Skip to content

RC4.5 tester-package qualification

Historical record. This preserves the source, tests and limitations recorded at the time; it is not current setup guidance. See the documentation index.

Local date: 2026-09-08; build work continues into 2026-09-09 UTC.

This records the initial 12b2485 RC4.5 checkpoint, before normal portable broker auto-start was added. Its package hashes and counts below are historical. The portable startup follow-up records the replacement packages and their qualification. The current distribution's SHA256SUMS.txt identifies the files to distribute.

Candidate identity

  • Release/channel marker: VIIPERRC4.5.
  • DS4Windows assembly, file, MSI and bundle version: 5.0.5.0.
  • Package version: 5.0.5.0-rc4.5.
  • Bundled VIIPER: v0.1.3-rc4.5, source 6205074f209e488c237d7eb475383f49d7f3cab9, PE 0.1.3.0.
  • VIIPER SHA-256: F4A86C6D00CDD30FACF4F7E2C0FB3C69761576119514678D055E65EA28CDBB95.
  • USB/IP remains the pinned 0.9.7.7 package and signed driver files.

This is an unsigned tester candidate composed through the existing local build path. The public workflow's required signing policy is unchanged. Automated qualification does not substitute for a disposable-machine install/upgrade/ repair/uninstall pass or the remaining physical-controller acceptance matrix.

Update and registration guards

The previous update policy offered any different prerelease to an installed prerelease. A locally installed RC4.5 could therefore be offered the older public RC4.3. Named VIIPER Beta/RC releases now compare phase and numeric ordinal, separately from the application's numeric Windows file version.

  • Initial regression run: 12 failed, 17 passed, proving the rollback cases.
  • First correction: all 29 release-policy cases passed.
  • Independent review identified a second rollback after a future transition from a named RC to a numeric prerelease. Added cases reproduced 10 failures among 41 policy cases, including partial numeric-tag parsing.
  • Final focused run: 54 passed, zero failed/skipped, covering all 41 policy cases, six new installer registration cases and seven existing helper cases. Evidence: rc45-upgrade-and-update-final.trx.

Only genuinely unmarked legacy prereleases retain the one-time bootstrap rule. Unknown marked prereleases do not receive guessed automatic ordering. Numeric prereleases require a strictly newer binary base; different suffixes at one base remain manual. Stable promotion at the same normalized binary version remains permitted. Selection is still publication-date based: republishing an older release may suppress a newer offer, but must not cause a downgrade.

The registration audit ties the duplicate-entry causes to actual install logs and production helper planning. No installer upgrade-family GUID was changed and no broad automatic registry cleaner was introduced. A separate, authorized machine repair reversibly hid 18 obsolete entries after backups; one current visible entry remains. It did not uninstall or delete the old cache, alter installed application bytes, or restart either running application.

Broker source and license provenance

The broker was rebuilt from a clean checkpoint with the release build tag, code generation excluded, and embedded VCS state modified=false. A repeat build produced the same executable hash. Version/help, the release Go suite, targeted command/tray race tests and changed-line lint passed. Existing unrelated format findings were not silently suppressed.

The previously unlicensed kong-yaml v0.2.0 dependency was explicitly advanced to v0.2.1-0.20240409173824-a56dc1466c79, whose source includes the upstream MIT grant. The existing nested YAML behavior is preserved with a resolver adapter and four regression cases. The local tray fork's complete Apache license and modification notice, the broker's GPL text, aggregate dependency notices and unabridged build notes are explicit package assets. Matching broker source is distributed alongside the installer. This closes that identified notice/pin gap; it is not a blanket legal certification.

The strict provenance format retains its original 17 key/value fields plus heading. Independently pinned hashes cover the full build notes and license documents. Narrow Git -text attributes preserve those byte-hashed artifacts across Windows/Linux checkouts. Four Python packaging fixtures passed, including byte preservation, managed-file ownership, ZIP inclusion and WiX harvesting of all required broker notice files.

