From 1e8596a4401572f81c525a568ea21a212fcb3671 Mon Sep 17 00:00:00 2001 From: Erik Date: Mon, 24 Aug 2026 18:09:54 +0200 Subject: [PATCH] =?UTF-8?q?fix(ui):=20retail=20tooltip=20rendering=20?= =?UTF-8?q?=E2=80=94=20formula-first=20compose=20and=20the=20two-pass=20si?= =?UTF-8?q?zing=20every=20tooltip=20was=20missing?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two corrections from the owner's retail-render oracle at the CA5 re-check, both against readings the TS-85 register row had recorded as settled: Compose (skill tooltips): retail is formula + newline + description — GetTooltip @0x004f1fe0's operator+ has the InqSkillFormula output as the LEFT operand; the old '"\n" + formula, no separator' reading had the operand order backwards and produced a leading blank line with the formula and description glued on one line. A formula-less skill (Salvaging) shows the bare description, matching the failed-InqSkillFormula branch. Sizing (ALL tooltips, per the owner's direction): retail sizes a tooltip in TWO passes (StartTooltip @0x0045DE90) — measure-wrap at the max width, resize the root through the authored ResizeTo clamps, then RecalculateGlyphList RE-WRAPS the text at its final clamped width and a second resize grows the root's HEIGHT for the extra lines. The branch the register called 'a structural no-op' IS that second pass; without it a description longer than the clamped popup stayed one clipped line, where retail shows three. ApplyTooltipText now ports the full chain, so every tooltip surface (items, options rows, character panel, world hover, map) wraps and grows exactly as retail. Pinned by BuildTooltip_FormulaFirstThenNewlineThenDescription, BuildTooltip_FormulaLessSkillShowsBareDescription, and LongTooltip_RewrapsAtTheClampedPopupWidth_AndGrowsHeightForTheExtraLines. TS-85 carries both dated corrections. Owner visual re-check owed: skill tooltip shows formula on line one, description below, long descriptions wrapping to three-plus lines inside the parchment. Full hermetic suite 15,332 passed / 0 failed. Co-Authored-By: Claude Fable 5 --- .../retail-divergence-register.md | 4 +- src/AcDream.App/Net/RetailSkillFormula.cs | 21 ++++--- .../UI/Layout/RetailTooltipPresenter.cs | 62 +++++++++++++++--- .../Net/RetailSkillFormulaTests.cs | 42 +++++++++++++ .../UI/Layout/RetailTooltipPresenterTests.cs | 63 +++++++++++++++++++ 5 files changed, 174 insertions(+), 18 deletions(-) diff --git a/docs/architecture/retail-divergence-register.md b/docs/architecture/retail-divergence-register.md index e0942e31..8f828cb9 100644 --- a/docs/architecture/retail-divergence-register.md +++ b/docs/architecture/retail-divergence-register.md @@ -423,7 +423,9 @@ AP-94..AP-112 for the confirmed retail-UI completion gaps. | # | Divergence | Where (file:line) | Why it is safe / justified | Risk if assumption breaks | Retail oracle | |---|---|---|---|---|---| -| TS-85 | **Filed 2026-08-16 at #409 (client-wide retail tooltip system); REWRITTEN at the same-day F3 review round; NARROWED again at the same-day live-failure round.** LIVE-FAILURE-ROUND NARROWING: the `m_TTText` READ side is now ported — `RetailTooltipPresenter.ResolveTooltipText` consults `UiElement.GetTooltipText()` (this port's `m_TTText`) BEFORE the authored `P0x49`, exactly as `StartTooltipAtMouse @0x00460DA3`/`@0x00460DDF` orders them, and the `P0x48`-absent fallback to the element's own layout (`@0x00460E7E`) is ported through `UiElement.SourceLayoutDid`. That lit up every acdream surface whose controller ALREADY writes runtime tooltip text (the four Options tabs, Configure Keyboard, the social pages) — live-verified 2026-08-16 on the Character tab. What remains deferred is the WRITE side at the retail `SetTooltip` call sites acdream has no analog for yet, enumerated below. Two sub-mechanisms of retail's tooltip system are unported. **(1) The `m_TTText`/`SetTooltip` runtime-text family (headed by the `P0xD0` truncated-text auto-tooltip):** the ORIGINAL filing argued this port's gap was "retail's dynamic `InqProperty(0x49)` override" — that framing is false. `UIElement::InqProperty @0x004638D0`, the BASE implementation every element uses unless its own class overrides the virtual, reads exactly the same authored property bags (`m_instanceProperties`, `m_curStateDesc`, `m_desc`) this port's `ElementReader` already walks generically — so an element with no literal `P0x49` gets NOTHING from retail's own default `InqProperty` either. The REAL second text source is the element's cached `m_TTText` field, set ONLY by the explicit, non-dat `UIElement::SetTooltip` call (`UIElement::StartTooltipAtMouse @0x00460D70` prefers `m_TTText` over the `InqProperty` fallback whenever it is non-empty). `SetTooltip` has 15+ known game-code call sites (Options rows `@0x00485E65`, chargen `@0x00481981`, the paperdoll endowment icon `@0x004C63A1`, the spellcast button `@0x004C6FE8`/`@0x004C6AAE`, and more), headed by the highest-volume one: `UIElement_Text::RecalculateTruncation @0x00466F80`, gated on authored `P0xD0` — an overflowing single/wrapped line calls `SetTooltip(this, ownText) @0x00467064` + sets enable bit 5 `@0x00467076`; a line that now fits calls `ClearTooltip @0x00467064`/clears the bit `@0x00466ff9`. `RecalculateTruncation`'s own truncation-POSITION computation (the rest of the function, `@0x004670a1` onward) walks a `GlyphList` per-line-position model (`FindCompleteLineFromY`/`FindPosFromLineAndPixels`/`FindPixelsFromPos`) this port's `UiText` has no equivalent of — `UiText` clips visually via a scissor rect (`DrawClippedText`'s `PushClip`) with no tracked "does this line overflow" state at all, so porting the auto-tooltip trigger requires building that state first. Sized as genuinely disproportionate for a single fix-round commit alongside F1-F2/F4-F11 and deferred here rather than shipped as a partial/unverified stub. A live-DAT sweep found 187 of the 430 elements authoring at least one tooltip-trigger property have NO literal `P0x49` `StringInfo` text; the live-failure round re-measured that set and found every one of the 187 authors BOTH popup-locator ids (`P0x47`+`P0x48`) — i.e. they are runtime-`SetTooltip` targets by construction, waiting only for text. **F12 correction (night-round review, 2026-08-17): this is 17 sites, not 15** — the original tally dropped `gmPaperDollUI::UpdateItemSlotTooltip @0x004A52EF` (mentioned two sentences below as its own closed row) and undercounted by one more besides. The 17 `SetTooltip` call sites, enumerated from the decomp at the live-failure round, split into: PORTED (an acdream controller already writes the text, and the presenter now reads it) — the Options rows `@0x00485E65`/`@0x00484803`/`@0x00487053`, chargen skills `@0x00481981`, the radar `@0x004D9605`; PORTED 2026-08-16 (hover-feedback completion round, docs/ISSUES.md #409/#411): inventory/shortcut item hover `UIElement_UIItem::UpdateTooltip @0x004E1CB0` — `UiItemSlot` now hardcodes the catalog's uniform popup locator (`P0x47=0x10000395`/`P0x48=0x21000041`, live-DAT-confirmed uniform across all 47 UIItem-type catalog prototypes, `TooltipLiveDatTests.UiItemCatalog_EveryPrototype_SharesTheSamePopupLocator`) and a `TooltipTextResolve` delegate wired at every physical-item construction site (inventory, external container, paperdoll — closing the separate `gmPaperDollUI::UpdateItemSlotTooltip @0x004A52EF` row below too, vendor, secure trade, toolbar), backed by