It could be that the 37k block is allocated right in the middle of the heap, and so no contiguous 38k block exists. I've added a function to the GC to print out the organisation of the heap and what is allocated/free. Call gc_dump_alloc_table after pyb_info to see it. It'll be large, but will tell you exactly what is going on.

Note that, for your test case above, the GC can be improved to make it work. At the moment a realloc is just an alloc followed by a memcpy followed by a free. This could be changed to see if the current block can be extended in place. See line 334 of gc.c.

but this is called first thing after running main.py, and nothing else is allocated, or maybe small objects for the strings printed are allocated which cause fragmentation :? anyway, I will try the function and send back the results.

But you are allocating 1k, then 2k, then 3k, ..., 36k before allocating the 37k, so this will fragment the memory quite a bit.

what if I free the memory before I look for a bigger slice, would that work :) ? oh right, need to copy it nevermind

Yes, if you changed realloc to free+alloc, then it should work....

I saw the comment you mentioned in gc.c I want to give this a try, so let's leave this issue open for now...

It should be easy to implement proper realloc: code from gc_nbytes can help determine if there are enough blocks after the current one to satisfy the allocation, and the last few lines of gc_alloc can be used to mark these blocks as used (ATB_FREE_TO_TAIL).

I tried the same test again, and it didn't fail this time:

void test_realloc()
{
    while (!usb_vcp_is_connected());
    printf("\n\ntesting realloc\n");
    void *mem = gc_alloc(1024);
    for (int i=1; i<=50; i++) { /* ~50K */
        mem = gc_realloc(mem, 1024*i);
        printf("realloc: %p %d\n\n", mem, 1024*i);
        if (mem==NULL) {
            pyb_info();
            while(1);
        }
    }
    pyb_info();

    mem = gc_realloc(mem, 1024);
    pyb_info();
    while(1);
}

result after first calls to realloc:

  102528 total
used: 52944 free: 49584
  1=40 2=5 m=3200
LFS free: 95744 bytes

result after realloc'ing down to 1024

GC:
  102528 total
used: 2768 free: 99760
  1=40 2=5 m=64
LFS free: 95744 bytes