nikon/Codesys-MCP-SP21-plus
0
0
Fork 0
Code Activity Actions

docs/SMOKE-TEST: collapse v1+v2 into a single end-of-day doc

Browse source 
Two parallel smoke-test files had drifted out of sync:

  docs/SMOKE-TEST-2026-04-25.md      (morning v1, 28 invocations,
                                      pre device-side fix)
  docs/SMOKE-TEST-2026-04-25-v2.md   (afternoon v2, 29 invocations,
                                      post device-side fix)

Neither covered the late-afternoon work (six new git_* tools,
mirror_export, launcher refuse-on-duplicate, list_project_libraries
rewrite). Consolidating to one doc named SMOKE-TEST-2026-04-25.md
that:

  - Covers the full 36 distinct tool invocations (28 OK / 4 FAIL /
    1 PARTIAL).
  - Reduces the open-bugs list from 5 to 4 (list_project_libraries
    is fixed; the read/write inconsistency now narrows to
    add_library's missing dedupe + placeholder).
  - Adds rows for the 7 new tools verified end-to-end on X33 (8
    library refs, 91 .st mirror files) and on GitSmokeTest (full
    init -> commit -> remote_add -> branch_set_upstream_to -> push
    round-trip against gitlab.usv.no via cached HTTPS creds).
  - Documents the new infrastructure (PDE-license rewrite, launcher
    refuse-on-duplicate) as separate sections rather than bugs.
  - Cites the late-afternoon commits in the "What this proves" table
    so the smoke test traces every fix back to its source.

The morning v1 is removed; the afternoon v2 file is renamed onto the
canonical filename to preserve git's rename history.
This commit is contained in:
Karstein Phobic Nyvold Kvistad 2026-04-25 23:32:26 +02:00
parent 0a4c1a0de8
commit 3e2467149f
2 changed files with 161 additions and 179 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

128
docs/SMOKE-TEST-2026-04-25-v2.md
View file

@ -1,128 +0,0 @@
# Smoke test v2 -- 2026-04-25 (post device-side login probe + write API fix)
End-to-end re-test of every MCP tool exposed by `codesys-mcp-persistent`
after the additional fixes landed on the [`sp21-plus-migration-notes`](https://github.com/phobicdotno/Codesys-MCP/tree/sp21-plus-migration-notes)
branch this afternoon. Supersedes the morning [SMOKE-TEST-2026-04-25.md](SMOKE-TEST-2026-04-25.md).
## Environment
- MCP server source: this fork @ `sp21-plus-migration-notes` HEAD
(commits `93a105a`..`b3bf4a8`).
- CODESYS launched: **`3.5.22.10` (SP22 Patch 1, 64-bit)** via `--runscript`
pointing at the rewritten watcher (v0.4.2 with KeyboardInterrupt hardening).
- Soft-PLC runtime up: `CODESYS Control Win V3 - x64` Windows service
(`CODESYSControlService.exe`, listening on port 11740). User-started
via the tray icon's "Start PLC" with one-shot UAC elevation.
- Test project: `\\files\karstein.kvistad\Documents\Claude\PLC\MCPTest\MCPTest.project`
-- created via `create_project` from the Standard template (device =
`PLCWinNT (CoDeSys SP Win V3)`, which IS the Control Win V3 device
descriptor on this machine despite the legacy display name).
- All test objects authored with prefix `MCPv2_` to avoid colliding with
user-fixed objects in the project; cleaned up at end of test.
