Debugging Instruments

Debugging and Instruments

Contents

• LLDB Debugging
• Memory Debugging
• Hang Diagnostics
• Build Failure Triage
• Instruments Overview
• Common Mistakes
• Review Checklist
• References

LLDB Debugging

Essential Commands

(lldb) po myObject              # Print object description (calls debugDescription)
(lldb) p myInt                  # Print with type info (uses LLDB formatter)
(lldb) v myLocal                # Frame variable — fast, no code execution
(lldb) bt                       # Backtrace current thread
(lldb) bt all                   # Backtrace all threads
(lldb) frame select 3           # Jump to frame #3 in the backtrace
(lldb) thread list              # List all threads and their states
(lldb) thread select 4          # Switch to thread #4

Use v over po when you only need a local variable value — it does not execute code and cannot trigger side effects.

Breakpoint Management

(lldb) br set -f ViewModel.swift -l 42          # Break at file:line
(lldb) br set -n viewDidLoad                     # Break on function name
(lldb) br set -S setValue:forKey:                 # Break on ObjC selector
(lldb) br modify 1 -c "count > 10"              # Add condition to breakpoint 1
(lldb) br modify 1 --auto-continue true          # Log and continue (logpoint)
(lldb) br command add 1                          # Attach commands to breakpoint
> po self.title
> continue
> DONE
(lldb) br disable 1                              # Disable without deleting
(lldb) br delete 1                               # Remove breakpoint

Expression Evaluation

(lldb) expr myArray.count                        # Evaluate Swift expression
(lldb) e -l swift -- import UIKit                # Import framework in LLDB
(lldb) e -l swift -- self.view.backgroundColor = .red  # Modify state at runtime
(lldb) e -l objc -- (void)[CATransaction flush]  # Force UI update after changes

After modifying a view property in the debugger, call CATransaction.flush() to see the change immediately without resuming execution.

Watchpoints

(lldb) w set v self.score                        # Break when score changes
(lldb) w set v self.score -w read               # Break when score is read
(lldb) w modify 1 -c "self.score > 100"         # Conditional watchpoint
(lldb) w list                                    # Show active watchpoints
(lldb) w delete 1                                # Remove watchpoint

Watchpoints are hardware-backed (limited to ~4 on ARM). Use them to find unexpected mutations — the debugger stops at the exact line that changes the value.

Symbolic Breakpoints

Set breakpoints on methods without knowing the file. Useful for framework or system code:

(lldb) br set -n "UIViewController.viewDidLoad"
(lldb) br set -r ".*networkError.*"              # Regex on symbol name
(lldb) br set -n malloc_error_break              # Catch malloc corruption
(lldb) br set -n UIViewAlertForUnsatisfiableConstraints  # Auto Layout issues

In Xcode, use the Breakpoint Navigator (+) to add symbolic breakpoints for common diagnostics like -[UIApplication main] or swift_willThrow.

Memory Debugging

Memory Graph Debugger Workflow

1. Run the app in Debug configuration.
2. Reproduce the suspected leak (navigate to a screen, then back).
3. Tap the Memory Graph button in Xcode's debug bar.
4. Look for purple warning icons — these indicate leaked objects.
5. Select a leaked object to see its reference graph and backtrace.

Enable Malloc Stack Logging (Scheme > Diagnostics) before running so the Memory Graph shows allocation backtraces.

Common Retain Cycle Patterns

Closure capturing self strongly:

// LEAK — closure holds strong reference to self
class ProfileViewModel {
    var onUpdate: (() -> Void)?

    func startObserving() {
       
            self.refresh()  // strong capture of self
        }
    }
}

// FIXED — use [weak self]
func startObserving() {
    [weak self] in
        self?.refresh()
    }
}

Strong delegate reference:

// LEAK — strong delegate creates a cycle
protocol DataDelegate: AnyObject {
    func didUpdate()
}

class DataManager {
    var delegate: DataDelegate?  // should be weak
}

// FIXED — weak delegate
class DataManager {
    weak var delegate: DataDelegate?
}

Timer retaining target:

// LEAK — Timer.scheduledTimer retains its target
timer = Timer.scheduledTimer(
    timeInterval: 1.0, target: self,
    selector: #selector(tick), userInfo: nil, repeats: true
)

// FIXED — use closure-based API with [weak self]
timer = Timer.scheduledTimer(withTimeInterval: 1.0, repeats: true) { [weak self] _ in
    self?.tick()
}

Instruments: Allocations and Leaks

• Allocations template: Track memory growth over time. Use the "Mark Generation" feature to isolate allocations created between user actions (e.g., open/close a screen).
• Leaks template: Automatically detects reference cycles at runtime. Run alongside Allocations for a complete picture.
• Filter by your app's module name to exclude system allocations.

Malloc Stack Logging

Enable in Scheme > Run > Diagnostics > Malloc Stack Logging (All Allocations). This records the call stack for every allocation, letting the Memory Graph Debugger and leaks CLI show where objects were created.

# CLI leak detection
leaks --atExit -- ./MyApp.app/MyApp
# Symbolicate with dSYMs for readable stacks

Hang Diagnostics

Identifying Main Thread Hangs

A hang occurs when the main thread is blocked for > 250ms (noticeable) or

1s (severe). Common detection tools:

• Thread Checker (Xcode Diagnostics): warns about non-main-thread UI calls
• os_signpost and OSSignposter: mark intervals for Instruments
• MetricKit hang diagnostics: production hang detection (see metrickit-diagnostics skill for MXHangDiagnostic)

import os

let signposter = OSSignposter(subsystem: "com.example.app", category: "DataLoad")

func loadData() async {
    let state = signposter.beginInterval("loadData")
    let result = await fetchFromNetwork()
    signposter.endInterval("loadData", state)
    process(result)
}

Using the Time Profiler

1. Product > Profile (Cmd+I) to launch Instruments.
2. Select the Time Profiler template.
3. Record while reproducing the slow interaction.
4. Focus on the main thread — sort by "Weight" to find hot paths.
5. Check "Hide System Libraries" to see only your code.
6. Double-click a heavy frame to jump to source.

