sentry_init() takes too much time and blocks UI thread

[nenseso]

Description
Using sentry-native version 0.6.1, I tested the project startup speed and found that it took a very long time, 2.9s. I checked it through instruments-App Launch and found that the function stack is as follows. Is there any way to optimize it?

When does the problem happen

• During build
• During run-time
• When capturing a hard crash

Environment

• OS: [e.g. macOS 13.5.2, Apple M1]
• Compiler: [AppleClang 14.0.3.14030022]
• CMake version and config: [3.24.2]

[supervacuus]

Hello @nenseso, and thanks for the report! Can you provide more context on your setup? I would be surprised if, in a release build of the Native SDK, sentry_init() with the crashpad-backend on macOS would take more than a couple of milliseconds to initialize (the Native SDK is also developed on M1-based macOS systems, so that would immediately be apparent). This must be an issue that is connected with your environment.

It also seems from your first screenshot that it is not the ChildPortHandshake::RunClient() path that is responsible for the significant delay, but rather the code (where the symbol couldn't be resolved) that executes in HandlerStarter::CommonStart() before that. Before the handshake, the crashpad_handler is started. Do you have a virus scanner or any (security-related) background process that monitors application starts?

[nenseso]

Thank you for your reply! I re-tested many times and found that instruments-App Launch seems to affect the execution time of senty_init()
Testing Method:

1. Add execution time test code before and after sentry_init()

    auto start = std::chrono::high_resolution_clock::now();

    sentry_init(options);
    
    auto end = std::chrono::high_resolution_clock::now();
    auto duration = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count();
    LOGI("sentry init duration: {} microseconds", duration);

2. After building the debug product package on the local machine, use instruments-App Launch or manually click to launch the application. Sentry starts very quickly, about a few hundred µs, as shown in the figure. (Native environment: Apple M1, macOS 13.5. 2)
   
3. After building the release product package on the build machine, use instruments-App Launch to launch the application. Sometimes the time can reach an astonishing 14.5 seconds! (Build machine environment: Apple M2 macOS 13.5.2), the address here cannot be resolved, although I have symbolized it
   

This is the result of my repeated testing using App Launch many times

Weight	Self Weight	Symbol Name
14.50 s 76.5%	0 s	sentry_init
11.15 s 69.7%	0 s	sentry_init
8.06 s 69.1%	0 s	sentry_init
3.00 s 4.0%	0 s	sentry_init
2.34 s 29.9%	0 s	sentry_init
2.26 s 3.8%	0 s	sentry_init

4. Manually click to pull up the release package that build machine had build. The startup speed of sentry is normal, as shown in the log.
   sentry init duration: 1602 microseconds
   sentry init duration: 989 microseconds
   sentry init duration: 367 microseconds
   sentry init duration: 356 microseconds
   sentry init duration: 332 microseconds
   sentry init duration: 342 microseconds
   sentry init duration: 342 microseconds

Through the above tests, I think this should not affect the normal startup speed of online users, but it is not clear why instruments-App Launch causes this result.