## Result table
| # | Tool | Result | Notes |
|---|--------------------------------------|-----------------|----------------------------------------------------------------------------------------|
| 1 | `get_codesys_status` | OK | State / Mode / PID / Session reported correctly. |
| 2 | `launch_codesys` | OK | Persistent mode; ready signal received within seconds. |
| 3 | `shutdown_codesys` | OK | Cleanly stops the spawned CODESYS instance. Prior smoke-test #5 (orphan from earlier launches) not reproduced this session. |
| 4 | `create_project` | OK | From Standard template; saved to the requested path. |
| 5 | `open_project` | OK | Reproducible from-cold open. Cross-project switch (closing one project to open another in the same instance) not exercised here -- prior smoke noted that as a separate bug. |
| 6 | `save_project` | OK | |
| 7 | `create_pou` (FunctionBlock / ST) | OK | `Application/MCPv2_FB`. |
| 8 | `create_dut` (Structure) | OK | `Application/MCPv2_ST`. |
| 9 | `create_gvl` | OK | `Application/MCPv2_GVL` with declarationCode populated. |
|10 | `create_method` | OK | `Application/MCPv2_FB/DoSomething : BOOL`. |
|11 | `create_property` | OK | `Application/MCPv2_FB/Counter : INT` (Get/Set sub-objects auto-created). |
|12 | `set_pou_code` | OK | Decl + impl wrote correctly to MCPv2_FB and MCPv2_ST. Verified via get_all_pou_code. |
|13 | `get_all_pou_code` | OK | Returned full code dump for every POU/DUT/GVL/Method/Property in the project. |
|14 | `rename_object` | **PARTIAL** | Renames the object's own internal declaration line (`TYPE old : ... END_TYPE` becomes `TYPE new : ... END_TYPE` automatically) BUT does NOT update other POUs that referenced the old name. Same as upstream. |
|15 | `delete_object` | OK | Cleaned up MCPv2_FB / MCPv2_GVL / MCPv2_STRenamed at end-of-test. |
|16 | `get_application_state` (offline) | OK | `State: none, Logged In: False` before connect. |
|17 | `add_library` | **PARTIAL** | Operation succeeds and CODESYS shows the entry in Library Manager, BUT it adds a SECOND `Standard` reference instead of detecting the existing one, and the new reference is not added as a `* (System)` placeholder so it pulls in unresolved transitive deps (e.g. IoStandard 3.1.3.1 yellow-warning). |
|18 | `list_project_libraries` | **FAIL** | Returns "No libraries found in the project (or Library Manager not found)" both before AND after a successful `add_library` -- the read path can't find the Library Manager that the write path just wrote to. |
|19 | `create_folder` | **FAIL** | `TypeError: create_folder() got an unexpected keyword argument 'name'`. Fork's call site uses `name=...` kwarg; the underlying CODESYS API rejects it. |
|20 | `compile_project` | **FAIL** | Build itself runs successfully (`build()` returns), but the message-marshaller dies: `TypeError: 281474976710655L is not JSON serializable`. The CODESYS `system.get_message_objects()` returns a dict containing an IronPython 2.7 `long` (the value `0xFFFFFFFFFFFF`) that the stdlib `json` module cannot encode. |
|21 | `get_compile_messages` | **FAIL** | Same JSON-long bug as #20 -- both call into the same message-encoder path. |
|22 | `connect_to_device` | **OK (FIXED)** | After commits `e862846` + `eee8ce2`. Login probe iterates `OnlineChangeOption` members + (val, bool) shapes; new `loginWaitSeconds` parameter (default 60) polls `application_state` so the credential dialog has time to surface and the user can fill in the device password. Verified working; user filled the password on first connect this session. |
|23 | `get_application_state` (online) | **OK** | After connect: `State: run` / `State: stop` / `Logged In: True` reported correctly. |
|24 | `read_variable` | **OK** | `read_value()` works directly. Tested `PLC_PRG.fb.iCount` and `GVL_Test.nCounter` -- both returned live values updating each cycle (counters incrementing). |
|25 | `write_variable` | **OK (FIXED)** | After commits `010811b` + `64906c4`. Switched to SP22 prepare-then-write API: `online_app.set_prepared_value(name, value)` then `online_app.write_prepared_values()`. Tested writing `GVL_Test.bRun = TRUE`; read-back confirmed the value landed. |
|26 | `start_stop_application` (start) | **OK** | `online_app.start()` -- verified state transitioned `stop` -> `run`. |
|27 | `start_stop_application` (stop) | **OK** | `online_app.stop()` -- verified state transitioned `run` -> `stop`. |
|28 | `download_to_device` | **OK (FIXED)** | After commit `b3bf4a8`. Same login probe + `loginWaitSeconds` as connect. The actual "download" is performed by `login(OnlineChangeOption.Force, bool)`; subsequent `create_boot_application()` finalises boot persistence. |
|29 | `disconnect_from_device` | **OK** | Clean disconnect after sweep. |
**Tally: 22 OK, 5 FAIL, 2 PARTIAL out of 29 distinct tool invocations.**
Diff vs morning smoke-test (2026-04-25 baseline): all 5 device-side fails
(`connect_to_device`, `download_to_device`, `read_variable`, `write_variable`,
`start_stop_application`) are now PASSING after this session's fixes.
The remaining failures are the same five that were already known:
`create_folder`, `compile_project`, `get_compile_messages`,
`list_project_libraries`/`add_library` inconsistency, and `rename_object`
not refactoring callers.