the new `ClientObject.GetTooltipDisplayName()` (NAME_APPROPRIATE + the `"%d %s"` stack-count prefix, matching the decomp exactly); and the SmartBox found-object world-hover tooltip `UIElement_SmartBoxWrapper::RecvNotice_SmartBoxObjectFound @0x004E5AD0` (`@0x004E5D74`/`@0x004E5DFB`) — `RetailTooltipPresenter.UpdateWorldHoverTooltip` ports its `PlayerModule::ShowTooltips`-gated, `GetAppropriateName`-only (no stack prefix — a real, decomp-confirmed asymmetry vs. the item-cell case) trigger — **TIMING CORRECTED at the 2026-08-17 morning gate round (user finding: retail world tooltips "lag"; ours popped instantly): the original "edge-fired (no dwell)" reading was a misread — the notice's immediate `StartTooltipAtMouse @0x004E5DFB` sits inside `if (s_pInstance->m_dragElement != 0)` (`@0x004E5D8E`; `m_dragElement` is a real, distinct PDB field in `acclient.h`'s `UIElementManager`, separate from the `m_pTooltipElement` family), so the immediate mount is DRAG-ONLY; the ordinary hover path merely STAGES the name (`SetTooltip @0x004E5D74` + `|= 0x20`) and the display rides `CheckTooltip @0x0045B6E0`'s mouse-idle dwell (`m_lastMouseMoveTime` stamped on EVERY move `@0x0045e736` + `m_tooltipDelay` 0.25 s `@0x0045f75d`), with found-object changes under an idle mouse swapping the popup same-frame via `SetTooltip`'s own text-change teardown (`@0x004617FF` → `ResetTooltip @0x0045C360` tail-calling `CheckTooltip`) and the 10 s `m_tooltipDuration` expiry requiring a fresh mouse move before re-arming (`SwitchMouseOver(null) @0x0045b7b2`) — all now ported, including the drag-immediate branch**, reusing the SAME popup locator since an exhaustive DAT sweep found `UIElement_SmartBoxWrapper` (class `0x10000030`) has no authored `ElementDesc` anywhere installed (`TooltipLiveDatTests.SmartBoxWrapper_HasNoAuthoredElementDesc_AnywhereInstalled`) — the popup-skin choice is therefore the best-evidenced inference, not a measured value, and is called out here as such. **BATCH B (2026-08-17) CLOSED the spellcasting and character-panel rows of this list. BATCH C (2026-08-17, Map/House toolbar panel) CLOSES THE LAST REMAINING ITEM: `gmMapUI::AddMapNote @0x004A1C51`'s 53 town-hotspot tooltips are now ported via `MapPageController.BuildTownMarkers` (`src/AcDream.App/UI/Layout/MapPageController.cs`), setting `UiButton.TooltipText` (retail's RUNTIME `m_TTText`/`SetTooltip` mechanism, not the DAT-authored `P0x49` path an earlier same-day cut mistakenly used and which never rendered live during verification); literal town names from `MapLocations.cs` (a verbatim port of `s_rgLocations`), not a DAT string-table lookup, matching `AddMapNote`'s own `StringInfo::SetLiteralValue` call. CORRECTED at the same-day morning gate round (user finding 3 — the retail screenshot's green hover highlight + special-font parchment tooltip): Batch C's "the town-marker template authors no locator of its own" claim was WRONG — the template (`0x100001F0` in `0x21000026`, `MapNoteLiveDatTests`) authors its OWN `P0x47=0x10000398`/`P0x48=0x21000041` (the fourth popup skin, whose incorporated text child `0x10000396` fonts `0x40000015` where the other three skins font `0x40000002`), a zero per-element delay `P0x50=0.0`, `P0x4B` TooltipOn, and `P0x13` RolloverEnabled with PassToChildren `Normal`/`Normal_rollover` states flipping the highlight child `0x100001F1`'s per-state `P0x3B` (the green `0x06004CC9` frame, byte-decoded A=FF R=00 G=FF B=00); the hardcoded shared-skin override was removed (the built marker's authored locator wins) and the rollover highlight + per-state-`P0x3B` + button PassToChildren cascade are now ported (`UiButton.CascadeStateToChildren`, `UiDatElement.TrySetRetailState`'s 0x3B honor). Sub-mechanism (1)'s `SetTooltip`-call-site enumeration is 16 of 17 known sites PORTED — `UIElement_Text::RecalculateTruncation @0x00466F80` (the headline, highest-volume site named at the top of sub-mechanism (1)) remains the ONE open item, exactly as this row's own sub-mechanism (1) text above already scoped it out (its own "Sized as genuinely disproportionate... deferred here" note). The prior "all 15 known sites accounted for" close (F12 correction, night-round review) was wrong twice over: the count is 17, not 15, and RecalculateTruncation was never actually ported — it was always the one deliberately-deferred item, not a closed one.** Batch B audit findings: the endowment icon `@0x004C63A1`, favorite `@0x004C7206`, and submenu `@0x004C67D8` sites turned out to be ALREADY CORRECT — all three are `UiCatalogSlot`-based and the pre-existing `Label`-driven `GetTooltipText` already carried retail's exact text (`SpellCastSubMenu::AddFavorite @0x004C7060`/`UpdateFromPlayerModule @0x004C6570` both build a single-arg `Formatted` PStringBase — plain spell name, no wrapper — for the favorite-bar/submenu case; `gmSpellcastingUI::UpdateEndowmentIcon @0x004C6120` mirrors the cast button's confirmed `"%s (%hs)"` literal at `data_7b64d8` — item name, then spell name in parens — via the identical two-value narrow/wide prep sequence, one address away in the same function family). The cast button `@0x004C6FE8`/`ClearTooltip @0x004C6AAE` was the one real gap (`UiButton` had no tooltip wiring at all): now ported via `SpellcastingUiController.UpdateCastAvailability`/`ComputeEndowmentCastState`, sourced from `gmSpellcastingUI::UpdateCastButtonTooltip @0x004C6A30`. Fully verified literal text: the no-selection states (`"Select a spell to cast"` @ `data_7b64ec`, `"You have no spells ready to cast"` @ `data_7b6520`) and the complete endowment-item branch (`"USE the %s"` @ `data_7b64c0`, `" on %s"` @ `data_7b6464`, `"You must select a target for the %s"` @ `data_7b6478`; `ItemUses::IsUseable_SelfTarget @0x004fcd30` is exactly `ItemUseability.AllowsSelfTarget`). NOT ported: the endowment branch's incompatible-target sub-state (`"You must select an appropriate\ntarget for the %s"` @ `data_7b6400`, gated by `ItemHolder::TargetCompatibleWithObject @0x00587520` — a ~400-line function with its own chat-message side effects, out of scope for a tooltip batch; a present target is optimistically treated as compatible, same text as the confirmed-compatible case). **CORRECTED at the night-round review (F3/F4, 2026-08-17): the plain-spell branch's wording is NOT unrecoverable — the "genuine `gmNoticeHandler` vtable SLOTS" claim above was itself the artifact. `PStringBase::sprintf`'s second argument at those three call sites is a raw pushed literal (a plain `push 0x7bXXXX; call sprintf`); Binary Ninja's pseudo-C rendering of that operand as `&gmSpellcastingUI::\`vftable'.RecvNotice_XXX` was a spurious symbol match, not the true operand — a direct capstone disassembly of the raw bytes at `0x4c6e48`/`0x4c6ea4`/`0x4c6f18`/`0x4c6f5d` resolves the actual constants: `"CAST %hs"` @0x7b63a4 (untargeted/self-cast at `0x4c6f35`, and targeted+compatible at `0x4c6e57` — both enabled, the latter appending `" on %s"` @0x7b6464 with the target's name), `"You must select an appropriate target for %hs"` @0x7b6348 (targeted+incompatible, disabled), `"You must select a target for %hs"` @0x7b63b8 (no target, disabled); `%hs` is the spell's own name in all four call sites (`CSpellBase::InqName`, the same call `0x5bbee0` throughout). Now ported: `RuntimeSpellCastState.EvaluateCastGate` (the four-state gate) + `SpellcastingUiController.ComputeSpellCastState`. Also corrected the endowment branch's "USE the %s" operand: it was NOT the bare item name (F4) — the vararg to `"USE the %s"`/`"You must select a target for the %s"`/the still-unported incompatible-target string is the SAME composed `"%s (%hs)"` string (item name, spell name) built once at `@0x004c6bb6-ef` from format literal `data_7b64d8`, byte-confirmed by all three sprintf call sites (`0x4c6c7f`/`0x4c6ca4`/`0x4c6d46`) reading the identical `[esp+0x18]` slot — now ported via `SpellcastingUiController.ComposeEndowmentName`.