RELEASE_NOTES.md

Release-ready branch

DEVolf 0.3 — EmWolf SSH controls and compact preview (2026-10-29)

• Add independent Start EmWolf and Stop EmWolf controls between Port and Connect. Start uses the Windows OpenSSH client to run /home/FC22/EmWolf as root on the selected Host; Stop sends terminal Ctrl+C in that session. Stop test retains its separate role. Existing EmWolf processes are detected before startup.
• Remove the explanatory Run log and Report preview text while preserving the result table and PDF data. Move file actions to a second connection row and verify the layout at 1366 x 768.
• Include autonomous EmWolf runs and stored result retrieval, Python ports of WolfEye algorithms, and x64/x86 build support added since 0.2. This release supplies the latest existing Windows x64 build.
• Validation: 226 of 229 tests passed. The three logging-format test failures also reproduce before the SSH/UI changes. All 63 focused EmWolf, tester and responsive-layout checks passed; packaged startup passed. SSH login and Ctrl+C were checked using a harmless shell loop on the board; the existing running EmWolf process was preserved and duplicate startup was refused.

Build updates and repository cleanup (2026-10-21)

• Add independent -UpdateWolfEye and -UpdateBeachEye switches to full and native builds. Both exact selected commits are recorded in their manifests.
• Use -UpdateSource in the dependency resolver; retain its old -UpdateBeachEye alias for existing scripts.
• Validate WolfEye update/pin transactions and require its signal-loader source.
• Move wire-contract tests into the normal tests/ suite so full builds run them.
• Exercise visible settings through the same typed editor API as the application, including exact byte preservation on both models; retain separate legacy JSON API coverage. The latest company JSON export represents negative admittance words as unsigned values, so it is not the visible editor snapshot.
• Consolidate current signal/numeric guidance and remove three superseded audit documents plus the duplicate signal diagnostic note. Application sources, EmWolf, inspection data and scenario files are retained.
• Full update build verified with BeachEye 0b5b4cce5755a5dba3db1b27413434fb0b6d6c7e and WolfEye master 7e58f4af439c6fa9c9db51e4558936264401732a: all 192 tests passed in 136.852 seconds; the x64 EXE passed packaged startup.

DEVolf 0.2 — native WolfEye core and signal fixes (2026-10-21)

• Separate WolfEye company algorithms into wolfeye_core.dll (ABI 1), with BeachEye device/settings integration in wolfeye_beacheye.dll (ABI 12). Both libraries are bundled in the Windows x64 executable.
• Execute the original company TCC calculator, pure AR stage transitions and SIGNAL/ADC loader from pinned WolfEye master 7e58f4a. BeachEye remains pinned to bee19ba on ver_50_remaster.
• Support INFO-selected 32/64 samples, including fixed 2560/3072-byte buffers from 32-sample devices. Keep all ADC cycles and use the correct time axis.
• Save separate raw SIGNAL/ADC diagnostics for hardware investigation.
• Preserve the Qt appearance, existing EmWolf/EWS2 source and user signal analysis, history, navigation and standalone viewers.
• Add -WolfEyeSourceDirectory for editable company-source development, with a separate build directory and unchanged release pins.
• Include the existing FC22AR scenario and documentation updates in this tree.
• Validation: 181 project tests and five additional wire-contract tests passed; the packaged executable passed startup and both embedded DLL hash checks. Physical relay validation and complete line-by-line parity with master are not claimed. The transport remains the existing native adapter.

The dated entries below describe earlier builds and their earlier ABI versions.

Company 32/64 signal sampling (2026-10-20)

• Pin WolfEye master to 7e58f4af439c6fa9c9db51e4558936264401732a. BeachEye remains at the already selected bee19ba2851658f03601375aec458a21633055bb.
• Read INFO flag 0x02 as the company's 32-sample mode; clear means 64. Decode SIGNAL/ADC using that stride, with 1600/3200 Sa/s for calibrated signals.
• Keep each capture's cycle size for history, plot timing, fault origin and analysis windows. Existing analysis calculations and nonoverlapping steps remain.
• Invoke the unchanged company ADC loader per cycle so multiple cycles are retained without the upstream compile-time stride or first-cycle overwrite.
• Bridge ABI 11 rejects older DLLs. Rebuild DLL and EXE together. Cover both models, flag transitions, all channels, malformed payloads, ADC cycles and mixed capture history in tests/test_signal_sampling.py. Hardware verification still requires a connected relay.
• Verification: all 174 root tests passed in 96.206 seconds, plus all 5 existing code/tests signal tests. build.ps1 rebuilt the x64 native DLL and Dist/DEVolf.exe; the packaged Qt startup smoke test passed.