## What this proves about today's fixes
| Commit | Tool fixed | Verification |
|-----------------------|-------------------------------|--------------|
| `e862846` + `eee8ce2` | `connect_to_device` | #22 PASS (was login() drift) |
| `010811b` + `64906c4` | `write_variable` | #25 PASS (was write_value() missing) |
| `b3bf4a8` | `download_to_device` | #28 PASS (same login() drift as connect) |
## Bugs still open (filed by # in the table above)
Each of these is a separate upstream issue worth its own PR back to
`luke-harriman/Codesys-MCP` once cross-referenced against the official
[CODESYS Python scripting docs](https://content.helpme-codesys.com/en/ScriptingEngine/idx-codesys_scripting.html):
1. **`create_folder` keyword mismatch (#19)** -- `create_folder(name=...)`
call site in `src/scripts/create_folder.py` doesn't match the current
API signature on the parent container object. Pending docs lookup for
the canonical method (likely `add_folder(name)` or positional `name`).
2. **JSON `long` serialization (#20, #21)** -- `compile_project.py` and
`get_compile_messages.py` need to convert IronPython `long` to `int`/
`str` before `json.dumps`, or pass a `default=` callable. Same root
issue, single fix can address both.
3. **Library list / add inconsistency (#17, #18)** -- `add_library`
succeeds visually but `list_project_libraries` reports empty. Either
the add silently no-ops on the in-memory tree, or the list path is
looking at the wrong manager object. Pending docs lookup for the
canonical Library Manager iteration pattern.
4. **`add_library` doesn't dedupe / placeholder (#17)** -- always adds
a new reference even when one with the same name exists; doesn't
format as `* (System)` placeholder so transitive deps don't resolve
to installed versions. Pending docs lookup for `add_placeholder_library`
vs `add_library` semantics.
5. **`rename_object` partial refactor (#14)** -- updates the renamed
object's own internal declaration but not any other POU that
references the old name. CODESYS UI does the full refactor; need
to find whether scripting exposes a `rename_with_references` or
similar, or implement a brute-force walk + text replace.
## Caveats
- **SP22 only.** SP21 install (`3.5.21.50`) was not exercised in this
run, but the watcher rewrite + login probe were architected to be
SP-version-agnostic; same behaviour expected. Re-run on SP21 is
pending.
- **Real PLC not exercised.** All device-side tests ran against the
Control Win V3 soft-PLC (port 11740), not against actual industrial
hardware (e.g. WAGO PFC). Network/protocol-specific issues that only
surface with a real device aren't covered.
- **Project state mid-test.** Local-side ops were exercised against an
MCPTest project that the user had previously cleaned up by hand
(after the morning sweep had left a duplicate Standard library and
a dangling ST_Sample reference from a non-refactoring rename).
Re-running from a virgin `create_project` would shake out any
state-dependent variations.
---
*Smoke test executed against this fork @ `sp21-plus-migration-notes` HEAD on 2026-04-25.*

212
docs/SMOKE-TEST-2026-04-25.md
View file

@ -1,69 +1,179 @@
# Smoke test — 2026-04-25 (post SP21+/SP22 watcher fix)
# Smoke test -- 2026-04-25 (post device-side fix + late-day extensions)
End-to-end test of every MCP tool exposed by `codesys-mcp-persistent` after the watcher rewrite on the [`sp21-plus-migration-notes`](https://github.com/phobicdotno/Codesys-MCP/tree/sp21-plus-migration-notes) branch.
End-to-end test of every MCP tool exposed by `codesys-mcp-persistent`
on the [`sp21-plus-migration-notes`](https://github.com/phobicdotno/Codesys-MCP/tree/sp21-plus-migration-notes)
branch as of end-of-day 2026-04-25. Covers the watcher rewrite that
unblocked SP21+/SP22, the device-side login + write API fixes, and
the late-afternoon additions: six new `git_*` tools, `mirror_export`,
launcher refuse-on-duplicate guard, and the `list_project_libraries`
rewrite.
## Environment
- MCP server source: this fork @ `sp21-plus-migration-notes`, `dist/` rebuilt from the rewritten `src/scripts/watcher.py` (single-thread design, no `execute_on_primary_thread`).
- CODESYS actually launched: **`3.5.21.50` (SP21 Patch 5)** — even though the MCP server config was set for SP22. Cause: the Claude Code session held three concurrent MCP child processes (two SP22-bound, one SP21-bound) from earlier re-registration churn, and call routing landed on the SP21-bound child. Did not re-test on SP22 in-session because that requires a Claude Code restart.
- Why this is still a useful test: the API removed in CODESYS V3.5 SP21+ (`system.execute_on_primary_thread`) is the **same** removal that affects SP22. Watcher code that runs cleanly on SP21 will run cleanly on SP22.