** **#430 CORRECTION (2026-08-24): the character-panel port below set the TEXT but omitted the popup LOCATOR on the runtime-built rows, so these tooltips never mounted until the CA5-adjacent fix gave rows the shared 0x10000395/0x21000041 skin (live-DAT probed; UiItemSlot precedent).** The character-panel `AttributeInfoRegion @0x004F1617` / `Attribute2ndInfoRegion @0x004F1777` / `SkillInfoRegion @0x004F222F` constructors are now fully ported through a new `UiClickablePanel.TooltipText` (the same settable-string seam as `UiButton.TooltipText`): the six hardcoded attribute descriptions (`SkillSystem::InqAttributeDescription @0x005c8e30`) and three hardcoded, pair-shared vitals descriptions (`SkillSystem::InqAttribute2ndDescription @0x005c8f70`) were byte-decoded from the retail string pool (the pseudo-C dump truncates them with "…"); skill tooltips compose `SkillInfoRegion::GetTooltip @0x004f1fe0`'s exact `"\n" + formula + description` (ported verbatim, including the confirmed lack of any separator between the formula and description text) from the ALREADY-DAT-parsed `DatReaderWriter.Types.SkillBase.Description`/`.Formula` fields (portal `0x0E000004`, the same resource `CharacterSheetProvider.SkillTable` already reads for skill names/costs) rather than hand-transcribed literals — no guessing was needed for the ~30+ skill description strings. The formula-to-text algorithm itself (`SkillSystem::InqSkillFormula @0x005c89b0`, e.g. producing `"( (Strength + Coordination) / 2 )"`) was fully recovered by byte-decoding six short literal fragments (`data_7e7930`/`7e7934`/`7e7940`/`7e7950`/`7e7954`/`797584`) the pseudo-C dump left completely unlabeled — they sit between two `gmSpellcastingUI` vtable declarations and Binary Ninja's type inference never recognized them as strings, so the raw hex had to be read directly as narrow ASCII (confirmed against the function's own directly-visible `" / %u"` and `"(%u x %s)"` literals, which needed no such recovery). Retail's runtime sites also SET the `P0x4B` on-bit themselves (`__bitfield164 |= 0x20`, eight sites) — the port models that as "runtime text present implies tooltip-on", so only the authored-text path consults the authored bit. **(2) The per-element wrap-width override:** `UIElement_Text::InqSizewMargins @0x00469660`'s `UITS_MAX_WIDTH` branch checks `GetAttribute_Int(this, 0x3D, ...)` before falling back to `RenderDevice::GetDisplayWidth()`; `RetailTooltipPresenter.ApplyTooltipText` always wraps at `UiRoot.EffectiveCanvasSize.X` (the confirmed fallback) and never checks for a `P0x3D` override — the live-DAT sweep found zero tooltip-bearing elements author one. | `src/AcDream.App/UI/Layout/RetailTooltipPresenter.cs` (`ResolveTooltipText`'s runtime-then-authored order; `ApplyTooltipText`'s wrap-width literal; `UpdateWorldHoverTooltip`/`TryBuildAndMountPopup` — the world-hover half added 2026-08-16); `src/AcDream.App/UI/Layout/ElementReader.cs` (`ElementInfo.TooltipText`'s own doc comment carries the same F3 correction); `src/AcDream.App/UI/UiItemSlot.cs` (`TooltipTextResolve`, `GetTooltipText`, the hardcoded popup-locator constants); `src/AcDream.Core/Items/ClientObject.cs` (`GetTooltipDisplayName`); `src/AcDream.App/UI/CursorFeedbackController.cs` (the related #411 found-cursor fix, same round — see that row); Batch B (2026-08-17) additions: `src/AcDream.App/UI/UiPanel.cs` (`UiClickablePanel.TooltipText`/`GetTooltipText`); `src/AcDream.App/UI/Layout/SpellcastingUiController.cs` (`UpdateCastAvailability`, `ComputeEndowmentCastState`); night-round review (F3/F4, 2026-08-17) additions: `src/AcDream.App/UI/Layout/SpellcastingUiController.cs` (`ComputeSpellCastState`, `ComposeEndowmentName`); `src/AcDream.Runtime/Gameplay/RuntimeSpellCastState.cs` (`EvaluateCastGate`, `SpellCastGate`); `src/AcDream.App/Net/RetailSkillFormula.cs` (`AttributeName`, `FormatFormula`, `BuildTooltip`); `src/AcDream.App/UI/Layout/CharacterSheet.cs` (`CharacterSkill.TooltipText`); `src/AcDream.App/UI/Layout/CharacterSheetProvider.cs` (`BuildLiveCharacterSkills`'s tooltip compose); `src/AcDream.App/UI/Layout/CharacterStatController.cs` (`AttributeDescriptions`, `Attribute2ndDescriptions`, `BuildAttributeRows`/`BuildSkillRows` row wiring) | The 243 elements WITH literal text — the "core" case #409 ships — cover every hover-text scenario the investigation's own landmark checks exercised (main-game-UI Appearance-page rotate/color/spin hints, etc.). F9 correction: "243 with literal text" is not automatically "243 showable" — `RetailTooltipPresenter.OnTooltipShow`'s real gate is the FULL conjunction of `TooltipEnabled` (P0x4B) AND non-null text (P0x49) AND both popup-locator ids (P0x47/P0x48) — so this was measured, not assumed: `TooltipLiveDatTests.ClientWideSweep_FindsKnownLandmarksAndAFloorCount`'s `Showable` column finds the intersection is exactly 243, i.e. every element authoring literal text also authors the other three properties together (they are evidently authored as one group in practice). The P0x3D sweep found zero authoring elements, so the unconditional display-width fallback is not an approximation for any element that exists today. | If a future DAT revision adds a `P0xD0`-truncated text element, a game-code `SetTooltip` caller, or authors a `P0x3D` override, it silently shows no tooltip / wraps at the wrong width instead of erroring — indistinguishable from "the element authors no tooltip at all" without re-running the sweep AND separately auditing which of the 187 no-literal-text elements would actually truncate at their authored width. | `UIElement::InqProperty @0x004638D0` (base authored-bag read, NOT a dynamic override); `UIElement::StartTooltipAtMouse @0x00460D70` (`m_TTText`-vs-`InqProperty` preference order); `UIElement_Text::RecalculateTruncation @0x00466F80` (`P0xD0` gate, `SetTooltip`/`ClearTooltip` sites); `UIElement_Text::InqSizewMargins @0x00469660` (`UITS_MAX_WIDTH` branch, `GetAttribute_Int(this, 0x3D, ...)`); Batch B (2026-08-17): `gmSpellcastingUI::UpdateCastButtonTooltip @0x004C6A30` (cast-button state machine); `gmSpellcastingUI::UpdateEndowmentIcon @0x004C6120` (endowment-icon `"%s (%hs)"` confirmation); `SpellCastSubMenu::AddFavorite @0x004C7060` / `UpdateFromPlayerModule @0x004C6570` (favorite/submenu plain-name confirmation); `ItemUses::IsUseable_SelfTarget @0x004fcd30`; `AttributeInfoRegion::AttributeInfoRegion @0x004f1530` / `Attribute2ndInfoRegion::Attribute2ndInfoRegion @0x004f1680` / `SkillInfoRegion::SkillInfoRegion @0x004f2140` / `SkillInfoRegion::GetTooltip @0x004f1fe0`; `SkillSystem::InqAttributeName @0x005c8d90` / `InqAttributeDescription @0x005c8e30` / `InqAttribute2ndName @0x005c8ed0` / `InqAttribute2ndDescription @0x005c8f70` / `InqSkillFormula @0x005c89b0` | +| TS-85 | **Filed 2026-08-16 at #409 (client-wide retail tooltip system); REWRITTEN at the same-day F3 review round; NARROWED again at the same-day live-failure round.