Sketch value fields: content-based key arbiter + the gate that can judge it

The reported defect: sketch dimension labels are "not editable" — you draw a
rectangle, its Width field opens, you type, and the as-drawn number is committed
instead. It affects every sketch tool, not just the rounded rectangle.

WHAT THIS ADDS

1. The arbiter (DesignPanel CHAR_HOOK -> DesignCanvas::inline_type_char ->
   SketchInlineEditor::type_char). Routes a key by what it IS, not by who the
   window manager focused: digits, sign, decimal separator and Backspace/Delete
   go to the open value field, Enter/Tab commit, letters stay tool shortcuts.
   This is FreeCAD Sketcher's rule (DrawSketchKeyboardManager::
   detectKeyboardEventHandlingMode), and the reason its sketcher behaves the same
   on every desktop: it never asks who has focus.

2. The [UX] trace (SNAPORCA_UXTRACE) in SketchInlineEditor: open/commit/refused/
   cancel, with the prefill and what the control actually held at Enter. It did
   not exist — the ladder below was written against a surface no build emitted,
   so it could only ever report "nothing opened". typed == prefill on a commit is
   the defect's signature and nothing else makes it visible.

3. A draw-then-edit trace in DesignSketchTool: four early returns can swallow the
   value-field chain and from outside they are indistinguishable.

4. scripts/CAD/check-gui-click-edit.py — types WITHOUT clicking the field, as a
   person does, across Line/Rectangle/Circle/Slot/Polygon/Ellipse/Arc plus label
   click-to-edit, and asserts committed == typed != prefill.

5. scripts/CAD/focus-loop.sh — sync/build/assert on behemoth. NOT the orcacad-gui
   rig: its image pins deps 216 non-CAD files behind cad-mainline, so today's CAD
   sources cannot build there without a deps rebuild.

WHAT IS PROVEN, AND WHAT IS NOT

Green under openbox: 28 checks, every tool, committed == typed != prefill.

But openbox CANNOT adjudicate this bug and the ladder says so in place. There the
field always wins the keyboard, so the same ladder also passes against a binary
with the arbiter compiled out — measured twice. Two ways of removing the keyboard
were tried and both are recorded as dead ends: XSetInputFocus loses to the field's
own re-focus CallAfter, and XSendEvent (xdotool --window) is dropped by GTK, which
made every run red regardless of the code.

Under metacity — same focus-stealing-prevention lineage as the user's mutter — the
mechanism appears in the WM's own log:

    Buggy client sent a _NET_ACTIVE_WINDOW message with a timestamp of 0

That is the activation being refused, which is exactly the reported symptom.
present_toplevel() already asks for a server timestamp, so a path is still falling
through to frame->Raise(), which sends time 0. That is the next thing to fix, and
it is tracked; the arbiter alone does not close it. metacity also aborts on this
window (frames.c:1239), so the gate needs a WM that survives before it can return
a verdict.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011FbJKJAJxxkhDTs9XdZzKA