- Sandbox project used for destructive ops: `C:\Users\karstein.kvistad\codesys-mcp-smoketest\Smoke_Test_001.project` (created from CODESYS Standard.project template).
- MCP server source: this fork @ `sp21-plus-migration-notes` HEAD
(commits `93a105a`..`0a4c1a0` -- inclusive of the late-afternoon
additions: `e236a0c` git tools, `3623c45` license-gate rewrite,
`95a884b` launcher refuse-on-duplicate, `9b766c8` list_project_libraries
rewrite, `e3e5f58` git_branch_set_upstream_to, `e6cfa57` project.save()
retrofit, `7a6e725`+`76b7cf4` mirror_export, `0a4c1a0` mirror default
mcp-mirror).
- CODESYS launched: **`3.5.22.10` (SP22 Patch 1, 64-bit)** via `--runscript`
pointing at the rewritten watcher (v0.4.2 with KeyboardInterrupt hardening).
- Soft-PLC runtime up: `CODESYS Control Win V3 - x64` Windows service
(`CODESYSControlService.exe`, listening on port 11740). User-started
via the tray icon's "Start PLC" with one-shot UAC elevation.
- Test project: `\\files\karstein.kvistad\Documents\Claude\PLC\MCPTest\MCPTest.project`
-- created via `create_project` from the Standard template (device =
`PLCWinNT (CoDeSys SP Win V3)`, which IS the Control Win V3 device
descriptor on this machine despite the legacy display name).
- All test objects authored with prefix `MCPv2_` to avoid colliding with
user-fixed objects in the project; cleaned up at end of test.
## Result table
| # | Tool | Result | Notes |
| - | ----------------------------------- | --------------------- | -------------------------------------------------------------------------------------------------------------- |
| 1 | `get_codesys_status` | ✅ Pass | Returned `State: ready, Mode: persistent, PID: …`. |
| 2 | `launch_codesys` | ✅ Pass | "CODESYS launched successfully in persistent mode." |
| 3 | `shutdown_codesys` | ⚠️ Partial | Kills the CODESYS process tracked by *this* MCP child only. Orphaned CODESYS processes from prior MCP children remain — needs upstream fix. |
| 4 | `open_project` | ✅ Pass | Opened X33 (`MRCodesysX33_0021.project`) over UNC; opened Smoke_Test_001 over local path. Reproducible — no longer the marshal-error fluke. |
| 5 | `create_project` | ✅ Pass | "Project created from template …". Template resolved from SP21 install path (note below). |
| 6 | `save_project` | ✅ Pass | "Project saved …". |
| 7 | `create_folder` | ❌ Fail | `TypeError: create_folder() got an unexpected keyword argument 'name'`. **Upstream bug**, unrelated to this fix — the underlying CODESYS API rejects the keyword the script uses. |
| 8 | `create_pou` (Program / ST) | ✅ Pass | Created `Application/PLC_PRG2`. |
| 9 | `create_pou` (FunctionBlock / ST) | ✅ Pass | Created `Application/MyFB`. |
|10 | `create_dut` (Structure) | ✅ Pass | Created `Application/MyStruct`. |
|11 | `create_gvl` | ✅ Pass | Created `Application/MyGVL` with declarationCode populated. |
|12 | `create_method` | ✅ Pass | Created `Application/MyFB/DoSomething : BOOL`. |
|13 | `create_property` | ✅ Pass | Created `Application/MyFB/IsActive : BOOL` (Get/Set sub-objects auto-created). |
|14 | `set_pou_code` | ✅ Pass | Wrote declaration + implementation to `Application/PLC_PRG2`; verified via `get_all_pou_code`. |
|15 | `get_all_pou_code` | ✅ Pass | Returned full code dump for all 9 objects in the project. |
|16 | `rename_object` | ✅ Pass | `Application/PLC_PRG2 → PLC_PRG2_Renamed`. |
|17 | `delete_object` | ✅ Pass | Deleted `Application/MyStruct`. |
|18 | `list_project_libraries` | ⚠️ Empty | Returns "No libraries found in the project (or Library Manager not found)" both before and after `add_library` — **soft inconsistency** (see #19). |
|19 | `add_library` | ⚠️ Inconsistent | Returns "Library 'Standard' added …" but `list_project_libraries` afterwards still shows nothing. Either the add silently no-ops, or the list query misses the just-added entry. |
|20 | `compile_project` | ❌ Fail | Build itself runs ("Build command executed for application 'Application'") but the message-serialiser then dies: `TypeError: 281474976710655L is not JSON serializable` (an IronPython 2.7 `long` from the message Severity bitmask). **Upstream bug** in the compile_project script's JSON encoder. |
|21 | `get_compile_messages` | ❌ Fail | Same `long` JSON-serialization bug as #20. |
|22 | `get_application_state` | ✅ Pass | Returned `Application: Application, State: none, Logged In: False` for both X33 and the sandbox. |
