Remove the unused AssetConfig from EngineState

[coghex]

Background

Engine.Asset.Types defines an AssetConfig record with four settings:

• acMaxTextureAtlases
• acMaxShaderPrograms
• acPreloadAssets
• acEnableHotReload

defaultEngineState constructs it positionally as:

AssetConfig 100 100 True True

and stores it in EngineState.assetConfig.

A complete search across src/, app/, test/, and test-headless/ finds no
read or update of assetConfig or any of its four selectors. The declarations
and default construction are their only occurrences. Git history likewise shows
introduction and relocation of the fields, but no implementation of the
advertised limits, preloading, or hot reload.

The live renderer uses separate mechanisms such as
BindlessConfig.bcMaxTextures, FontCache, and AssetPool; none consult
AssetConfig. The texture-infrastructure design document still describes hot
reload as an undecided future workflow rather than an implemented setting.

The dead record therefore advertises capabilities that do not exist, and its
positional construction makes the four unexplained literals particularly
misleading.

assetConfig is also recorded in two authoritative state documents:

• docs/persistence_state_inventory.md classifies it as excluded and rebuilt at
  boot.
• docs/engineenv_capability_inventory.md names it as part of
  main-render-private EngineState.

Those inventories must remain synchronized with the live record.

Requirements

1. Remove AssetConfig and its four field selectors from
   Engine.Asset.Types.
2. Remove EngineState.assetConfig and its default construction.
3. Remove imports or other source references made unnecessary by that deletion.
4. Update docs/persistence_state_inventory.md so its EngineState inventory
   exactly matches the remaining live fields.
5. Update docs/engineenv_capability_inventory.md so its descriptions of
   EngineState no longer name AssetConfig.
6. Preserve the existing main-render-thread ownership invariant for
   EngineState.
7. Preserve all live asset loading, texture limits, font caching, shader
   handling, rendering, and scene-management behavior.
8. Do not change save data or increment the save version; this field is
   currently excluded from persistence and has never carried runtime state.

Acceptance

• No declaration, construction, inventory row, or source reference remains:

! rg -n \
  '\b(AssetConfig|assetConfig|acMaxTextureAtlases|acMaxShaderPrograms|acPreloadAssets|acEnableHotReload)\b' \
  src app test test-headless \
  docs/persistence_state_inventory.md \
  docs/engineenv_capability_inventory.md

• Production and both test targets compile warning-clean:

cabal build lib:synarchy exe:synarchy \
  synarchy-test-headless synarchy-test-graphical \
  --ghc-options=-Werror

• Both affected inventory contracts remain valid:

python3 tools/persistence_inventory_audit.py
python3 tools/engine_env_capability_audit.py

• The patch contains no whitespace errors:

git diff --check

No behavioral test is required for deleting fields with no readers.

Out of scope

• Implementing atlas or shader-program caps, asset preloading, filesystem
  watching, or hot reload.
• Changing BindlessConfig, AssetPool, FontCache, ShaderProgram, or the
  live texture/font pipelines.
• Removing other unused asset types or registries.
• Changing the EngineEnv field set, RenderCapability, or the
  engineStateRef capability boundary.
• Changing persistence behavior or the save format.
• Folding in Remove unused legacy fields from EngineConfig #932's separate EngineConfig cleanup.

Related

• Remove unused legacy fields from EngineConfig #932 removes different legacy fields from EngineConfig; it explicitly
  preserves EngineState and does not own this work.
• [epic] Split EngineEnv into capability-scoped surfaces #537 and its completed child EngineEnv capability split E3: migrate render/window/Vulkan/asset consumers to a RenderCapability record #891 govern access to the containing
  engineStateRef; removing one unused nested field preserves that capability
  boundary.
• docs/texture_infrastructure.md treats development hot reload as a future
  decision, not an existing engine capability.

[coghex]

Automated cross-agent issue review

Reviewed against master@f116c8899f51 by Claude Fable 5.

Verdict: APPROVE

Corrections and spec additions below amend the implementation spec. Supporting context is non-normative. Open decisions remain unresolved and must not be guessed.

