Merge stopping the gimbal where it is when switching from a scan to manual

README.md

@@ -322,7 +322,7 @@ job. Details, payloads and safety rules: [docs/mqtt-api.md](docs/mqtt-api.md).
 
 | Command | Does |
 |---|---|
-| `mode auto` / `mode manual` | sweep the scan grid / stop and take the gimbal by hand (it stays homed and powered) |
+| `mode auto` / `mode manual` | sweep the scan grid / stop the gimbal where it is and take it by hand (it stays homed and powered) |
 | `mode pan <heading> <span> [<pitch>]` | watch one direction: `mode pan 90 30` covers 75°-105°, a picture every 6° |
 | `gimbal home` | find the endstops and centre - needed after every power-up |
 | `gimbal move <yaw>[,<pitch>]` | point at a **true bearing** (0 = north) and an elevation, in degrees; without a pitch only yaw moves. A yaw-only tower ignores the pitch and refuses pitch-only commands |

deploy.sh

@@ -219,6 +219,26 @@ ssh "$REMOTE_HOST" "mkdir -p '$REMOTE_DIR'"
 # build*/ covers every local build tree, not just build/. They are machine-specific
 # and the device regenerates its own build/ remotely, so shipping build-tui/ etc.
 # just pushed tens of MB of x86 object files the device can never use.
+#
+# Only what git knows about is copied: the tracked files and new ones not yet
+# added, as they are in the working tree, so uncommitted work still deploys.
+# Whatever .gitignore names stays behind - build trees, the editor's cache,
+# Claude Code worktrees, the 12-bit report's full-size example pictures - which
+# made a deploy 505 MB, where 7 MB was needed, while the whole folder was
+# mirrored. The set is staged locally so the rsync below keeps its --delete: a
+# file git no longer has, or clutter an older deploy left, is removed from the
+# tower, and the excludes still protect the tower's own files as before.
+git rev-parse --git-dir >/dev/null 2>&1 || {
+    echo "error: deploy from a git checkout - it decides what is copied" >&2
+    exit 1
+}
+STAGE="$(mktemp -d)"
+trap 'rm -rf "$STAGE"' EXIT
+chmod 0755 "$STAGE"     # rsync gives the tower's project folder these permissions
+# --ignore-missing-args: a tracked file deleted but not yet committed is listed
+git ls-files -z --cached --others --exclude-standard |
+    rsync -a --from0 --files-from=- --ignore-missing-args ./ "$STAGE/"
+
 echo ">> syncing to $REMOTE_HOST:$REMOTE_DIR"
 rsync -avz --delete \
     --exclude 'build*/' \
@@ -235,7 +255,7 @@ rsync -avz --delete \
     --exclude 'config/camera_profiles/**/*.xml' \
     --exclude 'config/camera_profiles/**/*.txt' \
     --exclude 'images/' \
-    ./ "$REMOTE_HOST:$REMOTE_DIR/"
+    "$STAGE/" "$REMOTE_HOST:$REMOTE_DIR/"
 
 # .git is not synced, so leave the revision beside the tree for the tools that
 # run THERE and need to know what they are building (install-tower, flash.sh).

docs/deployment.md

@@ -76,7 +76,11 @@ and which mode it starts in are the tower's own `config.ini` `[Startup] home` /
 ```
 
 It syncs the tree (leaving `.git` behind, and writing the revision into
-`.revision` beside it instead), then configures and builds **on the tower**. The
+`.revision` beside it instead), then configures and builds **on the tower**. Only
+files git knows about are synced - tracked ones and new ones not yet added, as
+they are in the working tree - so nothing `.gitignore` names (build trees, editor
+caches, `.claude/worktrees/`, the camera bench's example pictures) reaches the
+tower, and a file left there by an older deploy is removed. The
 program already running keeps running the old binary until it is restarted:
 
 ```bash

src/core/Application.cpp

@@ -1849,6 +1849,7 @@ struct Application::Impl {
         if (m == Mode::Manual) camera->stop();
         else camera->start();
         scheduler->setMode(m, pan);
+        if (m == Mode::Manual && was != Mode::Manual) stopScanMove();
         if (m == Mode::Pan) {
             const auto& h = scheduler->panHeadingsNow();
             LOG_INFO << "mode pan: " << panText(pan) << " - " << h.size()
@@ -1861,6 +1862,22 @@ struct Application::Impl {
         return true;
     }
 
+    // Leaving a scan: stop the move it last sent, which the firmware would
+    // otherwise finish - a slow return to the first waypoint can take many
+    // seconds the operator spends waiting for the gimbal to answer. The firmware
+    // ramps the axis down and holds where it comes to rest (it does not resume
+    // the move), and the next MOVE or jog supersedes a stop still ramping down.
+    // Sent whether or not telemetry shows motion, since a MOVE sent this tick
+    // is not in it yet. Never during a homing, which a STOP ends in ERROR.
+    void stopScanMove() {
+        const MotorTelemetry t = motor->telemetry();
+        if (t.yaw.state == AxisState::Homing ||
+            (t.pitch_present && t.pitch.state == AxisState::Homing))
+            return;
+        if (anyLeftoverMove()) LOG_INFO << "stopping the gimbal where it is";
+        motor->sendCommand("STOP ALL");
+    }
+
     // A pan the gimbal can actually do: its pitch inside the homed window (not
     // clamped to it - that would image an elevation nobody asked for), and at
     // least one of its headings inside the reachable yaw arc.

src/core/HelpText.cpp

@@ -35,7 +35,7 @@ const std::vector<HelpSection>& helpCatalog() {
                 "Pitch defaults to 0 (level); a yaw-only tower ignores it. Key p pans",
                 "30 deg around the current aim.",
                 "Example:  mode pan 90 30"}},
-            {"mode manual", "Stop scanning; aim by hand (key m). Cancels a running procedure.", {}},
+            {"mode manual", "Stop scanning, and the gimbal where it is; aim by hand (key m). Cancels a running procedure.", {}},
         }},
         {"Capture", "Images and cameras. Manual mode, homed gimbal.", {
             {"snapshot", "Take one image now (key Space).", {