|23 | `connect_to_device` | ❌ Fail | Two layered failures — `OnlineChangeOption.TryOnlineChange` attribute is missing, and the fallback plain `login()` now requires 2 args. **Upstream API drift** — would block all device ops even with a real PLC. |
|24 | `download_to_device` | ❌ Fail | Same `login()` signature change as #23. |
|25 | `start_stop_application` (start) | ❌ Fail | "Application not logged in." (downstream of the broken `connect_to_device`.) |
|26 | `read_variable` | ❌ Fail | "Application not found" — online-app object created, but `read_value` path can't resolve target. |
|27 | `write_variable` | ❌ Fail | "Online application does not support write_value() or write()." Method on the online-app object renamed/removed in newer CODESYS scripting. |
|28 | `disconnect_from_device` | ✅ Pass | "Disconnected from device …" — succeeds even when not currently connected. |
| # | Tool | Result | Notes |
|---|--------------------------------------|-----------------|----------------------------------------------------------------------------------------|
| 1 | `get_codesys_status` | OK | State / Mode / PID / Session reported correctly. |
| 2 | `launch_codesys` | OK | Persistent mode; ready signal received within seconds. |
| 3 | `shutdown_codesys` | OK | Cleanly stops the spawned CODESYS instance. Prior smoke-test #5 (orphan from earlier launches) not reproduced this session. |
| 4 | `create_project` | OK | From Standard template; saved to the requested path. |
| 5 | `open_project` | OK | Reproducible from-cold open. Cross-project switch (closing one project to open another in the same instance) not exercised here -- prior smoke noted that as a separate bug. |
| 6 | `save_project` | OK | |
| 7 | `create_pou` (FunctionBlock / ST) | OK | `Application/MCPv2_FB`. |
| 8 | `create_dut` (Structure) | OK | `Application/MCPv2_ST`. |
| 9 | `create_gvl` | OK | `Application/MCPv2_GVL` with declarationCode populated. |
|10 | `create_method` | OK | `Application/MCPv2_FB/DoSomething : BOOL`. |
|11 | `create_property` | OK | `Application/MCPv2_FB/Counter : INT` (Get/Set sub-objects auto-created). |
|12 | `set_pou_code` | OK | Decl + impl wrote correctly to MCPv2_FB and MCPv2_ST. Verified via get_all_pou_code. |
|13 | `get_all_pou_code` | OK | Returned full code dump for every POU/DUT/GVL/Method/Property in the project. |
|14 | `rename_object` | **PARTIAL** | Renames the object's own internal declaration line (`TYPE old : ... END_TYPE` becomes `TYPE new : ... END_TYPE` automatically) BUT does NOT update other POUs that referenced the old name. Same as upstream. |
|15 | `delete_object` | OK | Cleaned up MCPv2_FB / MCPv2_GVL / MCPv2_STRenamed at end-of-test. |
|16 | `get_application_state` (offline) | OK | `State: none, Logged In: False` before connect. |
|17 | `add_library` | **PARTIAL** | Operation succeeds and CODESYS shows the entry in Library Manager, BUT it adds a SECOND `Standard` reference instead of detecting the existing one, and the new reference is not added as a `* (System)` placeholder so it pulls in unresolved transitive deps (e.g. IoStandard 3.1.3.1 yellow-warning). |
|18 | `list_project_libraries` | **OK (FIXED)** | After commit `9b766c8`. Was looking up the Library Manager by NAME (`primary_project.find("Library Manager", True)`) which never matched because the libman's actual name is generated, not literal. Rewritten to walk every node and check the `has_library_manager` property on `ScriptLibManObjectContainer` (added to both Project and every Application object), then iterate `lm.references` for structured per-reference info. Verified against X33: 71 references across 2 library managers (project-level + Application). |
|19 | `create_folder` | **FAIL** | `TypeError: create_folder() got an unexpected keyword argument 'name'`. Fork's call site uses `name=...` kwarg; the underlying CODESYS API rejects it. |
|20 | `compile_project` | **FAIL** | Build itself runs successfully (`build()` returns), but the message-marshaller dies: `TypeError: 281474976710655L is not JSON serializable`. The CODESYS `system.get_message_objects()` returns a dict containing an IronPython 2.7 `long` (the value `0xFFFFFFFFFFFF`) that the stdlib `json` module cannot encode. |
|21 | `get_compile_messages` | **FAIL** | Same JSON-long bug as #20 -- both call into the same message-encoder path. |