.claude/loopspec/sketch-focus-arbiter.spec.md

@@ -0,0 +1,86 @@
+# DELEGATION SPECIFICATION: HARNESS-DRIVEN VALIDATION LOOP
+slug: sketch-focus-arbiter · repo: /home/tommaso/projects/apps/orca_cad · branch: cad-mainline
+
+## 1. TARGET GOAL
+
+**Functional Objective.** Keyboard input in the Design tab is routed by WHAT THE KEY IS, not by
+which widget the window manager decided to focus. Adopted from FreeCAD's
+`DrawSketchKeyboardManager::detectKeyboardEventHandlingMode`
+(src/Mod/Sketcher/Gui/DrawSketchKeyboardManager.cpp), which never queries focus at all:
+
+  - digit, `-`, `.`, `,`  -> the open value field
+  - Backspace / Delete    -> the open value field (when one is open)
+  - Enter / Return / Tab  -> commit the field, control returns to the view
+  - a letter              -> the sketch-tool shortcut map, as today
+  - Esc                   -> the existing CadLevel LIFO (DesignInteraction.hpp), unchanged
+  - anything else         -> sticky: whoever had it keeps it
+
+Observable postcondition: for EVERY sketch tool that opens a value field, a value typed
+immediately after the field appears — with NO click into the field — is the value committed.
+Today the prefill is committed instead whenever the WM withholds focus.
+
+**Target Files / Scope (writable).**
+  src/slic3r/GUI/CAD/DesignPanel.cpp        (the arbiter lives in the existing wxEVT_CHAR_HOOK)
+  src/slic3r/GUI/CAD/DesignCanvas.cpp/.hpp  (forwarding entry points only)
+  src/slic3r/GUI/CAD/SketchInlineEditor.cpp/.hpp (accept a programmatically delivered character)
+  scripts/CAD/check-gui-click-edit.py       (F2P oracle — authoring exception, see §4)
+Everything else read-only. No dependency additions, no reformatting.
+
+**Open Bindings.**
+  - The in-canvas ImGui field on wip/in-canvas-value-field is NOT in scope. Default: the arbiter
+    is implemented against the CURRENT wxFrame field on cad-mainline, because content-based
+    routing makes the window's focus irrelevant either way. If it later moves in-canvas the
+    arbiter is unchanged.
+  - Tools whose field is opened by a toolbar button rather than a gesture (Constrain path) are
+    covered by the same arbiter but are not in the F2P tool list. Default: assert them in P2P only.
+
+## 2. HARNESS ENVIRONMENT & GROUND TRUTH
+
+The rig container `orcacad-gui` on nativedev IS the harness. Xvfb `:11` + openbox, the app under
+test, `xdotool` for synthetic input, and an MCP socket at `/tmp/mcp.sock` that reports sketch
+state as JSON. It is a closed loop: drive input, read geometry back, assert. No window manager
+politics, no human.
+
+  Harness interface (ordered; each slot one invocation, one exit code):
+    S1 sync    docker cp <file> orcacad-gui:/OrcaSlicer/<path>
+    S2 build   docker exec orcacad-gui ninja -C /OrcaSlicer/build orca-slicer
+    S3 restart docker exec orcacad-gui /OrcaSlicer/scripts/CAD/start-headless-gui.sh
+    S4 F2P     docker exec -e DISPLAY=:11 orcacad-gui python3 /tmp/check-gui-click-edit.py --attach
+    S5 P2P     docker exec -e DISPLAY=:11 orcacad-gui python3 /tmp/check-gui-sketching.py
+
+**F2P.** `scripts/CAD/check-gui-click-edit.py`. For each of Line, Rectangle, Circle, Slot,
+Polygon, Ellipse and Rounded rectangle: arm the tool, draw it, and type a value that differs
+from the prefill WITHOUT clicking the field. Assert the committed value equals the typed value.
+The ladder must FAIL against unmodified cad-mainline — that is what proves it asserts something.
+
+**P2P.** `scripts/CAD/check-gui-sketching.py`, the existing gesture ladder, minus anything red at
+baseline. NOTE: it calls `focus_field()` — one click into the field before typing — which is the
+workaround this whole task removes. It stays green as a regression guard; it is NOT evidence.
+
+**Test Integrity Constraint.** `focus_field()` in check-gui-sketching.py must NOT be deleted to
+make things pass, and check-gui-click-edit.py must NOT be weakened. Either invalidates the run.
+
+## 3. VERIFICATION COMMANDS
+1. Static: `docker exec orcacad-gui ninja -C /OrcaSlicer/build orca-slicer` (warnings delta only;
+   this repo configures no linter — the compiler is the static gate. Absolute-zero is NOT the gate.)
+2. Harness: `docker exec -e DISPLAY=:11 orcacad-gui python3 /tmp/check-gui-click-edit.py --attach`
+3. Regression: `docker exec -e DISPLAY=:11 orcacad-gui python3 /tmp/check-gui-sketching.py`
+
+## 4. CONVERGENCE LOOP — ceiling 8 iterations
+EDIT (scoped) -> EXECUTE S1..S5 -> PARSE the ladder's per-tool assertions and the [UX]/[KEYTRACE]
+lines -> PATCH from the parsed cause. On ceiling without convergence: stop, report the last diff
+and the unresolved failure set. Do not report success.
+
+F2P authoring exception: check-gui-click-edit.py is writable, and must be shown RED against
+unmodified source before any source edit counts.
+
+## 5. TERMINATION CRITERIA
+- [ ] S2 exits 0, and introduces no compiler warning absent from the baseline.
+- [ ] S4 ALL_PASSED — every tool commits the typed value, no click into the field.
+- [ ] S5 shows zero regressions against its recorded baseline pass count.
+- [ ] F2P proven red without the fix (source stashed, ladder re-run, must FAIL).
+
+## 6. GUARDRAILS
+Zero-assumption: no completion claim without captured stdout and exit codes. Oracle supremacy:
+the ladder's verdict overrides my judgement. Blast radius: §1 files only. Baseline obligation:
+run §3 once before the first edit and record it.