** LIVE-FAILURE-ROUND NARROWING: the `m_TTText` READ side is now ported — `RetailTooltipPresenter.ResolveTooltipText` consults `UiElement.GetTooltipText()` (this port's `m_TTText`) BEFORE the authored `P0x49`, exactly as `StartTooltipAtMouse @0x00460DA3`/`@0x00460DDF` orders them, and the `P0x48`-absent fallback to the element's own layout (`@0x00460E7E`) is ported through `UiElement.SourceLayoutDid`. That lit up every acdream surface whose controller ALREADY writes runtime tooltip text (the four Options tabs, Configure Keyboard, the social pages) — live-verified 2026-08-16 on the Character tab. What remains deferred is the WRITE side at the retail `SetTooltip` call sites acdream has no analog for yet, enumerated below. Two sub-mechanisms of retail's tooltip system are unported. **(1) The `m_TTText`/`SetTooltip` runtime-text family (headed by the `P0xD0` truncated-text auto-tooltip):** the ORIGINAL filing argued this port's gap was "retail's dynamic `InqProperty(0x49)` override" — that framing is false. `UIElement::InqProperty @0x004638D0`, the BASE implementation every element uses unless its own class overrides the virtual, reads exactly the same authored property bags (`m_instanceProperties`, `m_curStateDesc`, `m_desc`) this port's `ElementReader` already walks generically — so an element with no literal `P0x49` gets NOTHING from retail's own default `InqProperty` either. The REAL second text source is the element's cached `m_TTText` field, set ONLY by the explicit, non-dat `UIElement::SetTooltip` call (`UIElement::StartTooltipAtMouse @0x00460D70` prefers `m_TTText` over the `InqProperty` fallback whenever it is non-empty). `SetTooltip` has 15+ known game-code call sites (Options rows `@0x00485E65`, chargen `@0x00481981`, the paperdoll endowment icon `@0x004C63A1`, the spellcast button `@0x004C6FE8`/`@0x004C6AAE`, and more), headed by the highest-volume one: `UIElement_Text::RecalculateTruncation @0x00466F80`, gated on authored `P0xD0` — an overflowing single/wrapped line calls `SetTooltip(this, ownText) @0x00467064` + sets enable bit 5 `@0x00467076`; a line that now fits calls `ClearTooltip @0x00467064`/clears the bit `@0x00466ff9`. `RecalculateTruncation`'s own truncation-POSITION computation (the rest of the function, `@0x004670a1` onward) walks a `GlyphList` per-line-position model (`FindCompleteLineFromY`/`FindPosFromLineAndPixels`/`FindPixelsFromPos`) this port's `UiText` has no equivalent of — `UiText` clips visually via a scissor rect (`DrawClippedText`'s `PushClip`) with no tracked "does this line overflow" state at all, so porting the auto-tooltip trigger requires building that state first. Sized as genuinely disproportionate for a single fix-round commit alongside F1-F2/F4-F11 and deferred here rather than shipped as a partial/unverified stub. A live-DAT sweep found 187 of the 430 elements authoring at least one tooltip-trigger property have NO literal `P0x49` `StringInfo` text; the live-failure round re-measured that set and found every one of the 187 authors BOTH popup-locator ids (`P0x47`+`P0x48`) — i.e. they are runtime-`SetTooltip` targets by construction, waiting only for text. **F12 correction (night-round review, 2026-08-17): this is 17 sites, not 15** — the original tally dropped `gmPaperDollUI::UpdateItemSlotTooltip @0x004A52EF` (mentioned two sentences below as its own closed row) and undercounted by one more besides. The 17 `SetTooltip` call sites, enumerated from the decomp at the live-failure round, split into: PORTED (an acdream controller already writes the text, and the presenter now reads it) — the Options rows `@0x00485E65`/`@0x00484803`/`@0x00487053`, chargen skills `@0x00481981`, the radar `@0x004D9605`; PORTED 2026-08-16 (hover-feedback completion round, docs/ISSUES.md #409/#411): inventory/shortcut item hover `UIElement_UIItem::UpdateTooltip @0x004E1CB0` — `UiItemSlot` now hardcodes the catalog's uniform popup locator (`P0x47=0x10000395`/`P0x48=0x21000041`, live-DAT-confirmed uniform across all 47 UIItem-type catalog prototypes, `TooltipLiveDatTests.UiItemCatalog_EveryPrototype_SharesTheSamePopupLocator`) and a `TooltipTextResolve` delegate wired at every physical-item construction site (inventory, external container, paperdoll — closing the separate `gmPaperDollUI::UpdateItemSlotTooltip @0x004A52EF` row below too, vendor, secure trade, toolbar), backed by the new `ClientObject.GetTooltipDisplayName()` (NAME_APPROPRIATE + the `"%d %s"` stack-count prefix, matching the decomp exactly); and the SmartBox found-object world-hover tooltip `UIElement_SmartBoxWrapper::RecvNotice_SmartBoxObjectFound @0x004E5AD0` (`@0x004E5D74`/`@0x004E5DFB`) — `RetailTooltipPresenter.UpdateWorldHoverTooltip` ports its `PlayerModule::ShowTooltips`-gated, `GetAppropriateName`-only (no stack prefix — a real, decomp-confirmed asymmetry vs. the item-cell case) trigger — **TIMING CORRECTED at the 2026-08-17 morning gate round (user finding: retail world tooltips "lag"; ours popped instantly): the original "edge-fired (no dwell)" reading was a misread — the notice's immediate `StartTooltipAtMouse @0x004E5DFB` sits inside `if (s_pInstance->m_dragElement != 0)` (`@0x004E5D8E`; `m_dragElement` is a real, distinct PDB field in `acclient.h`'s `UIElementManager`, separate from the `m_pTooltipElement` family), so the immediate mount is DRAG-ONLY; the ordinary hover path merely STAGES the name (`SetTooltip @0x004E5D74` + `|= 0x20`) and the display rides `CheckTooltip @0x0045B6E0`'s mouse-idle dwell (`m_lastMouseMoveTime` stamped on EVERY move `@0x0045e736` + `m_tooltipDelay` 0.25 s `@0x0045f75d`), with found-object changes under an idle mouse swapping the popup same-frame via `SetTooltip`'s own text-change teardown (`@0x004617FF` → `ResetTooltip @0x0045C360` tail-calling `CheckTooltip`) and the 10 s `m_tooltipDuration` expiry requiring a fresh mouse move before re-arming (`SwitchMouseOver(null) @0x0045b7b2`) — all now ported, including the drag-immediate branch**, reusing the SAME popup locator since an exhaustive DAT sweep found `UIElement_SmartBoxWrapper` (class `0x10000030`) has no authored `ElementDesc` anywhere installed (`TooltipLiveDatTests.SmartBoxWrapper_HasNoAuthoredElementDesc_AnywhereInstalled`) — the popup-skin choice is therefore the best-evidenced inference, not a measured value, and is called out here as such. **BATCH B (2026-08-17) CLOSED the spellcasting and character-panel rows of this list. BATCH C (2026-08-17, Map/House toolbar panel) CLOSES THE LAST REMAINING ITEM: `gmMapUI::AddMapNote @0x004A1C51`'s 53 town-hotspot tooltips are now ported via `MapPageController.BuildTownMarkers` (`src/AcDream.App/UI/Layout/MapPageController.cs`), setting `UiButton.TooltipText` (retail's RUNTIME `m_TTText`/`SetTooltip` mechanism, not the DAT-authored `P0x49` path an earlier same-day cut mistakenly used and which never rendered live during verification); literal town names from `MapLocations.cs` (a verbatim port of `s_rgLocations`), not a DAT string-table lookup, matching `AddMapNote`'s own `StringInfo::SetLiteralValue` call. CORRECTED at the same-day morning gate round (user finding 3 — the retail screenshot's green hover highlight + special-font parchment tooltip): Batch C's "the town-marker template authors no locator of its own" claim was WRONG — the template (`0x100001F0` in `0x21000026`, `MapNoteLiveDatTests`) authors its OWN `P0x47=0x10000398`/`P0x48=0x21000041` (the fourth popup skin, whose incorporated text child `0x10000396` fonts `0x40000015` where the other three skins font `0x40000002`), a zero per-element delay `P0x50=0.0`, `P0x4B` TooltipOn, and `P0x13` RolloverEnabled with PassToChildren `Normal`/`Normal_rollover` states flipping the highlight child `0x100001F1`'s per-state `P0x3B` (the green `0x06004CC9` frame, byte-decoded A=FF R=00 G=FF B=00); the hardcoded shared-skin override was removed (the built marker's authored locator wins) and the rollover highlight + per-state-`P0x3B` + button PassToChildren cascade are now ported (`UiButton.CascadeStateToChildren`, `UiDatElement.TrySetRetailState`'s 0x3B honor). Sub-mechanism (1)'s `SetTooltip`-call-site enumeration is 16 of 17 known sites PORTED — `UIElement_Text::RecalculateTruncation @0x00466F80` (the headline, highest-volume site named at the top of sub-mechanism (1)) remains the ONE open item, exactly as this row's own sub-mechanism (1) text above already scoped it out (its own "Sized as genuinely disproportionate... deferred here" note). The prior "all 15 known sites accounted for" close (F12 correction, night-round review) was wrong twice over: the count is 17, not 15, and RecalculateTruncation was never actually ported — it was always the one deliberately-deferred item, not a closed one.** Batch B audit findings: the endowment icon `@0x004C63A1`, favorite `@0x004C7206`, and submenu `@0x004C67D8` sites turned out to be ALREADY CORRECT — all three are `UiCatalogSlot`-based and the pre-existing `Label`-driven `GetTooltipText` already carried retail's exact text (`SpellCastSubMenu::AddFavorite @0x004C7060`/`UpdateFromPlayerModule @0x004C6570` both build a single-arg `Formatted` PStringBase — plain spell name, no wrapper — for the favorite-bar/submenu case; `gmSpellcastingUI::UpdateEndowmentIcon @0x004C6120` mirrors the cast button's confirmed `"%s (%hs)"` literal at `data_7b64d8` — item name, then spell name in parens — via the identical two-value narrow/wide prep sequence, one address away in the same function family). The cast button `@0x004C6FE8`/`ClearTooltip @0x004C6AAE` was the one real gap (`UiButton` had no tooltip wiring at all): now ported via `SpellcastingUiController.UpdateCastAvailability`/`ComputeEndowmentCastState`, sourced from `gmSpellcastingUI::UpdateCastButtonTooltip @0x004C6A30`. Fully verified literal text: the no-selection states (`"Select a spell to cast"` @ `data_7b64ec`, `"You have no spells ready to cast"` @ `data_7b6520`) and the complete endowment-item branch (`"USE the %s"` @ `data_7b64c0`, `" on %s"` @ `data_7b6464`, `"You must select a target for the %s"` @ `data_7b6478`; `ItemUses::IsUseable_SelfTarget @0x004fcd30` is exactly `ItemUseability.AllowsSelfTarget`). NOT ported: the endowment branch's incompatible-target sub-state (`"You must select an appropriate\ntarget for the %s"` @ `data_7b6400`, gated by `ItemHolder::TargetCompatibleWithObject @0x00587520` — a ~400-line function with its own chat-message side effects, out of scope for a tooltip batch; a present target is optimistically treated as compatible, same text as the confirmed-compatible case). **CORRECTED at the night-round review (F3/F4, 2026-08-17): the plain-spell branch's wording is NOT unrecoverable — the "genuine `gmNoticeHandler` vtable SLOTS" claim above was itself the artifact. `PStringBase::sprintf`'s second argument at those three call sites is a raw pushed literal (a plain `push 0x7bXXXX; call sprintf`); Binary Ninja's pseudo-C rendering of that operand as `&gmSpellcastingUI::\`vftable'.RecvNotice_XXX` was a spurious symbol match, not the true operand — a direct capstone disassembly of the raw bytes at `0x4c6e48`/`0x4c6ea4`/`0x4c6f18`/`0x4c6f5d` resolves the actual constants: `"CAST %hs"` @0x7b63a4 (untargeted/self-cast at `0x4c6f35`, and targeted+compatible at `0x4c6e57` — both enabled, the latter appending `" on %s"` @0x7b6464 with the target's name), `"You must select an appropriate target for %hs"` @0x7b6348 (targeted+incompatible, disabled), `"You must select a target for %hs"` @0x7b63b8 (no target, disabled); `%hs` is the spell's own name in all four call sites (`CSpellBase::InqName`, the same call `0x5bbee0` throughout). Now ported: `RuntimeSpellCastState.EvaluateCastGate` (the four-state gate) + `SpellcastingUiController.ComputeSpellCastState`. Also corrected the endowment branch's "USE the %s" operand: it was NOT the bare item name (F4) — the vararg to `"USE the %s"`/`"You must select a target for the %s"`/the still-unported incompatible-target string is the SAME composed `"%s (%hs)"` string (item name, spell name) built once at `@0x004c6bb6-ef` from format literal `data_7b64d8`, byte-confirmed by all three sprintf call sites (`0x4c6c7f`/`0x4c6ca4`/`0x4c6d46`) reading the identical `[esp+0x18]` slot — now ported via `SpellcastingUiController.ComposeEndowmentName`.** **CA5-GATE CORRECTIONS (2026-08-24, owner retail-render oracle): (a) `SkillInfoRegion::GetTooltip`'s compose is `formula + " +" + description` — the row below has the operand order backwards (" +"+formula, "no separator"); the operator+ left operand is the InqSkillFormula output, and retail renders formula-first-line/description-below. (b) The "grow further if vertical scroll overflow" branch this row calls "a structural no-op" is NOT one — it is retail's SECOND sizing pass (`StartTooltip @0x0045DE90`: measure-wrap at max width → clamped root resize → `RecalculateGlyphList` RE-WRAP at the final clamped width → height growth for the extra lines), which is what makes long tooltips multi-line; both now ported in `RetailTooltipPresenter.ApplyTooltipText`/`RetailSkillFormula.BuildTooltip`.** **#430 CORRECTION (2026-08-24): the character-panel port below set the TEXT but omitted the popup LOCATOR on the runtime-built rows, so these tooltips never mounted until the CA5-adjacent fix gave rows the shared 0x10000395/0x21000041 skin (live-DAT probed; UiItemSlot precedent).** The character-panel `AttributeInfoRegion @0x004F1617` / `Attribute2ndInfoRegion @0x004F1777` / `SkillInfoRegion @0x004F222F` constructors are now fully ported through a new `UiClickablePanel.TooltipText` (the same settable-string seam as `UiButton.TooltipText`): the six hardcoded attribute descriptions (`SkillSystem::InqAttributeDescription @0x005c8e30`) and three hardcoded, pair-shared vitals descriptions (`SkillSystem::InqAttribute2ndDescription @0x005c8f70`) were byte-decoded from the retail string pool (the pseudo-C dump truncates them with "…"); skill tooltips compose `SkillInfoRegion::GetTooltip @0x004f1fe0`'s exact `"\n" + formula + description` (ported verbatim, including the confirmed lack of any separator between the formula and description text) from the ALREADY-DAT-parsed `DatReaderWriter.Types.SkillBase.Description`/`.Formula` fields (portal `0x0E000004`, the same resource `CharacterSheetProvider.SkillTable` already reads for skill names/costs) rather than hand-transcribed literals — no guessing was needed for the ~30+ skill description strings. The formula-to-text algorithm itself (`SkillSystem::InqSkillFormula @0x005c89b0`, e.g. producing `"( (Strength + Coordination) / 2 )"`) was fully recovered by byte-decoding six short literal fragments (`data_7e7930`/`7e7934`/`7e7940`/`7e7950`/`7e7954`/`797584`) the pseudo-C dump left completely unlabeled — they sit between two `gmSpellcastingUI` vtable declarations and Binary Ninja's type inference never recognized them as strings, so the raw hex had to be read directly as narrow ASCII (confirmed against the function's own directly-visible `" / %u"` and `"(%u x %s)"` literals, which needed no such recovery). Retail's runtime sites also SET the `P0x4B` on-bit themselves (`__bitfield164 |= 0x20`, eight sites) — the port models that as "runtime text present implies tooltip-on", so only the authored-text path consults the authored bit. **(2) The per-element wrap-width override:** `UIElement_Text::InqSizewMargins @0x00469660`'s `UITS_MAX_WIDTH` branch checks `GetAttribute_Int(this, 0x3D, ...)