|22 | `connect_to_device` | **OK (FIXED)** | After commits `e862846` + `eee8ce2`. Login probe iterates `OnlineChangeOption` members + (val, bool) shapes; new `loginWaitSeconds` parameter (default 60) polls `application_state` so the credential dialog has time to surface and the user can fill in the device password. Verified working; user filled the password on first connect this session. |
|23 | `get_application_state` (online) | **OK** | After connect: `State: run` / `State: stop` / `Logged In: True` reported correctly. |
|24 | `read_variable` | **OK** | `read_value()` works directly. Tested `PLC_PRG.fb.iCount` and `GVL_Test.nCounter` -- both returned live values updating each cycle (counters incrementing). |
|25 | `write_variable` | **OK (FIXED)** | After commits `010811b` + `64906c4`. Switched to SP22 prepare-then-write API: `online_app.set_prepared_value(name, value)` then `online_app.write_prepared_values()`. Tested writing `GVL_Test.bRun = TRUE`; read-back confirmed the value landed. |
|26 | `start_stop_application` (start) | **OK** | `online_app.start()` -- verified state transitioned `stop` -> `run`. |
|27 | `start_stop_application` (stop) | **OK** | `online_app.stop()` -- verified state transitioned `run` -> `stop`. |
|28 | `download_to_device` | **OK (FIXED)** | After commit `b3bf4a8`. Same login probe + `loginWaitSeconds` as connect. The actual "download" is performed by `login(OnlineChangeOption.Force, bool)`; subsequent `create_boot_application()` finalises boot persistence. |
|29 | `disconnect_from_device` | **OK** | Clean disconnect after sweep. |
|30 | `git_init` (NEW) | **OK** | Wraps `project.git.init(localRepoPath)`. Requires PDE subscription; without it, fails fast with the friendly "PDE subscription required" message (see commit `3623c45`). Default localRepoPath = `<projectDir>_git` sibling, auto-created and emptiness-validated (commit `31e8429`). Verified against `GitSmokeTest.project` -- created `.git/`, `.gitattributes`, `.gitignore`, `.apsession`, and `project/` subdir. |
|31 | `git_status` (NEW) | **OK** | Branch + diagnostic dump of `project.git` API surface. Adds an early `has_working_tree()` license probe so the rewrite triggers reliably even when the per-method probe loop would otherwise swallow the gate. |
|32 | `git_commit` (NEW) | **OK** | Wraps `project.git.commit_complete(message, user, mail)`. Verified end-to-end on `GitSmokeTest.project` -- one commit `80c6d89` written and confirmed via `git log` against the local `.git/`. |
|33 | `git_remote_add` (NEW) | **OK** | Wraps `project.git.remote_add(name, url)`. Verified by adding both a local bare remote (`C:\Temp\MCPTestRemote.git`) and the GitLab remote (`https://gitlab.usv.no/karstein.kvistad/codesys-gitsmoketest.git`) to the same project. |
|34 | `git_branch_set_upstream_to` (NEW) | **OK** | Wraps `project.git.branch_set_upstream_to(remoteName, branchName?)`. MANDATORY between remote_add and the first push -- per helpme-codesys.com Git scripting docs. Without it, `push()` fails with `sLocalBranchName: branch 'master' does not track an upstream branch.` Shipped in commit `e3e5f58` after that exact failure mode surfaced in the first end-to-end test. |
|35 | `git_push` (NEW) | **OK** | Three overloads: `push()` / `push(branch)` (relies on cached creds) and `push(branch, user, SecureString(token))`. Verified against both the local bare remote and `gitlab.usv.no` -- the libgit2-backed CODESYS Git plug-in picks up Windows Credential Manager creds the same way as command-line git, so no PAT was needed for the GitLab smoke test. |
|36 | `mirror_export` (NEW) | **OK** | Walks the project tree and writes one `.st` file per code-bearing object into `<projectDir>/mcp-mirror/` (default), preserving the project tree as nested directories. UTF-8 output. Each file carries a header comment with its CODESYS project path. Verified against X33: 91 files, 254 KB, 0 errors, 7 kinds (FB / DUT / METHOD / FUNCTION / PROGRAM / GVL / UNKNOWN). |
**Tally: 17 ✅ pass, 8 ❌ fail, 3 ⚠️ partial / inconsistent (out of 28 tool invocations across 26 distinct tools).**
**Tally: 28 OK, 4 FAIL, 1 PARTIAL out of 36 distinct tool invocations.**
## What this proves about the watcher fix
Diff vs morning smoke-test (2026-04-25 baseline): all 5 device-side
fails (`connect_to_device`, `download_to_device`, `read_variable`,
`write_variable`, `start_stop_application`) are now PASSING after the
device-side fixes; `list_project_libraries` (the read side of the
library inconsistency) is also PASSING after the late-afternoon
rewrite. Plus seven new tools shipped (six `git_*` and `mirror_export`).