Company core process audit (2026-10-17)

• Preserve EmWolf, FC22 Tester and the user's analysis/report integrations.
• Pin pristine WolfEye master alongside pristine BeachEye; bridge ABI 10.
• Execute original company TCC calculation, AR stage rules and raw ADC loader.
• Bind settings directly to Configure so JSON export cannot clamp or omit UI values.
• Display company DNP3 measurement indices and advance transport sequence only after a valid matching response; resend CRC failures with the same sequence.
• Document remaining adapter/UI differences in the parity audit. No new test methods were added.

Unmodified company DLL and transport cleanup (2026-10-17)

• Removed Legacy TCP, Legacy HTTP, discovery and automatic fallback from both the application and the native provider. Relay requests always use BeachEye's sequenced protocol. Failed INFO handshakes are reported and closed.
• Removed all company-source rewrites from dependency resolution. A separate pristine checkout of majidnow/BeachEye ver_50_remaster is required and checked before and after compilation.
• Removed custom legacy settings, measurement, event and signal interpretations, extra settings JSON fields and byte-difference patching. Settings JSON and wire bytes now use the unmodified company implementation.
• Bridge ABI is 9; rebuild the DLL and EXE together. Earlier entries below document superseded behavior. No new tests were added.

Remove extra frame and measurement restrictions (2026-10-15)

• General frame creation and queued requests no longer reject data at the firmware file-chunk size. Both directions use the two-byte wire length, with the sequence byte included for native requests. File upload partitioning remains aligned with the company source.
• Complete Measurements blocks with different point counts use wire-index labels instead of failing a model/build profile check. Empty channels and repeated blocks retain all received values. Known profiles retain company names. Status cards remain unavailable when only indexed readings exist, rather than displaying an invented zero or OFF state for an unidentified point.
• Event reception retains the deployed 5112-byte CC16 correction, current CC32 support, dynamic frame assembly, CRC and sequence checks. Truncated or corrupt data remains distinguishable from a valid larger response.
• Event file downloads use the reported file length without the former desktop 1 MiB rejection, and chunk sequences wrap within their uint16 wire field.

FC22 interface consolidation (2026-10-15)

• Four fixed workflow steps, grouped form inputs and a single editable test table replace the crowded results sidebar. Sources, verdicts and rule tooltips stay next to observations; the complete diagnostic view is optional.
• Setup retains the WI03 revision 00 procedure, existing DUT autofill and shared EmWolf board connection. Custom Test Lot JSON and equipment adapters remain available in expandable settings. Reports expose optional approval fields and enable Open/Print after a PDF exists.
• This release is built from the current devolf tree, retaining ABI 8 event decoding and full 16-bit receive lengths. The 5112-byte short-counter stream, fragmented 5512-byte socket response and frame sizes through 65535 bytes remain covered by the native regressions.

This branch contains the cleaned, release-oriented DEVolf product tree. It is based on the wolfeye branch at commit 1718877, but removes the repository's nested legacy layouts and keeps one canonical application source.

Product features

• Synced with company WolfEye master at 989f9ff and pinned BeachEye ver_50_remaster at 222c24e; see the synchronization manifest for the commit map and previous pins.

• User-facing product identity and the default executable name are DEVolf; compatibility-facing WolfEye/BeachEye package, DLL, and schema names remain unchanged.

• BeachEye is now resolved exclusively from an authenticated private GitHub repository into an external cache. The bundled mirror, generated DLL, and source-diff patches are no longer tracked.

• Complete UV/OV stage and voltage auto-reclose settings added to the native mapping and FC22AR protection editor.

• Independent digital-output recognition time added beside pulse duration.

• Fast-trip TCC max-current mode/multiplier and min/max-bounded curve plotting.

• Binary, analog, and 32-bit counter event decoding using the current BeachEye stream API from the same selected ver_50_remaster dependency for every device build.

• Native BeachEye/DataProvider TCP transport with automatic one-time fallback to the legacy protocol when the native handshake fails.

• FC22AR and FC22S device-specific protection menus.

• Shared setting synchronization across device views.

• TCC element editor and curve chart using the company IEC/ANSI formulas.

• Settings read/write support for both the current 3,439-byte layout and the deployed pre-menu-split 3,243-byte revision-50 layout.

• Byte-exact settings round trips so unchanged payloads remain unchanged.

• Write completion based on the company's status response (0x00 or 0x09), without a second read that can race or report a false failure.

• Company-source measurement/event decoding: FC22AR build 51 uses the ver_50_remaster 410-BI/122-AI packed profile (300 bytes), with serializer declaration order and no automatic ver_51_compatible DLL switch.

• FC22AR and FC22S share the same company branch for builds 50, 51 and later. INFO retains its 16-bit build; the company's 8-bit revision argument saturates instead of wrapping at 256. Simulated read/write and measurement coverage includes builds 50, 51, 52, 100, 255, 256, 257, 306 and 65535 for both models. Future compatibility follows the current branch's >50 map and wire format.

