Fix scroll and scrollUntilVisible on Android

Use ADB input swipe for reliable scrolling instead of Appium gestures,
which are unreliable on many Android devices/emulators. Log a warning
when falling back to the Appium scroll path due to missing ADB.

Also distinguish "element not found" errors from infrastructure failures
in scrollUntilVisible so connection errors are propagated immediately
instead of being silently swallowed until all scrolls are exhausted.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: maggialejandro

CHANGELOG.md

@@ -10,6 +10,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 - `runFlow: when` conditions with variable expressions (e.g., `${output.element.id}`) were never expanded, causing conditions to always evaluate as false and silently skip conditional blocks
 - iOS real device: `acceptAlertButtonSelector` matched "Don't Allow" instead of "Allow" — `CONTAINS[c] 'Allow'` matched both buttons, causing WDA to reject permission dialogs. Changed to `BEGINSWITH[c] 'Allow'` with `OK` fallback for older iOS versions
+- Android: `scroll` and `scrollUntilVisible` did not scroll — Appium `/appium/gestures/scroll` endpoint is unreliable on many devices. Replaced with ADB `input swipe` for direct OS-level input injection (falls back to Appium if ADB is unavailable). Also added on-screen bounds verification to prevent false positives from off-screen elements in the Android view hierarchy
 
 ## [1.0.7] - 2026-02-20
 

pkg/driver/uiautomator2/commands.go

@@ -2,6 +2,7 @@ package uiautomator2
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"strconv"
 	"strings"
@@ -395,51 +396,72 @@ func (d *Driver) scroll(step *flow.ScrollStep) *core.CommandResult {
 		direction = "down"
 	}
 
-	// Get screen size for dynamic scroll area
+	// Get screen size for scroll coordinates
 	width, height, err := d.screenSize()
 	if err != nil {
 		return errorResult(err, "Failed to get screen size")
 	}
 
-	// Use most of screen for scroll area (leave margins)
-	area := uiautomator2.NewRect(0, height/8, width, height*3/4)
-
-	// /appium/gestures/scroll already uses scroll semantics — no inversion needed
-	if err := d.client.ScrollInArea(area, direction, 0.5, 0); err != nil {
+	// Use ADB input swipe for reliable scrolling
+	if err := d.scrollBySwipe(direction, width, height); err != nil {
 		return errorResult(err, fmt.Sprintf("Failed to scroll: %v", err))
 	}
 
 	return successResult(fmt.Sprintf("Scrolled %s", direction), nil)
 }
 
+// isElementNotFoundError returns true if the error indicates the element was simply
+// not found (expected during scrolling). Returns false for infrastructure errors
+// (connection refused, request failures, etc.) which should be propagated immediately.
+func isElementNotFoundError(err error) bool {
+	if errors.Is(err, context.DeadlineExceeded) {
+		return true
+	}
+	msg := strings.ToLower(err.Error())
+	notFoundPhrases := []string{"not found", "no elements match", "no such element", "could not be located", "context deadline exceeded"}
+	for _, phrase := range notFoundPhrases {
+		if strings.Contains(msg, phrase) {
+			return true
+		}
+	}
+	return false
+}
+
 func (d *Driver) scrollUntilVisible(step *flow.ScrollUntilVisibleStep) *core.CommandResult {
 	direction := strings.ToLower(step.Direction)
 	if direction == "" {
 		direction = "down"
 	}
 
-	maxScrolls := 10
+	maxScrolls := step.MaxScrolls
+	if maxScrolls <= 0 {
+		maxScrolls = 10
+	}
 
-	// Get screen size for dynamic scroll area
+	// Get screen size for scroll coordinates
 	width, height, err := d.screenSize()
 	if err != nil {
 		return errorResult(err, "Failed to get screen size")
 	}
 
-	// Use most of screen for scroll area (leave margins)
-	area := uiautomator2.NewRect(0, height/8, width, height*3/4)
-
 	for i := 0; i < maxScrolls; i++ {
 		// Try to find element (short timeout - includes page source fallback)
 		_, info, err := d.findElement(step.Element, true, 1000)
 		if err == nil && info != nil {
-			// Element found - return success
-			return successResult(fmt.Sprintf("Element found after %d scrolls", i), info)
+			// On Android, UIAutomator can find elements that exist in the view hierarchy
+			// but are off-screen (e.g., in ScrollView). Verify the element is actually
+			// visible on screen by checking its bounds overlap with the viewport.
+			if isElementOnScreen(info, width, height) {
+				return successResult(fmt.Sprintf("Element found after %d scrolls", i), info)
+			}
+			// Element exists in hierarchy but is off-screen - continue scrolling
+		} else if err != nil && info == nil && !isElementNotFoundError(err) {
+			return errorResult(err, "Failed to find element")
 		}
 
