Stack Overflow Error: Process Exited with STATUS_STACK_OVERFLOW (0xc00000fd)

[gitctrlx]

Description:

The program encounters a stack overflow error after running a series of episodes, causing it to crash with the following message:

thread 'main' has overflowed its stack
error: process didn't exit successfully: `target\debug\rl-examples.exe` (exit code: 0xc00000fd, STATUS_STACK_OVERFLOW)

This issue typically indicates that the program is consuming more stack memory than is allocated, which could be due to excessive recursion, large stack-allocated variables, or improper thread stack size settings.

---

Steps to Reproduce:

1. Navigate to the project directory:

   cd burn-rl-examples

2. Compile and run the program:

   cargo run

3. Observe the program processing a series of episodes. For example:

   {"episode": 0, "reward": 44.0000, "duration": 44}
   {"episode": 1, "reward": 82.0000, "duration": 82}
   ...
   {"episode": 52, "reward": 87.0000, "duration": 87}

4. The program crashes after episode 52 with the following error:

   thread 'main' has overflowed its stack
   error: process didn't exit successfully: `target\debug\rl-examples.exe` (exit code: 0xc00000fd, STATUS_STACK_OVERFLOW)
   

---

Expected Behavior:

The program should handle memory allocation and recursion (if any) efficiently and execute without crashing due to stack overflow.

---

Observed Behavior:

The program crashes after a specific number of episodes due to stack overflow.

---

Suggested Fixes:

1. Optimize recursive logic or convert to iterative implementations.
2. Allocate large variables on the heap using Box or Vec instead of the stack.
3. Explicitly increase the thread stack size using std::thread::Builder for custom threads.
4. Check for and fix any unintended infinite recursion.

[yunjhongwu]

I can't reproduce this issue on Linux/iOS with either burn 0.15 or 0.16 (rustc 1.84). Not sure if this is a Windows-specific issue, but I don't have a Windows machine to check.

[Dateb]

I think the program allocates a bit too much on the stack here and there. I got a stack overflow on Linux as well. In my case, I used a env with a larger state space (game screens), which was allocated on the stack due to using copyable data (array). After allocating to the heap via a Box, it worked very well!