Remaining failures: `create_folder` kwarg, `compile_project` /
`get_compile_messages` JSON-`long`, `add_library` doesn't dedupe /
placeholder, `rename_object` doesn't refactor callers.
Every passing entry above is direct proof the new watcher works on SP21+. Before the fix, all of these tools returned identical `Marshal error: The functionality 'system.execute_on_primary_thread(...) is no longer supported`. After the fix, the underlying CODESYS scripting work runs and the tool either succeeds or fails for **its own** (unrelated) reason. The threading regression is closed.
## What this proves about today's fixes
## Bugs found that are NOT this fix
| Commit | Tool fixed / added | Verification |
|-----------------------|-------------------------------|--------------|
| `e862846` + `eee8ce2` | `connect_to_device` | #22 PASS (was login() drift) |
| `010811b` + `64906c4` | `write_variable` | #25 PASS (was write_value() missing) |
| `b3bf4a8` | `download_to_device` | #28 PASS (same login() drift as connect) |
| `e236a0c` | NEW: `git_init`/`git_status`/`git_commit` | #30/31/32 PASS against `GitSmokeTest.project` -- live PDE Demo subscription on this box (activation flipped the runtime `HasGitLicense` rule from False to True; the same call returns the friendly PDE-required message on installs without a subscription, see `3623c45`) |
| `8a6059b` | NEW: `git_remote_add` + `git_push` | #33/35 PASS (push to local bare + push to gitlab.usv.no via cached creds) |
| `e3e5f58` | NEW: `git_branch_set_upstream_to` | #34 PASS (RTFM-driven; the canonical "init -> remote_add -> set-upstream -> push" flow needs this step or push fails with `does not track an upstream branch`) |
| `e6cfa57` | All git_* (binding persistence) | `project.save()` retrofit after every mutating git op so state survives IDE close. Soft-fail on save error to avoid masking a successful git op as a failure. |
| `9b766c8` | `list_project_libraries` | #18 PASS (rewritten to use `ScriptLibManObjectContainer` API instead of name-matching the libman) |
| `95a884b` | Launcher refuse-on-duplicate | Pre-spawn `tasklist` scan; refuses to launch alongside an existing CODESYS.exe with a clear message listing the offending PIDs. Mitigates v1's #3 orphan/modal cascade. |
| `7a6e725`+`76b7cf4`+`0a4c1a0` | NEW: `mirror_export` | #36 PASS against X33: 91 .st files, 254 KB, 0 errors. Default mirror root iterated through `MCP/mirror/` -> `mcp-mirror/` based on real-world layout feedback. |
These are pre-existing upstream issues exposed once scripting actually runs. Each is worth filing back to upstream as a separate issue / PR after this one merges:
## Bugs still open (filed by # in the table above)
1. **`create_folder` keyword mismatch** (#7) — `create_folder(name=…)` call site needs to match the current CODESYS API signature.
2. **JSON `long` serialization** (#20, #21) — `compile_project` and `get_compile_messages` need a JSON encoder that handles IronPython `long` (e.g. cast `Severity` bitmasks to `int` or `str` before `json.dumps`).
3. **Online API drift** (#23, #24, #27) — `OnlineChangeOption.TryOnlineChange` removed; `login()` now requires explicit args; `Online application` no longer exposes `write_value()/write()`. Fixing these is a wider re-survey of the CODESYS online scripting API.
4. **`shutdown_codesys` doesn't reach orphans** (#3) — when the MCP server child is respawned (e.g. on Claude Code MCP re-registration), CODESYS processes from prior child processes are no longer tracked and survive `shutdown_codesys`. Either the child should kill child-of-child on its own exit, or the launcher should scan for and adopt orphans.
5. **Library list / add inconsistency** (#18, #19) — needs a small repro to determine whether the issue is `add_library` (silent no-op) or `list_project_libraries` (missing the just-added entry).