Common Hang Causes

Build Failure Triage

Reading Compiler Diagnostics

• Start from the first error — subsequent errors are often cascading.
• Search for the error code (e.g., error: cannot convert) in the build log.
• Use Report Navigator (Cmd+9) for the full build log with timestamps.

SPM Dependency Resolution

# Common: version conflict
error: Dependencies could not be resolved because root depends on 'Package' 1.0.0..<2.0.0

# Fix: check Package.resolved and update version ranges
# Reset package caches if needed:
rm -rf ~/Library/Caches/org.swift.swiftpm
rm -rf .build
swift package resolve

Module Not Found / Linker Errors

Build settings to inspect first:

• FRAMEWORK_SEARCH_PATHS
• OTHER_LDFLAGS
• SWIFT_INCLUDE_PATHS
• BUILD_LIBRARY_FOR_DISTRIBUTION (for XCFrameworks)

Instruments Overview

Template Selection Guide

xctrace CLI for CI Profiling

# Record a trace from the command line
xcrun xctrace record --device "My iPhone" \
    --template "Time Profiler" \
    --output profile.trace \
    --launch MyApp.app

# Export trace data as XML for automated analysis
xcrun xctrace export --input profile.trace --xpath '/trace-toc/run/data/table'

# List available templates
xcrun xctrace list templates

# List connected devices
xcrun xctrace list devices

Use xctrace in CI pipelines to catch performance regressions automatically. Compare exported metrics between builds.

Common Mistakes

DON'T: Use print() for debugging instead of os.Logger

print() output is not filterable, has no log levels, and is not automatically stripped from release builds. It pollutes the console and makes it impossible to isolate relevant output.

// WRONG — unstructured, not filterable, stays in release builds
print("user tapped button, state: \(viewModel.state)")
print("network response: \(data)")

// CORRECT — structured logging with Logger
import os

let logger = Logger(subsystem: "com.example.app", category: "UI")

logger.debug("Button tapped, state: \(viewModel.state, privacy: .public)")
logger.info("Network response received, bytes: \(data.count)")

Logger messages appear in Console.app with filtering by subsystem and category, and .debug messages are automatically excluded from release builds.

DON'T: Forget to enable Malloc Stack Logging before memory debugging

Without Malloc Stack Logging, the Memory Graph Debugger shows leaked objects but cannot display allocation backtraces, making it difficult to find the code that created them.

// WRONG — open Memory Graph without enabling Malloc Stack Logging
// Result: leaked objects visible but no allocation backtrace

// CORRECT — enable BEFORE running:
// Scheme > Run > Diagnostics > check "Malloc Stack Logging: All Allocations"
// Then run, reproduce the leak, and open Memory Graph

DON'T: Debug optimized code expecting full variable visibility

In Release (optimized) builds, the compiler may inline functions, eliminate variables, and reorder code. LLDB cannot display optimized-away values.

// WRONG — profiling with Debug build, debugging with Release build
// Debug builds: extra runtime checks distort perf measurements
// Release builds: variables show as "<optimized out>" in debugger

// CORRECT approach:
// Debugging: use Debug configuration (full symbols, no optimization)
// Profiling: use Release configuration (realistic performance)

DON'T: Stop on every loop iteration without conditional breakpoints

Breaking on every iteration wastes time and makes it hard to find the specific case you care about.

// WRONG — breakpoint on line inside loop, stops 10,000 times
for item in items {
    process(item)  // breakpoint here stops on EVERY item
}

// CORRECT — use a conditional breakpoint:
// (lldb) br set -f MyFile.swift -l 42 -c "item.id == targetID"
// Or in Xcode: right-click breakpoint > Edit > add Condition

DON'T: Ignore Thread Sanitizer warnings

Thread Sanitizer (TSan) warnings indicate data races that may only crash intermittently. They are real bugs, not false positives.

// WRONG — ignoring TSan warning about concurrent access
var cache: [String: Data] = [:]  // accessed from multiple threads

// CORRECT — protect shared mutable state
actor CacheActor {
    var cache: [String: Data] = [:]

    func get(_ key: String) -> Data? { cache[key] }
    func set(_ key: String, _ value: Data) { cache[key] = value }
}

Enable TSan: Scheme > Run > Diagnostics > Thread Sanitizer. Note: TSan cannot run simultaneously with Address Sanitizer.

Review Checklist

• Using os.Logger instead of print() for diagnostic output
• Malloc Stack Logging enabled before memory debugging sessions
• Memory Graph Debugger checked after dismiss/dealloc flows
• Delegates declared as weak var to prevent retain cycles
• Closures stored as properties use [weak self] capture lists
• Timers use closure-based API with [weak self]
• Thread Sanitizer enabled in test schemes
• No synchronous I/O or heavy computation on the main thread
• Time Profiler run on Release build for performance baselines
• Build failures triaged from the first error in the build log
• OSSignposter used for custom performance intervals
• Conditional breakpoints used for loop/collection debugging

References

• Logging (unified logging system)
• Logger
• OSSignposter
• Generating log messages from your code
• Recording performance data (signposts)
• Diagnosing memory, thread, and crash issues early
• Data races
• Reducing your app's memory use
• Profiling apps using Instruments
• Analyzing the performance of your shipping app
• LLDB command reference: references/lldb-patterns.md
• Instruments template guide: references/instruments-guide.md