Sims inherited readonly media child context, 2026-11-15

Evidence: scratch/runs/20261115-sims-inherited-readonly-media-child-context/. This task follows TIMsingle1118476, then Q3capture1170070. Preparation is restricted to new tests, this document and candidates until both explicitly release source/build/runtime and their actual-wait/both-EOF/PID-group evidence is personally checked. No workers or boxes are created.

The preceding original remains permanently UNKNOWN/CLOSED, with zero input and no native Quit established. Its genuine type-library registration and child exit zero precede rejection of resident d:\maxis.ini; no completion or merge was accepted. The published readonly-cache repair preserves source provenance. Its explicit C-cwd generated control qualifies mutable transport only. It does not qualify the ordinary inherited D-cwd route.

Read-only investigation

spawnVfsChild currently writes a file-only mutable snapshot and passes the parent's actual cwd to the child. Provider-backed source files are correctly omitted. No readonly directory tree or per-drive metadata accompanies that snapshot, and the child correctly refuses an absent d:\ cwd. Adding a bare directory would conceal the lost media context.

The actual disc mount has label THESIMS_COMPLETE, serial 2782189353, drive type 5 and volume size 335261 sectors. Its provider is a ChunkCache over the real CUE data-track provider. The provider cannot be reconstructed by copying enumerable object properties; the underlying range reader is a closure. Reopening original paths alone would also miss entry replacement or adoption in the live parent filesystem.

A candidate transport will capture the actual readonly directories, provider entries and per-drive metadata separately from mutable state. It reopens pinned sources with the existing real mount plan and checks the resulting layout, without downgrading synchronous providers to asynchronous RPC. The parent's actual source ownership must fail if replaced or made writable. Profile inputs use the mount plan's ordinary pre-execution materialization while retaining borrowed provenance. Path validation, readonly writes and the existing completion guard remain strict. No original launch is qualified until the generated native parent/child inherits genuine D cwd, reads its profile and verifies real merged C-file and registry writes.

Baseline limitations retained

The predecessor documents stale media-CLI source-text assertions and an older native modal fixture naming removed $esp. Neither is claimed passing. No original outcome or inherited-D-cwd repair is claimed during preparation.

Prepared controls

The host-only generated-media candidate passes (light-cue-owner-candidate): file-only export demonstrably omits the actual d:\empty cwd, whereas pinned mount-plan replay preserves its empty directory, profile source provider, volume identity and readonly failures. Mutable exports still omit disc files; a resident untrusted D file still prevents completion. Changed source inode, same-shaped entry replacement, writable-drive transition, metadata change, unowned readonly drive, namespace collision and adversarial paths are refused. No WASM, original executable or native UI is instantiated by this control. The final host candidate uses an actual generated MODE1/2352 CUE/BIN, preserving the same closure-backed provider shape as the Sims source. Source-handle revocation is rejected even after the profile has been cached. File timestamps are transported from the parent's real entries rather than accepting fresh remount timestamps; the remounted disc's real track table and callable track provider are retained. Existing adoptFrom independently reproduces a missing source-owner record (light-adoption-baseline, expected failure). Its registry and disc-table transfer remain required integration work, not a passing claim.

The 73,728-byte generated native fixture is prepared but unrun. Its parent stages the generated child and DLL through actual guest reads/writes, then calls CreateProcess with NULL cwd. The DLL initializer must see the inherited D directory and readonly profile before the child entry point. Parent and child separately check volume label/serial/type and denied disc writes; the parent must verify child-written C bytes and a real registry DWORD after wait. This fixture is intended to distinguish early bootstrap from a late cwd patch. It also has an 84,672-byte generated MODE1/2352 BIN and CUE so native testing can take the actual disc-provider route. Generated inputs remain outside Git; the new focused test is their recovery recipe.

The additional data-only genuine Disc1 check (light-genuine-media-host) also passes: 308 source files and 37 directories remount with inherited d:\\setup, the real 16-byte profile, label THESIMS_COMPLETE, serial 2782189353, CD-ROM drive type 5 and 335261-sector size. Its BIN SHA-256 is 298dedb913aa31d71d38adc984892f07908910ed0f7ca4a30498725dc49b041a; the CUE is pinned separately in its compact log. The remounted profile retains its provider, readonly create/delete fail and mutable export writes zero disc files. This check executes neither WASM nor a PE and remains host-data evidence, not native qualification or an original launch. The independent deadline and actual process/pipe waits are retained alongside the log.

Native reproduction and repair

After both explicit predecessor releases and personal hashed wait/PID-group verification, the unchanged native route reproduced the missing d:\\empty directory (native-inherited-d-seeded-baseline). The parent passed NULL cwd to CreateProcess. The child loaded the real generated guard DLL but failed later cwd validation; its parent wait required deadline shutdown. This remains a failed baseline, with native, watchdog, supervisor and outer waits retained. One earlier wrapper attempt failed before runtime loading because its ownership environment contained a filename instead of JSON; it is not native evidence.

lib/readonly-media-context.js now records the actual mount recipe, source file identity/hash and live provider owners, real directory/file layout, timestamps and media metadata. A native child receives this separate private, launch-token-bound context. It reopens and validates the same pinned media via the existing mount plan before DLL initialization, then validates the inherited readonly cwd against the restored tree. VirtualFS.adoptFrom carries source ownership and the real disc table with the existing shared provider entries. Unowned, revoked, replaced, writable or conflicting contexts fail closed. Neither mutable export nor completion/merge path guards gain an exception.