-		// /appium/gestures/scroll already uses scroll semantics — no inversion needed
-		if err := d.client.ScrollInArea(area, direction, 0.3, 0); err != nil {
-			return errorResult(err, fmt.Sprintf("Failed to scroll: %v", err))
+		// Use ADB input swipe for reliable scrolling (Appium gestures/scroll is unreliable)
+		if err := d.scrollBySwipe(direction, width, height); err != nil {
+			return errorResult(err, "Failed to scroll")
 		}
 
 		time.Sleep(300 * time.Millisecond)
@@ -448,6 +470,63 @@ func (d *Driver) scrollUntilVisible(step *flow.ScrollUntilVisibleStep) *core.Com
 	return errorResult(fmt.Errorf("element not found"), fmt.Sprintf("Element not found after %d scrolls", maxScrolls))
 }
 
+// scrollBySwipe performs a scroll gesture using ADB input swipe for reliability.
+// Falls back to Appium gestures/scroll if ADB is not available.
+func (d *Driver) scrollBySwipe(direction string, screenWidth, screenHeight int) error {
+	centerX := screenWidth / 2
+	startY := screenHeight * 3 / 5
+	endY := screenHeight * 2 / 5
+	durationMs := 300
+
+	// Calculate swipe coordinates based on direction
+	// Swipe direction is opposite of scroll direction:
+	// scroll DOWN (see content below) = swipe finger UP
+	// scroll UP (see content above) = swipe finger DOWN
+	var fromX, fromY, toX, toY int
+	switch direction {
+	case "up":
+		fromX, fromY = centerX, endY
+		toX, toY = centerX, startY
+	case "down":
+		fromX, fromY = centerX, startY
+		toX, toY = centerX, endY
+	case "left":
+		centerY := screenHeight / 2
+		fromX, fromY = screenWidth*2/5, centerY
+		toX, toY = screenWidth*3/5, centerY
+	case "right":
+		centerY := screenHeight / 2
+		fromX, fromY = screenWidth*3/5, centerY
+		toX, toY = screenWidth*2/5, centerY
+	default:
+		fromX, fromY = centerX, startY
+		toX, toY = centerX, endY
+	}
+
+	// Prefer ADB shell for reliable input injection
+	if d.device != nil {
+		cmd := fmt.Sprintf("input swipe %d %d %d %d %d", fromX, fromY, toX, toY, durationMs)
+		_, err := d.device.Shell(cmd)
+		return err
+	}
+
+	// Fallback to Appium gestures if no ADB access — this path is unreliable
+	// on many Android devices/emulators, so log a warning to aid debugging.
+	logger.Warn("ADB not available, falling back to Appium scroll (may be unreliable)")
+	area := uiautomator2.NewRect(0, screenHeight/8, screenWidth, screenHeight*3/4)
+	return d.client.ScrollInArea(area, direction, 0.5, 0)
+}
+
+// isElementOnScreen checks if an element's bounds overlap with the visible screen area.
+// Returns false if bounds have no area (zero width or height) or are entirely off-screen.
+func isElementOnScreen(info *core.ElementInfo, screenWidth, screenHeight int) bool {
+	b := info.Bounds
+	if b.Width == 0 || b.Height == 0 {
+		return false
+	}
+	return b.X+b.Width > 0 && b.X < screenWidth && b.Y+b.Height > 0 && b.Y < screenHeight
+}
+
 func (d *Driver) swipe(step *flow.SwipeStep) *core.CommandResult {
 	// Check if coordinate-based swipe (percentage or absolute)
 	if step.Start != "" && step.End != "" {

pkg/driver/uiautomator2/commands_test.go

@@ -1,6 +1,7 @@
 package uiautomator2
 
 import (
+	"context"
 	"errors"
 	"fmt"
 	"net/http"
@@ -4067,6 +4068,34 @@ func TestLaunchAppViaShellAmStartErrorWithArgs(t *testing.T) {
 // Verify MockUIA2Client satisfies UIA2Client at compile time.
 var _ UIA2Client = (*MockUIA2Client)(nil)
 
+func TestIsElementNotFoundError(t *testing.T) {
+	tests := []struct {
+		name     string
+		err      error
+		expected bool
+	}{
+		{"context deadline exceeded", context.DeadlineExceeded, true},
+		{"wrapped deadline exceeded", fmt.Errorf("element 'x' not found: %w", context.DeadlineExceeded), true},
+		{"element not found", fmt.Errorf("element not found"), true},
+		{"no elements match", fmt.Errorf("no elements match selector"), true},
+		{"no such element", fmt.Errorf("no such element: An element could not be located"), true},
+		{"could not be located", fmt.Errorf("An element could not be located on the page"), true},
+		{"appium deadline with no such element", fmt.Errorf("context deadline exceeded: no such element: An element could not be located on the page using the given search parameters"), true},
+		{"connection refused", fmt.Errorf("connection refused"), false},
+		{"send request failed", fmt.Errorf("send request failed"), false},
+		{"EOF", fmt.Errorf("unexpected EOF"), false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := isElementNotFoundError(tt.err)
+			if got != tt.expected {
+				t.Errorf("isElementNotFoundError(%q) = %v, want %v", tt.err, got, tt.expected)
+			}
+		})
+	}
+}
+
 // Verify uiautomator2.DeviceInfo is used correctly.
 var _ = &uiautomator2.DeviceInfo{}