` before falling back to `RenderDevice::GetDisplayWidth()`; `RetailTooltipPresenter.ApplyTooltipText` always wraps at `UiRoot.EffectiveCanvasSize.X` (the confirmed fallback) and never checks for a `P0x3D` override — the live-DAT sweep found zero tooltip-bearing elements author one. | `src/AcDream.App/UI/Layout/RetailTooltipPresenter.cs` (`ResolveTooltipText`'s runtime-then-authored order; `ApplyTooltipText`'s wrap-width literal; `UpdateWorldHoverTooltip`/`TryBuildAndMountPopup` — the world-hover half added 2026-08-16); `src/AcDream.App/UI/Layout/ElementReader.cs` (`ElementInfo.TooltipText`'s own doc comment carries the same F3 correction); `src/AcDream.App/UI/UiItemSlot.cs` (`TooltipTextResolve`, `GetTooltipText`, the hardcoded popup-locator constants); `src/AcDream.Core/Items/ClientObject.cs` (`GetTooltipDisplayName`); `src/AcDream.App/UI/CursorFeedbackController.cs` (the related #411 found-cursor fix, same round — see that row); Batch B (2026-08-17) additions: `src/AcDream.App/UI/UiPanel.cs` (`UiClickablePanel.TooltipText`/`GetTooltipText`); `src/AcDream.App/UI/Layout/SpellcastingUiController.cs` (`UpdateCastAvailability`, `ComputeEndowmentCastState`); night-round review (F3/F4, 2026-08-17) additions: `src/AcDream.App/UI/Layout/SpellcastingUiController.cs` (`ComputeSpellCastState`, `ComposeEndowmentName`); `src/AcDream.Runtime/Gameplay/RuntimeSpellCastState.cs` (`EvaluateCastGate`, `SpellCastGate`); `src/AcDream.App/Net/RetailSkillFormula.cs` (`AttributeName`, `FormatFormula`, `BuildTooltip`); `src/AcDream.App/UI/Layout/CharacterSheet.cs` (`CharacterSkill.TooltipText`); `src/AcDream.App/UI/Layout/CharacterSheetProvider.cs` (`BuildLiveCharacterSkills`'s tooltip compose); `src/AcDream.App/UI/Layout/CharacterStatController.cs` (`AttributeDescriptions`, `Attribute2ndDescriptions`, `BuildAttributeRows`/`BuildSkillRows` row wiring) | The 243 elements WITH literal text — the "core" case #409 ships — cover every hover-text scenario the investigation's own landmark checks exercised (main-game-UI Appearance-page rotate/color/spin hints, etc.). F9 correction: "243 with literal text" is not automatically "243 showable" — `RetailTooltipPresenter.OnTooltipShow`'s real gate is the FULL conjunction of `TooltipEnabled` (P0x4B) AND non-null text (P0x49) AND both popup-locator ids (P0x47/P0x48) — so this was measured, not assumed: `TooltipLiveDatTests.ClientWideSweep_FindsKnownLandmarksAndAFloorCount`'s `Showable` column finds the intersection is exactly 243, i.e. every element authoring literal text also authors the other three properties together (they are evidently authored as one group in practice). The P0x3D sweep found zero authoring elements, so the unconditional display-width fallback is not an approximation for any element that exists today. | If a future DAT revision adds a `P0xD0`-truncated text element, a game-code `SetTooltip` caller, or authors a `P0x3D` override, it silently shows no tooltip / wraps at the wrong width instead of erroring — indistinguishable from "the element authors no tooltip at all" without re-running the sweep AND separately auditing which of the 187 no-literal-text elements would actually truncate at their authored width. | `UIElement::InqProperty @0x004638D0` (base authored-bag read, NOT a dynamic override); `UIElement::StartTooltipAtMouse @0x00460D70` (`m_TTText`-vs-`InqProperty` preference order); `UIElement_Text::RecalculateTruncation @0x00466F80` (`P0xD0` gate, `SetTooltip`/`ClearTooltip` sites); `UIElement_Text::InqSizewMargins @0x00469660` (`UITS_MAX_WIDTH` branch, `GetAttribute_Int(this, 0x3D, ...)`); Batch B (2026-08-17): `gmSpellcastingUI::UpdateCastButtonTooltip @0x004C6A30` (cast-button state machine); `gmSpellcastingUI::UpdateEndowmentIcon @0x004C6120` (endowment-icon `"%s (%hs)"` confirmation); `SpellCastSubMenu::AddFavorite @0x004C7060` / `UpdateFromPlayerModule @0x004C6570` (favorite/submenu plain-name confirmation); `ItemUses::IsUseable_SelfTarget @0x004fcd30`; `AttributeInfoRegion::AttributeInfoRegion @0x004f1530` / `Attribute2ndInfoRegion::Attribute2ndInfoRegion @0x004f1680` / `SkillInfoRegion::SkillInfoRegion @0x004f2140` / `SkillInfoRegion::GetTooltip @0x004f1fe0`; `SkillSystem::InqAttributeName @0x005c8d90` / `InqAttributeDescription @0x005c8e30` / `InqAttribute2ndName @0x005c8ed0` / `InqAttribute2ndDescription @0x005c8f70` / `InqSkillFormula @0x005c89b0` | | TS-84 | Chargen 3D preview (Campaign CC slice CC6a foundation): `ChargenClothingTable`'s composer skips retail's ~8-branch Setup-id substitution chain (`ClothingTable::BuildObjDesc @ 0x005A7900`'s Umbraen/Penumbraen/Undead/Anakshay fallback) when a garment's `ClothingBaseEffects` has no entry for the resolved body Setup. MEASURED (not assumed) against the installed EoR dat across all 26 heritage/gender combinations via `ChargenAppearanceCatalogInstalledDatTests`, with the measurement now PINNED by a real assertion rather than diagnostic-only output (review fix round F7): the 9 standard heritages whose UI actually shows clothing controls resolve every default gear choice with zero coverage gaps. Undead is a real gap — its default gear choices (both genders) have NO base-effect entry on **ALL FOUR clothing slots — headgear, trousers, shirt, AND footwear** (not the three-slot "headgear/trousers/footwear" this row originally understated, with a self-contradicting "4 of 4 non-shirt slots" aside — corrected at the review fix round F2) — for Undead's own live body Setup (male 0x02001A9C / female 0x02001AA0), because that Setup is one of the skeleton/zombie variants the un-ported chain exists to redirect. The four measured missing clothing-table ids are identical on both genders and in a fixed order: `0x10000009, 0x100000F9, 0x10000001, 0x10000007` (Headgear, Trousers, Shirt, Footwear — the factory's own composition order). Gear Knight and both Olthoi variants also show gaps under a synthetic "select every offered option" sweep, but retail hides the clothing controls entirely for those three heritages (`gmCGAppearancePage::Update @ 0x0047E8F0`'s `m_pClothesButton->SetVisible(0)` branches for `mHeritageGroup == 6` and `== 0xc \|\| == 0xd`), so a real chargen selection never reaches them — not a live gap. | `src/AcDream.Core/CharGen/ChargenClothingTable.cs`; `src/AcDream.Core/CharGen/ChargenAppearanceFactory.cs` (`ComposeClothingSlot`) | CC6a is explicitly the rendering-foundation slice (index→ObjDesc factory + static-pose offscreen renderer, no page mount yet); porting the ~8-branch substitution chain is bounded follow-up work once CC6b wires real clothing-slot UI, not a blocker for the foundation deliverable — and the installed-DAT test proves the gap is narrow (one heritage, all four of ITS slots) rather than pervasive. | Undead's default clothing preview renders the bare body mesh for ALL FOUR slots — headgear, trousers, shirt, AND footwear (no clothing part/texture override applied on any of them, though the dye subpalette contribution — gated on a DIFFERENT lookup — is unaffected) — until the chain, or an equivalent per-heritage default-clothing-Setup map, is ported. | `ClothingTable::BuildObjDesc @ 0x005A7900` (Umbraen/Penumbraen/Undead/Anakshay Setup-substitution branches); `gmCGAppearancePage::Update @ 0x0047E8F0` (clothes-button visibility gate); `tests/AcDream.Content.Tests/CharGen/ChargenAppearanceCatalogInstalledDatTests.cs` | | TS-73 | **NARROWED 2026-08-11 at Campaign OP slice OP4.