• Alternate source/revision build options and old branch API paths removed. Fourteen unused native source/header files removed; six required company CRC/JSON/configuration files remain. The 60-entry mapping records removals and the active implementations replacing them.

• Signal analysis, event handling, firmware/bootloader transfer, EmWolf test scenarios, FC22 tester sessions, and PDF reports.

• One 64-bit Windows package using Python 3.11, PySide6, and PyInstaller 6. The Python 3.8/MinGW legacy build mode has been removed.

Event reception fix (2026-10-13)

• Reported invalid 5512 bytes failure covered with ring-decoder and socket regressions. Receive frames accept the full 16-bit length with CRC/sequence validation; outgoing frames at that stage retained the former 5 KiB policy (superseded by the frame restriction removal above).
• Ring counter events, full/wrapped boundaries and empty slots corrected. Truncated AI/CC records fail explicitly. Serialized events use the same pinned company implementation and point maps.
• Event downloads guard repeated clicks, retry interrupted requests/chunks twice, validate sequences/file sizes and release busy state on every exit. Invalid data and device metadata are saved under the local DEVolf diagnostics folder. Event filters reapply on reload and timestamps show milliseconds.
• Native bridge ABI is now 7; rebuild the DLL/application before running. Tests use synthetic responses; the original physical-device response has not yet been captured. See event verification details.

Real-device Counter compatibility (2026-10-13)

• Actual AR51 capture (5,112 bytes) now decodes to 627 visible events, including all four Counters. Company history confirms the older serializer advanced by 10 bytes for a 12-byte Counter, leaving only its low 16 bits in the stream.
• Event wire sizes come from the pinned company API. The complete stream is validated before selecting CC16/CC32; packet size, INFO model and build do not choose its layout. Current 32-bit values stay intact and transmitted widths are retained. Short Counter values are identified in viewer tooltips.
• ABI 8 requires a native/application rebuild. The real capture passed both models and native/chunked service paths. Source selection remains the same pinned ver_50_remaster; no alternate dependency branch or DLL was added.

Connection reliability (2026-10-14)

• Native TCP connects directly and obtains identity through company INFO; optional HTTP identity is only probed for the legacy fallback. Unavailable or malformed HTTP metadata no longer blocks the native handshake.
• Stopping during HTTP discovery or TCP establishment prevents a late successful connection from being published or used for requests.
• Windows nonblocking connect checks both writable and exception socket sets and reports the actual endpoint and Winsock error. The Python bridge retains that detail instead of replacing it with a generic failure message.
• Regression coverage includes a complete AR51 INFO/settings handshake over localhost TCP, cancellation, text-valued legacy identity and failed socket error reporting. Physical device reachability requires its actual IP/error; this software validation does not establish the current device's availability.

Updating company source during builds (2026-10-14)

• Added -UpdateBeachEye to release builds, native-only builds and source resolution. It selects the latest ver_50_remaster commit once and records it after preparing the private checkout. Normal builds repeat the pin.
• Updated the company pin to 18d5e16. SEF defaults now match 10 seconds for both models. New upstream ROCOF struct attributes do not change serialized Settings or JSON lists, so no new wire settings are assumed.
• Added guidebuild.txt with English build commands and source-update, runtime, output-name, wheelhouse and toolchain/cache options.

Differences from wolfeye

Area	wolfeye branch	This branch
Repository layout	Multiple app copies, including a nested wolfeye/ tree	One canonical product tree at repository root
Connection mode	Manual native/legacy selection	Native-first connection with one automatic legacy fallback
Protection UI	Older combined configuration presentation	Separate FC22AR/FC22S menus with synchronized shared fields
TCC workflow	Basic element editing	Dedicated editor and chart with company curve calculations
Legacy settings	Current 3,439-byte layout only	Current and deployed 3,243-byte layouts with preserved offsets
Settings write	Follow-up readback after status	Validates status directly and clears failed/pending writes safely
Build-51 measurements	Company branch point maps	Uses the selected ver_50_remaster Database with the INFO build number
Release contents	Historical and duplicated project trees	Source, tests, docs, official scenarios, and build scripts only

Verification

Run build.ps1 with Python 3.11 x64, CMake, MSYS2 UCRT64 GCC/Ninja, and authenticated access to the private BeachEye repository. The build executes the tests and performs a packaged EXE startup smoke test. The old Windows x86/Windows 7 build and its toolchain parameters were removed. The device-level protocol fallback and deployed settings compatibility remain.

The native build now uses GCC/Ninja explicitly because the pinned BeachEye headers do not compile under MSVC. The GCC cache is separate from the previous Visual Studio build directory.