Also wondering why tests aren't formatted with black like everything else is now?

I'm not able to reproduce this problem (but I do see that it fails sometimes on Travis CI). What arch/OS can you reproduce it on?

It'd be good to try and fix this properly (if possible) rather than adjust the test.

I can reproduce the issue using the unix coverage variant on Mac. (This can also be seen in #5773 where that build is added to CI).

Based on the discussion in #2927, it sounds like having too small of a stack can lead to this error and debug builds (and therefore coverage builds) require a much bigger stack. It is the test itself that is setting a 2K stack, so I don't think the problem is elsewhere.

If there was a way to detect coverage builds at runtime, then I suppose we could make a better test condition, but I think what I currently have here seems reasonable enough.

I can reproduce the issue using the unix coverage variant on Mac.

Ok, thanks for the hint, I can now also reproduce it on Mac. And I have a different fix: the problem seems to be that on Mac the minimum stack size allowed by pthreads (the variable ``) is much less than on Linux, and so the subtraction of 8192 bytes (to give room to detect a stack overflow) takes the stack size down to 0. To fix this I did:

--- a/ports/unix/mpthreadport.c
+++ b/ports/unix/mpthreadport.c
@@ -221,7 +221,11 @@ void mp_thread_create(void *(*entry)(void *), void *arg, size_t *stack_size) {
 
     // adjust stack_size to provide room to recover from hitting the limit
     // this value seems to be about right for both 32-bit and 64-bit builds
-    *stack_size -= 8192;
+    if (*stack_size >= 2 * 8192) {
+        *stack_size -= 8192;
+    } else {
+        *stack_size /= 2;
+    }
 
     // add thread to linked list of all threads
     thread_t *th = malloc(sizeof(thread_t));

And this seems to get all thread tests passing on Mac.

Thanks for the insight. I will update the PR.

updated

Thanks a lot! Merged in 5e6cee0