Description

TL;DR: When using doctest with ESP32, either

• #define DOCTEST_THREAD_LOCAL to disable thread local storage,
  or, alternatively
• #define DOCTEST_CONFIG_NO_MULTITHREADING, which is safer and more efficient if you don't use assertions across multiple threads.

Background

• I'm using doctest with PlatformIO for both native and ESP32 builds. The ESP32 build uses an Arduino-style main, with setup() and empty loop()
• ESP32 has an IPC task for each CPU which is created during startup.
• Before running any code that I can see in the debugger, the ESP32 crashes with this error:

Guru Meditation Error: Core  0 panic'ed (Unhandled debug exception). 
Debug exception reason: Stack canary watchpoint triggered (ipc0) 
Core  0 register dump:
PC      : 0x403799ea  PS      : 0x00060036  A0      : 0x8037a0e4  A1      : 0x3fcf0aa0  
A2      : 0x3fc8e2c4  A3      : 0xffffffff  A4      : 0x00060023  A5      : 0xb33fffff  
A6      : 0x00000000  A7      : 0x00060023  A8      : 0x80379add  A9      : 0x3fc8f320  

While the ESP-IDF supports thread-local storage, using the C++11 thread_local, they come at the cost of increasing the stack usage of all tasks in the system, regardless of whether they use the thread-local variable nor not.

In this case, the thread-locals used by doctest are enough to blow up the stack of the ipc tasks, which have a 4K stack by default.

One possible solution would be to increase the stack size of all tasks to accommodate the thread locals. If you are targeting IDF directly, this could be a workable solution. However, I'm using PlatformIO, and presently it's not possible to reconfigure ESP-IDF at that level, since the IDF is a pre-compiled, and hence a pre-configured library wherein stack sizes for the system tasks are fixed.

Disabling thread local storage, or disabling multithreading in doctest, using the #defines above are the workarounds I found.

This wasn't an easy find :-). I hope this is of use to anyone else using doctest on an architecture that allocates thread locals on the stack of each thread.