The identical generated seeded route passes after repair (native-inherited-d-repaired): its real guard DLL initializer sees D cwd and reads the readonly profile; parent and child verify label, serial, drive type and denied disc writes. Child exit is 0. The existing guarded completion merges one real C file and 78 registry/INI keys, then the native parent verifies the file bytes and DWORD and exits 0. This is inherited D-cwd qualification, with NULL child cwd, rather than the earlier explicit C-cwd control. Inputs are recovered with the new test's --prepare-native command; use its generated setup.exe, control.cue, --media-exe=D:\\SETUP.EXE and explicit generated guard.dll seed with ordinary test/run.js --spawn-processes --real-ticks. --assert-native-log checks the retained actual run, including both guest exits and the completed merge. The test's ordinary host run now checks adoption too.

Two additional existing harness failures reproduce against unchanged committed HEAD and remain separate: test-cli-child-state-completion.js omits the pre-existing VFS_MOUNT mock, and test-shell-cli-routing.js expects shell status 33 but obtains 0. Neither is reported passing or repaired by this task.

Both required full flock builds pass. The shared module is eff47c5b9342ec18b157cd99c1c870977c6e6cb1fafc6589c19c1b3a7170e858; the committed-HEAD-plus-whole-owned module is 8340c85fd34df408001f1c05502672613ab4103974368801f6945de8abe62d1b. They differ because the shared build includes preserved foreign WAT WIP. The inherited-D native control passes on both; the final native qualification uses the HEAD module, with 138 unique WAT source pins verified. An initial 180-second HEAD-build watchdog expired while gates were progressing; the 600-second bounded run passed. This is not recorded as a source/build failure.

Focused adoption, lazy VFS, core/export, readonly-profile, media autorun, explicit child mounts and native spawn regression pass. Installer controller policy, HTTP transport and bootstrap tests pass. A production controller run during concurrent checks correctly refused input when its independent UP helper did not arm; the serial production rerun passes all 13 cases. The original must use the qualified HEAD module, the unchanged source/host-map Worker preloads, and new exact input pins. At this checkpoint no new original has launched.

Published repair and one authorized original

Repair 8f2c2cb4f98473f86797e82fd0c34ff76b77e21c is published on main. A fresh native run using only published source again qualifies inherited D cwd. The exact two Worker preloads pass the actual authority round trip; either missing preload is rejected. The original launch pins 3173 existing inputs, including Setup SHA-256 fe95a9fc8b2cdb5add76fbd326b1a11801eaa43c7d908f20cbdf413fd4d8dfde and the genuine 614532-byte IKernel. It uses the qualified HEAD module, real ticks, default settings, the current native partial overlay, a 900-second native deadline, independent 920-second watchdog and 180-second Quit reserve.

Exactly one newly authorized original ran. The complete current 800x600 scene was personally readable through an explicit image/png item: the branded Sims backdrop, InstallShield Wizard, and the front error "The InstallShield Engine (iKernel.exe) could not be launched. (0x80040154)". Identical PNG items sometimes rendered flat; this is not a native-blank claim. Selected unchanged image installer-closefresh.png has SHA-256 b368d52c088daae3bd9c5410189d1a60e79bd059c04e8c48bf23b162a2ffae33. The first review expired before binding, so it caused no bind, intent or input. A separately captured, personally reviewed fresh scene supported one permanent binding and one 120 ms physical pointer gesture on the visible error-dialog X. Its fresh nonce was CONSUMED, independent UP was ACKed, and the release helper has its actual wait and both EOFs. No gameplay or installation success follows from those receipts.

After the X gesture, the native parent traps on unimplemented CoFreeAllLibraries and exits 1. Native, watchdog and supervisor actual waits and both EOFs are recorded. This is the next concrete cleanup failure, not successful native Cancel/Quit. The original parent also reported class-not- registered 80040154; its IKernel child was launched with inherited d:\\setup, but no completed child registration/merge was observed before closure. Thus generated native D-context qualification is established, while successful original installation and original child registration remain unproved. The earlier original's LoadTypeLib/RegisterTypeLib/TreatAs success is not borrowed as evidence for this run.

Actual consumed parent/child source hashes match the published repair, and their nonempty main/Worker host maps verify. Post-run input readback finds only the runtime-rewritten overlay index changed; pinned media, DLLs and IKernel bytes remain unchanged. The overlay contains ordinary engine directories, temporary-file whiteout and retained IKernel file records; this is a filesystem effect, not an installation-completion claim.

One closure limitation is permanent: the invoking launcher ended with observed tool exit 143 before writing actual-outer-wait.json. The outer process had recorded its actual supervisor-child wait, but its own parent wait receipt is missing. All recorded native/child/watchdog/supervisor/outer PIDs and groups are personally absent; absence does not reconstruct the missing wait. This is written UNKNOWN for that closure receipt. No subsequent control requests, input, rebind, replay or second original are permitted. The prior UNKNOWN original also stays closed. The run retains the known native exit separately from this outer-wait limitation. No boxes were created.

Next work is the real IKernel child bootstrap/registration boundary and genuine CoFreeAllLibraries cleanup behavior. A future original requires new explicit authorization; this task's single original authorization is consumed.