** `RuntimeCharacterOptionsState.TrySetOption`'s port of `CPlayerModule::OnChanged @0x0059A8E0`'s local side-effect switch (step 2) still covers only the two `PlayerModule`-state-mutating cases (`case 2`/`case 0x12` fellowship mutual exclusion) — that part is unchanged. Of the four presentation-binding cases, TWO are now closed: `0x07 ViewCombatTarget` (re-pointed `ICombatGameplaySettingsSource` reads `RuntimeCharacterOptionsState` live — `CharacterOptionCombatSettingsSource`, `src/AcDream.App/Combat/LiveCombatAttackOperations.cs`) and `0x30 DisableDistanceFog` (`WeatherSystem.DisableDistanceFogSource`, a poll bound once in `GameWindow.cs`, forces `FogMode.Off` in `WeatherSystem.Snapshot`) — NEITHER lives inside `TrySetOption`'s own switch; both are separate App-layer poll bindings, so the literal claim in this row's title ("this Runtime-only seam can reach") stays true, but the user-observable symptom is fixed for these two ids. The remaining two, `0x04 DisableMostWeatherEffects` and `0x05 PersistentAtDay`, stay open — see TS-6 (weather-particle subsystem not yet located) and TS-75 (day/night force) respectively; this row no longer duplicates either. | `src/AcDream.Runtime/Gameplay/RuntimeCharacterState.cs` (`RuntimeCharacterOptionsState.TrySetOption`) | The remaining two options are correctly scoped to their OWN pre-existing/new rows (TS-6, TS-75) rather than re-litigated here. | Toggling `DisableMostWeatherEffects`/`PersistentAtDay` writes the bit and dirties/auto-saves it correctly, but produces NONE of retail's immediate local presentation change (weather doesn't stop, day/night doesn't force) — see TS-6/TS-75 for why. `ViewCombatTarget`/`DisableDistanceFog` are retired from this row's risk: both now behave correctly. | `CPlayerModule::OnChanged @0x0059A8E0`; `docs/research/2026-08-10-character-options-map.md` §1.5 | | TS-75 | "Always Daylight Outdoors" (`PlayerOption PersistentAtDay`, `CPlayerModule::OnChanged` case `0x05` → `LScape::SetDay(value)`) has no acdream consumer. The campaign plan's own Group-B binding table cites `RuntimeWorldEnvironmentDefinition.ForcedDayGroupIndex` as the target seam — **that citation is a mechanism mismatch, corrected here**: `ForcedDayGroupIndex` selects which WEATHER-VARIETY day-group (`RuntimeWorldDayGroupDefinition`, e.g. a Clear/Overcast/Rain/Snow/Storm pick) is always chosen — the SAME deterministic-per-day-RNG mechanism `WeatherSystem`'s own roll uses (see TS-6) — NOT retail's time-of-day day/night force. No acdream mechanism currently overrides the sky cycle's TIME to stay in daytime lighting; wiring this option correctly needs that mechanism built first, not just a poll into the wrong field. | `src/AcDream.Runtime/World/RuntimeWorldEnvironmentState.cs` (`RuntimeWorldEnvironmentDefinition.ForcedDayGroupIndex` — NOT the right target); no current consumer exists | Filed rather than silently wired to the wrong field — a poll into `ForcedDayGroupIndex` would have SILENTLY changed the character's weather-variety odds instead of forcing daytime, an incorrect fix masquerading as a correct one (CLAUDE.md's "no workarounds" rule). | Toggling the option writes the bit and dirties/auto-saves it correctly, but night still falls normally — no observable daylight-forcing behavior. | `CPlayerModule::OnChanged @0x0059A8E0` case 5; `LScape::SetDay` (not yet located in the decomp) | diff --git a/src/AcDream.App/Net/RetailSkillFormula.cs b/src/AcDream.App/Net/RetailSkillFormula.cs index 70dc47ef..18aab4d5 100644 --- a/src/AcDream.App/Net/RetailSkillFormula.cs +++ b/src/AcDream.App/Net/RetailSkillFormula.cs @@ -186,12 +186,19 @@ internal static class RetailSkillFormula /// /// Retail SkillInfoRegion::GetTooltip @ 0x004f1fe0, called once /// from SkillInfoRegion::SkillInfoRegion @ 0x004f2140's - /// UIElement::SetTooltip at 0x004f222f. Composition is exactly - /// "\n" + formula + description — retail concatenates the - /// description directly onto the formula line with NO separator between - /// them (ported verbatim, not "fixed": append_n_chars runs - /// immediately after the formula assignment with no intervening - /// literal). SkillSystem::InqSkillDescription @ 0x005c8770 reads + /// UIElement::SetTooltip at 0x004f222f. Composition is + /// formula + "\n" + description — CORRECTED 2026-08-24 at the + /// CA5 gate: the original Batch-B reading ("\n" + formula, no + /// separator) had the operator+ operand order backwards + /// (PVar3 = operator+(&local_c, ...) — local_c, the + /// InqSkillFormula output, is the LEFT operand; the "\n" literal is the + /// right), and the owner's retail-client render confirms: formula on + /// its own first line, description below (wrapping to further lines + /// when the DAT text is long). A formula-less skill (Salvaging) shows + /// the bare description with no leading break, matching retail's + /// failed-InqSkillFormula branch, which appends the description + /// to the still-empty output. + /// SkillSystem::InqSkillDescription @ 0x005c8770 reads /// SkillBase._description — the same DAT field /// already /// exposes, so no hand-transcription was needed for the ~30+ skill @@ -204,7 +211,7 @@ internal static class RetailSkillFormula string? formula = FormatFormula(skillBase.Formula); string description = skillBase.Description.Value ?? string.Empty; - string tooltip = (formula is null ? string.Empty : "\n" + formula) + description; + string tooltip = (formula is null ? string.Empty : formula + "\n") + description; return tooltip.Length == 0 ? null : tooltip; } } diff --git a/src/AcDream.App/UI/Layout/RetailTooltipPresenter.cs b/src/AcDream.App/UI/Layout/RetailTooltipPresenter.cs index d217dcaa..cef1ced8 100644 --- a/src/AcDream.App/UI/Layout/RetailTooltipPresenter.cs +++ b/src/AcDream.App/UI/Layout/RetailTooltipPresenter.cs @@ -626,15 +626,27 @@ public sealed class RetailTooltipPresenter : IDisposable ? bitmapFont.MeasureWidth : static s => s.Length * 8f; - float wrapWidth = MathF.Max(1f, _host.EffectiveCanvasSize.X); - var wrapped = UiText.WrapWords(tooltipText, measure, wrapWidth); - text.LinesProvider = () => wrapped - .Select(line => new UiText.Line(line, text.DefaultColor)) - .ToArray(); - - float measuredWidth = wrapped.Count == 0 ? 0f : wrapped.Max(measure); + // CA5-gate correction (2026-08-24, owner retail-render oracle): + // retail sizes a tooltip in TWO passes, and the second is what makes + // long tooltips multi-line. StartTooltip @0x0045DE90: + // 1. MEASURE — InqSizewMargins(..., UITS_MAX_WIDTH): wrap at the + // authored P0x3D max width, else the display width, producing + // the measured extent. + // 2. Resize the ROOT by the measured-vs-authored delta through + // ResizeTo, where the popup skin's authored max/min CLAMP. + // 3. RecalculateGlyphList — the text RE-WRAPS at its FINAL + // (possibly clamped) width. + // 4. A second ResizeTo grows the root's HEIGHT (width unchanged) + // when the re-wrapped glyph extent needs more than the text + // child's current height. + // The pre-correction port did only pass 1, so a description longer + // than the clamped popup stayed one clipped line. float lineHeight = text.DatFont?.LineHeight ?? text.Font?.LineHeight ?? 