Each of these is a separate upstream issue worth its own PR back to
`luke-harriman/Codesys-MCP` once cross-referenced against the official
[CODESYS Python scripting docs](https://content.helpme-codesys.com/en/ScriptingEngine/idx-codesys_scripting.html):
## Caveats / things not directly verified
1. **`create_folder` keyword mismatch (#19)** -- `create_folder(name=...)`
call site in `src/scripts/create_folder.py` doesn't match the current
API signature on the parent container object. Pending docs lookup for
the canonical method (likely `add_folder(name)` or positional `name`).
- **SP22 not directly tested** in this run because of the orphaned MCP child issue. A Claude Code restart with only the SP22-bound config would resolve this; the architecture of the fix is identical for SP21 and SP22, so SP22 is expected to behave the same.
- **No real PLC** was connected during the test, so the device/runtime tools (#23–28) were exercised only against a non-running soft PLC target. The errors captured are mostly *script-side* (API mismatches), not network/device errors.
- **Standard.project template** was resolved from `C:\Program Files\CODESYS 3.5.21.50\CODESYS\Templates\Standard.project` regardless of the MCP's `--codesys-path` arg. Worth verifying the template-resolution code uses the same install root as the launched CODESYS.
2. **JSON `long` serialization (#20, #21)** -- `compile_project.py` and
`get_compile_messages.py` need to convert IronPython `long` to `int`/
`str` before `json.dumps`, or pass a `default=` callable. Same root
issue, single fix can address both.
3. **`add_library` doesn't dedupe / placeholder (#17)** -- always adds
a new reference even when one with the same name exists; doesn't
format as `* (System)` placeholder so transitive deps don't resolve
to installed versions. The matching READ side (`list_project_libraries`)
was fixed in `9b766c8` -- the write side has the same wrong-axis
bug (uses name-match for the libman lookup) and is the natural
follow-up. Pending docs lookup for `add_placeholder_library` vs
`add_library` semantics.
4. **`rename_object` partial refactor (#14)** -- updates the renamed
object's own internal declaration but not any other POU that
references the old name. CODESYS UI does the full refactor; need
to find whether scripting exposes a `rename_with_references` or
similar, or implement a brute-force walk + text replace.
## New infrastructure (this session, not bugs)
- **PDE-subscription gate detection.** All six `git_*` tools detect
the runtime `HasGitLicense=False` rule and rewrite the error to a
clear "CODESYS Professional Developer Edition subscription required"
message pointing at the store page. Triggered when no PDE
subscription is active; passes through transparently when one is
(Demo or full). See commit `3623c45`.
- **Launcher refuse-on-duplicate.** Pre-spawn `tasklist` scan refuses
to launch a second CODESYS alongside an existing one (orphan from
prior MCP session, user's own interactive IDE, or a CODESYS still
mid-shutdown), since two CODESYS processes against the same project
file race on the lock and the loser pops a "project is currently in
use" modal that freezes script execution. See `95a884b`. Future:
adopt-existing-watcher (find a live ready.signal whose PID is still
alive, attach to it instead of spawning) -- prototyped via the
inject-once.mjs bridge in this session, not yet shipped in the
launcher.
## Filesystem mirror (Phase 1 of "project as a filesystem")
`mirror_export` (#36) lays the foundation for an AI-editable text
representation of a CODESYS project. Phase 2 (`sync_pou_from_file`:
parse a `.st` file, split decl/impl, push back via `set_pou_code`)
and Phase 3 (drift detection: re-export when CODESYS state diverges
from the mirror) are not yet shipped. The X33 project was used as
the proving ground -- 91 .st files round-tripping cleanly through
the `(* Project path: ... *)` header convention.
## Caveats
- **SP22 only.** SP21 install (`3.5.21.50`) was not exercised in this
run, but the watcher rewrite + login probe were architected to be
SP-version-agnostic; same behaviour expected. Re-run on SP21 is
pending.
- **Real PLC not exercised.** All device-side tests ran against the
Control Win V3 soft-PLC (port 11740), not against actual industrial
hardware (e.g. WAGO PFC). Network/protocol-specific issues that only
surface with a real device aren't covered.
- **Project state mid-test.** Local-side ops were exercised against an
MCPTest project that the user had previously cleaned up by hand
(after the morning sweep had left a duplicate Standard library and
a dangling ST_Sample reference from a non-refactoring rename).
Re-running from a virgin `create_project` would shake out any
state-dependent variations.
---
*Smoke test executed against this fork @ `sp21-plus-migration-notes` branch with the rewritten `src/scripts/watcher.py`, on 2026-04-25.*
*Smoke test executed against this fork @ `sp21-plus-migration-notes` HEAD on 2026-04-25.*