Staging confirmed all six byte-pinned artifacts (binary and five documents) match their raw working-tree Git blob identities. The two original CRLF license documents have file-specific whitespace attributes; the aggregate's original trailing blank lines are retained. Neither document was reformatted to satisfy source whitespace checks or silently given a new release hash.

Final package acceptance

  • rc45-complete.trx: final Release/x64 source rebuilt, 4,463 passed, zero failures, 11 existing opt-in skips (4,474 total).
  • rc45-repeat.trx: separate repeat, 4,463 passed, zero failures, the same 11 skips.
  • rc45-allocation.trx: all 152 allocation-named cases passed, no failures or skips. No allocation assertion was disabled or loosened.
  • rc45-xbox-process-interop.trx: all eight opt-in process interop cases passed, no failures/skips, with all eight success markers. The separately compiled, release-tagged Go test peer came from clean broker source 6205074f209e488c237d7eb475383f49d7f3cab9. It exercised real authentication, input/feedback streaming and retirement with ephemeral loopback listeners, synthetic keys and fake native attachment. No hardware/driver was contacted; every test peer exited. This establishes source-level protocol compatibility, not physical input or a server startup test of the packaged product EXE. The peer and synthetic key are confined to private _build test staging.
  • Both full runs retain the three existing CI exclusions (CheckSettingsSave, CheckWriteProfile, CheckJaysProfileRead); no new exclusion was added.
  • Self-contained Windows x64 publish succeeded in a new, empty Desktop staging directory. Published PE versions and channel marker match the identity above.

  • The published file inventory contained no reparse points, user profiles, controller stores, logs, lab-data, dumps, traces or private key files. The portable ZIP was composed from that new publish, not a running lab directory.

  • WiX MSI and Burn builds succeeded with zero warnings/errors. Default ICE validation remained enabled, with the pre-existing ICE61 exception unchanged.

  • The final bundle was extracted and its MSI, setup-helper and bootstrapper payload hashes verified by the build gate. Required payload, provenance, localization and release-marker validation passed before atomic publication.
  • USB/IP reboot-boundary simulation, exact-SID startup-task ownership/rollback simulation, and installer transaction-state simulation passed. These are simulations, not a real installation or reboot of the build host.
  • Two resource-only child probes against packaged assemblies passed from both package and outside working directories: localized and fallback strings, TaskScheduler satellites and rejection of an unrelated assembly. They did not initialize the application or controller service.
  • Independent audit checked all 547 published files; the installer manifest matches the other 546 files by size and SHA-256. Every one of the portable ZIP's 546 files matches the fresh publish byte-for-byte (the additional package-manifest.json was created later for the installer). No private profiles, authentication keys, captures, test peers or lab clients ship.
  • The final installer is 200,191,091 bytes, SHA-256 982023115E9C5D11D0391D9EFB631D9818617C519B18909457A465AFB74F0228. Authenticode explicitly reports NotSigned.
  • The portable ZIP is 132,192,631 bytes, SHA-256 77F2D07A90ED3F54757F7E6C9D33398D17012940BA0B5766E5387CA44DBA3480.

The distribution folder includes both matching source archives, preserved license notices, this qualification record and the cumulative RC assessment. SOURCE-REVISIONS.txt and SHA256SUMS.txt identify the final source checkpoint and exact delivered artifacts. No public tag, push or release was performed.

Earlier source-checkpoint results in the controller rollup are historical evidence, distinct from the new runs recorded here.

Remaining acceptance boundaries

The live portable b93 session was not replaced for packaging. No actual MSI transaction or new controller/game acceptance is claimed on this host in this release-preparation pass. Bluetooth headset audio remains unsupported; the native-driver backend and sub-millisecond end-to-end latency are not shipping claims. See the RC4.5 assessment for the full feature scope, demonstrated hardware evidence and focused tester scenarios.