14f; - float measuredHeight = wrapped.Count * lineHeight; + + float measureWrapWidth = MathF.Max(1f, _host.EffectiveCanvasSize.X); + var measured = UiText.WrapWords(tooltipText, measure, measureWrapWidth); + float measuredWidth = measured.Count == 0 ? 0f : measured.Max(measure); + float measuredHeight = measured.Count * lineHeight; float requestedWidth = root.Width + (measuredWidth - authoredTextWidth); float requestedHeight = root.Height + (measuredHeight - authoredTextHeight); @@ -651,10 +663,40 @@ public sealed class RetailTooltipPresenter : IDisposable if (root.AuthoredResizeMinWidth is { } minWidth && requestedWidth < minWidth) requestedWidth = minWidth; + float authoredRootWidth = root.Width; + float authoredRootHeight = root.Height; root.Width = requestedWidth; root.Height = requestedHeight; - text.Width = measuredWidth; - text.Height = measuredHeight; + + // The text child follows the root's ACTUAL growth (retail's + // anchored resize) — the clamp is what makes these differ from the + // measured extents. + float textFinalWidth = MathF.Max( + 1f, authoredTextWidth + (requestedWidth - authoredRootWidth)); + float textFinalHeight = + authoredTextHeight + (requestedHeight - authoredRootHeight); + + // Pass 3: re-wrap at the final clamped width (RecalculateGlyphList). + var wrapped = UiText.WrapWords(tooltipText, measure, textFinalWidth); + text.LinesProvider = () => wrapped + .Select(line => new UiText.Line(line, text.DefaultColor)) + .ToArray(); + text.Width = wrapped.Count == 0 ? 0f : MathF.Min(textFinalWidth, wrapped.Max(measure)); + float rewrappedHeight = wrapped.Count * lineHeight; + + // Pass 4: grow the root's height by the re-wrap's overflow beyond + // the text child's post-resize height (width unchanged), through + // the same ResizeTo clamps. + if (rewrappedHeight > textFinalHeight) + { + float grownHeight = root.Height + (rewrappedHeight - textFinalHeight); + if (root.AuthoredResizeMaxHeight is { } maxH2 && grownHeight > maxH2) + grownHeight = maxH2; + if (root.AuthoredResizeMinHeight is { } minH2 && grownHeight < minH2) + grownHeight = minH2; + root.Height = grownHeight; + } + text.Height = rewrappedHeight; } /// Retail UIElementManager::StartTooltip @0x00459700: the diff --git a/tests/AcDream.App.Tests/Net/RetailSkillFormulaTests.cs b/tests/AcDream.App.Tests/Net/RetailSkillFormulaTests.cs index 66b10dc1..1e1e3547 100644 --- a/tests/AcDream.App.Tests/Net/RetailSkillFormulaTests.cs +++ b/tests/AcDream.App.Tests/Net/RetailSkillFormulaTests.cs @@ -143,6 +143,48 @@ public sealed class RetailSkillFormulaTests Assert.Equal(0u, resolver.Resolve(0x999u, new Dictionary())); } + [Fact] + public void BuildTooltip_FormulaFirstThenNewlineThenDescription() + { + // CA5 gate correction (2026-08-24): retail renders the formula on + // its own FIRST line with the description below — GetTooltip + // @0x004f1fe0's operator+ has the InqSkillFormula output as the + // LEFT operand (formula + a newline), then appends the description; + // confirmed against the owner's retail-client render. The original + // reading (newline-first + formula, glued description) produced a leading + // blank line and a single glued line. + SkillFormula formula = Formula(w: 0, x: 1, y: 1, z: 2); + formula.Attribute1 = DatReaderWriter.Enums.AttributeId.Strength; + formula.Attribute2 = DatReaderWriter.Enums.AttributeId.Coordination; + var skillBase = new SkillBase + { + Formula = formula, + Description = { Value = "Description text." }, + }; + + string? tooltip = RetailSkillFormula.BuildTooltip(skillBase); + + Assert.NotNull(tooltip); + Assert.False(tooltip!.StartsWith('\n')); + int split = tooltip.IndexOf('\n'); + Assert.True(split > 0); + Assert.Equal("Description text.", tooltip[(split + 1)..]); + } + + [Fact] + public void BuildTooltip_FormulaLessSkillShowsBareDescription() + { + // Retail's failed-InqSkillFormula branch (z==0) appends the + // description to a still-empty string — no leading break. + var skillBase = new SkillBase + { + Formula = Formula(w: 7, x: 1, y: 1, z: 0), + Description = { Value = "Salvage things." }, + }; + + Assert.Equal("Salvage things.", RetailSkillFormula.BuildTooltip(skillBase)); + } + private static SkillFormula Formula(int w, int x, int y, uint z) => new() { AdditiveBonus = w, diff --git a/tests/AcDream.App.Tests/UI/Layout/RetailTooltipPresenterTests.cs b/tests/AcDream.App.Tests/UI/Layout/RetailTooltipPresenterTests.cs index 12ac3c0d..4cb985a3 100644 --- a/tests/AcDream.App.Tests/UI/Layout/RetailTooltipPresenterTests.cs +++ b/tests/AcDream.App.Tests/UI/Layout/RetailTooltipPresenterTests.cs @@ -464,6 +464,69 @@ public sealed class RetailTooltipPresenterTests return target; } + [Fact] + public void LongTooltip_RewrapsAtTheClampedPopupWidth_AndGrowsHeightForTheExtraLines() + { + // CA5-gate correction: retail sizes tooltips in TWO passes + // (StartTooltip @0x0045DE90) — measure-wrap at the max width, resize + // the root through the authored clamps, then RE-WRAP at the final + // clamped width and grow the root's HEIGHT for the extra lines. + // Pre-correction, a description longer than the clamped popup stayed + // one clipped line (the owner's retail client shows three). + var (root, _, _) = CreateHarness(); + string longText = string.Join( + " ", Enumerable.Repeat("description", 40)); // far wider than 200px + var target = new HoverTarget + { + Left = 100, Top = 100, Width = 40, Height = 20, + AuthoredTooltipEnabled = true, + AuthoredTooltipText = longText, + AuthoredTooltipRootElementId = PopupRootId, + AuthoredTooltipLayoutDid = PopupLayoutDid, + }; + root.AddChild(target); + int childrenBefore = root.Children.Count; + + // Clamp the popup to 200px wide via a max-width-authored skin. + HoverAndDwellWithClampedPopup(root, maxWidth: 200); + + UiElement popup = root.Children[^1]; + Assert.True(root.Children.Count > childrenBefore, "popup did not mount"); + Assert.True(popup.Width <= 200f, $"popup width {popup.Width} escaped the clamp"); + var text = FindText(popup); + Assert.NotNull(text); + var lines = text!.LinesProvider(); + Assert.True(lines.Count >= 3, + $"expected the clamped width to force >=3 lines, got {lines.Count}"); + // Every re-wrapped line must fit the clamped popup. + Assert.All(lines, l => Assert.True(l.Text.Length * 8f <= 200f + 8f)); + // The root grew to hold the extra lines. + float lineHeight = 14f; + Assert.True(popup.Height >= lines.Count * lineHeight, + $"popup height {popup.Height} does not fit {lines.Count} lines"); + } + + private static void HoverAndDwellWithClampedPopup(UiRoot root, int maxWidth) + { + // A presenter whose popup skin authors ResizeMaxWidth, mirroring the + // parchment skins' bounded frames. + var presenter = new RetailTooltipPresenter(root, (_, _) => + { + ImportedLayout layout = BuildPopup(); + layout.Root.AuthoredResizeMaxWidth = maxWidth; + return layout; + }); + HoverAndDwell(root); + } + + private static UiText? FindText(UiElement element) + { + if (element is UiText text) return text; + foreach (UiElement child in element.Children) + if (FindText(child) is { } found) return found; + return null; + } + private static void HoverAndDwell(UiRoot root) { root.OnMouseMove(110, 110);