I was still having troubles with lldb accepting breakpoints [1]; the breakpoints were just not being hit. Thinking about the issue, it seemed to me that the code might be requesting a too large amount of memory and an easy eay to do that is to track all allocations made by Lua. It was easy enough to test:
static void *l_alloc(void *ud,void *ptr,size_t osize,size_t nsize)
{
(void)ud;
fprintf(stderr,"osize=%zu nsize=%zu\n",osize,nsize);
if (nsize == 0)
{
free(ptr);
return NULL;
}
else
return realloc(ptr,nsize);
}
int main(int argc,char *argv[])
{
/* ... */
L = lua_newstate(l_alloc,NULL);
/* ... */
return 0;
}
Then it was a simple matter of running the program and check each allocation:
... osize=0 nsize=64 osize=0 nsize=40 osize=0 nsize=64 osize=0 nsize=0 osize=0 nsize=64 osize=0 nsize=16 osize=0 nsize=64 osize=0 nsize=0 osize=0 nsize=160 osize=0 nsize=148 osize=0 nsize=64 osize=0 nsize=16 osize=16 nsize=32 osize=0 nsize=16 osize=0 nsize=64 osize=0 nsize=18446744073709551600 ...
Hmmmmm [2] …
Okay, force a core dump we can examine:
static void *l_alloc(void *ud,void *ptr,size_t osize,size_t nsize)
{
(void)ud;
(void)osize;
if (nsize > 10uL * 1024uL * 1024uL)
abort(); /* dump core! */
if (nsize == 0)
{
free(ptr);
return NULL;
}
else
return realloc(ptr,nsize);
}
And the problem is immedately resolved.
The Linux version used epoll() [3] for events, while for whatever reason I don't remember, the Mac OS-X version used select() [4] [Yes, I know that's a Linux reference and not a Mac OS-X reference, but the call is the same between the two systems, and I couldn't find a version of the Mac OS-X page online. —Sean], and that code path was … a bit problematic.
The fix was easy—use poll() [5] for events. With that change, the code worked fine on Mac OS-X (not that we use Mac OS-X in production, but it makes it easy to test in development if it works on the Mac).
[2] https://www.youtube.com/watch?v=gCxrkl2igGY&t=32s
[3] http://man7.org/linux/man-pages/man7/epoll.7.html
[4] http://man7.org/linux/man-pages/man2/select.2.html