Spec additions / clarifications

• Claude Fable 5: Import fates under requirement 3, verified against the snapshot: Engine.Core.Defaults's import Engine.Asset.Types (src/Engine/Core/Defaults.hs:9) uses no other name from that module and becomes fully unused after deleting the construction at line 104 — it must be removed, and the acceptance build enforces this (-Wall at synarchy.cabal:77 plus --ghc-options=-Werror makes an unused import fatal). Conversely, Engine.Core.State's import Engine.Asset.Types (src/Engine/Core/State.hs:12) must REMAIN — assetPoolRef ∷ IORef AssetPool (src/Engine/Core/State.hs:122) still depends on it.
• Claude Fable 5: While editing docs/engineenv_capability_inventory.md under requirement 5, note both passages to touch cite stale line numbers — src/Engine/Core/State.hs:389 (§1, line 47) and :389-394 (§3, line 185) — while the EngineState record actually sits at State.hs:404-409 in the current snapshot, and will shift again after the field deletion. Refresh or de-precise those citations in the same pass; no audit enforces them, so this is hygiene, not a gate.
• Claude Fable 5: Optional, outside the acceptance rg by design: docs/code_health_findings.md's status legend (line 8: [ ] not filed · [#N] filed as issue N, e.g. ### [#931] CH-1. at line 34) means the origin finding ### CH-18. (line 244) should gain a [#952] marker. The acceptance rg correctly excludes that file, so its remaining AssetConfig mentions are intentional and must not be deleted.

Supporting context

• Claude Fable 5: Zero-reader claim verified by repo-wide grep of all six identifiers (AssetConfig, assetConfig, and the four ac* selectors): the only hits are the declaration (src/Engine/Asset/Types.hs:63-68, deriving (Show) only), the field (src/Engine/Core/State.hs:407), the construction (src/Engine/Core/Defaults.hs:104), the two inventory docs, and docs/code_health_findings.md CH-18 — including docs/history/, so no hidden consumer exists.
• Claude Fable 5: Single-construction claim verified: defaultEngineState (Defaults.hs:58-106) is the only EngineState construction; Engine.Core.Init.hs:263 merely does newIORef defaultEngineState; a grep of every EngineState occurrence in *.hs shows only type signatures and EngineState(..) imports — no positional constructor application or pattern anywhere, so the field deletion touches exactly the three source sites.
• Claude Fable 5: tools/persistence_inventory_audit.py lists EngineState in ROOT_RECORDS (line 56, parsing data EngineState = EngineState from src/Engine/Core/State.hs) and its check loop (lines 1540-1550) flags only live fields lacking a doc row — a stale doc row for a removed field does not fail the audit, so requirement 4's row deletion is enforced by the issue's rg acceptance, not the audit; the pairing is consistent.
• Claude Fable 5: tools/engine_env_capability_audit.py contains no reference to assetConfig/AssetConfig; its only EngineState mentions are comments and an error string about modules naming engineStateRef/rcEngineStateRef (lines 644-763), so requirement 5's prose edit in the capability inventory §3 cannot trip the audit or its self-test.
• Claude Fable 5: Acceptance commands are executable as written: test-headless/ exists (150 files); synarchy-test-graphical (synarchy.cabal:934, hs-source-dirs test) and synarchy-test-headless (synarchy.cabal:980, hs-source-dirs test-headless) are real test-suite targets.
• Claude Fable 5: Requirement 8 (no save version bump) verified: docs/persistence_state_inventory.md:135 classifies assetConfig as Exclude/rebuilt-at-boot, EngineState has no deriving Serialize and is never persisted, so no save-format or currentSaveVersion impact exists.
• Claude Fable 5: The Related-section claims hold: docs/texture_infrastructure.md:244 lists hot reload under Open questions ('Default to no for simplicity'), confirming the false-capability-advertisement premise; Remove unused legacy fields from EngineConfig #932 targets EngineConfig, a different record, and this issue's out-of-scope list explicitly excludes folding it in, so no duplicate or contradiction is visible from the dossier.

Reviewer summaries

• Claude Fable 5 — APPROVE: The issue is a fully verified, self-contained one-PR dead-code deletion. Every factual claim checks out against the snapshot: AssetConfig (src/Engine/Asset/Types.hs:63-68) and EngineState.assetConfig (src/Engine/Core/State.hs:407) have exactly one construction site (AssetConfig 100 100 True True at src/Engine/Core/Defaults.hs:104, inside the sole EngineState construction defaultEngineState) and zero readers anywhere in src/, app/, test/, or test-headless/. The two inventory docs reference it exactly where the issue says (persistence_state_inventory.md:135; engineenv_capability_inventory.md:50 and :188), both audit tools tolerate the removal (persistence audit parses live fields and requires rows, never the reverse; capability audit has no hard-coded assetConfig reference), and all acceptance commands name real paths and real cabal targets. Requirements are implementation-neutral, the out-of-scope list cleanly fences off Remove unused legacy fields from EngineConfig #932 and the live pipelines, and no open decisions remain. Approving with three non-blocking spec additions clarifying import fates and adjacent doc hygiene.