
- F2: byte-re-derived gmCharGenMainUI::RecvNotice_
  CharGenVerificationResponse @0x004e9030's jump table — Pending is an
  explicit switch case landing on the SAME NameDBDown label as
  Corrupt/DatabaseDown, and Undef/out-of-range falls through the
  function's own unsigned-underflow default arm to that identical
  label. Retail's dispatch has NO silent branch. ApplyCreationResponse
  now produces a real rejection for Pending/Undef instead of a silent
  reset; ReconcileDialogs maps them to NameDBDown. Corrects the wrong
  "retail swallows Pending" claim everywhere it was repeated (plan doc,
  Core.Net doc comment, Runtime doc comments).
- F3: skill rows now use the key/value template with
  CharGenState::GetSkillScore @0x005C4B50 as the value (ported via the
  new RetailSkillFormula.CalculateChargenScore /
  ChargenSkillScoreResolver, wired through a new GetSkillScore
  binding), not template 0/name-only; bucket headers are unconditional.
  Writing this fix's own regression test surfaced a second, more severe
  bug: CharacterCreationSummaryPage never wired _list.TemplateResolver
  at all, so RebuildListbox has been a silent no-op since CC5 shipped —
  fixed by threading templateResolver through the page's constructor,
  matching every sibling UiTemplateListBox owner.
- F4: added the missing _errorMessageDialogContext one-outstanding
  guard to the 0xF643 rejection dialog, matching
  MakeErrorMessageDialog's own guard @0x004e8cc4 and the other four
  sibling dialogs' shape (registered in CloseAllDialogs, suppress-
  callback checked).
- F5: the Summary preview camera now seeds/re-derives retail's
  zoomed-OUT eye (byte-decoded (0,-2.5,0.95) at gmCGSummaryPage::
  InitializePage ~0x0047bd14-0x0047bd44) instead of Appearance's
  zoomed-in default, via a new ChargenPreviewController
  useZoomedOutEye flag.
- F6: retired AP-225 outright — re-derived the ListenToElementMessage
  length gate is NUL-inclusive, so MaxNameLength=32 was always
  byte-correct, not merely internally consistent.
- F7: amended AP-221 to cover the Summary preview's duplicate
  one-shot-composition binding gap (CC5 duplicated the pattern instead
  of closing it).
- F8: byte-decoded GetRandomReal @0x00563940's fmul operand at
  0x007cd650 — an 8-byte double, not a 4-byte float — is EXACTLY
  1.0/32767.0, not 1/32768. Added RollShadeLocked
  (_random.Next(32768) * (1.0/32767.0)) and switched all six shade
  rolls onto it.
- F9: evaluated porting retail's exact empty-name-commit no-op
  (NUL-inclusive length==1 skips SetName entirely) and rejected it —
  it would fight the F1 field-sync model by spontaneously reverting an
  emptied field on the next unrelated revision bump. Kept the clear,
  documented the tradeoff, filed AP-227.
- F11: filed AP-226 documenting retail's static pcProfessions/pcGender/
  pcHeritage/pcTown label tables versus acdream's DAT-sourced labels,
  including the non-human-heritage-renders-bare-"Heritage:" retail
  quirk.
- F12: added exclude-current determinism (count-2 lists), Random-
  clears-name, repeat-identical-rejection-reshows, and RebuildListbox
  content tests (the last one found F3's TemplateResolver bug).
- F13: threaded an optional Random through GameRuntimeDependencies ->
  LiveSessionController -> RuntimeCharacterCreationState, matching the
  existing TimeProvider injection shape, closing the Slice-K
  determinism hazard on a bot-reachable Randomize* command family.
- F14: RandomizeCharacterLocked now assigns _heritageId unconditionally
  before the TryGetHeritage gate, matching retail's SetHeritageGroup
  @0x005C67A0 (mHeritageGroup written before the DAT lookup).

Gates: Runtime 1726/0 (was 1722/0), App 5242/3 skips (was 5240/3),
Headless 166/0, Core.Net 993/994 (the one failure, NakEmissionTests
LossSoak, is a known pre-existing flake — passes standalone), full
solution Release build green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-08-16 01:13:52 +02:00
parent a975efd1d5
commit 0c8e1e7df1
14 changed files with 720 additions and 141 deletions
Download patch file Download diff file Expand all files Collapse all files

9
docs/architecture/retail-divergence-register.md
View file

File diff suppressed because one or more lines are too long