use subproc_main for final statistics output to avoid dereferencing the heap

include/mimalloc/internal.h

@@ -239,6 +239,7 @@ void          _mi_heap_page_reclaim(mi_heap_t* heap, mi_page_t* page);
 // "stats.c"
 void          _mi_stats_init(void);
 void          _mi_stats_done(mi_stats_t* stats);
+void          _mi_stats_print(mi_stats_t* stats, mi_output_fun* out, void* arg) mi_attr_noexcept;
 void          _mi_stats_merge_thread(mi_tld_t* tld);
 void          _mi_stats_merge_from(mi_stats_t* to, mi_stats_t* from);
 mi_msecs_t    _mi_clock_now(void);

src/init.c

@@ -784,7 +784,7 @@ void mi_cdecl _mi_process_done(void) {
   //_mi_page_map_unsafe_destroy(_mi_subproc_main());
 
   if (mi_option_is_enabled(mi_option_show_stats) || mi_option_is_enabled(mi_option_verbose)) {
-    mi_stats_print(NULL);
+    _mi_stats_print(&_mi_subproc_main()->stats, NULL, NULL);  // use always main subproc at process exit to avoid dereferencing the heap (as it may be destroyed by now)
   }
   _mi_allocator_done();
   _mi_verbose_message("process done: 0x%zx\n", tld_main.thread_id);

src/options.c

@@ -405,8 +405,10 @@ void mi_register_output(mi_output_fun* out, void* arg) mi_attr_noexcept {
 // add stderr to the delayed output after the module is loaded
 static void mi_add_stderr_output(void) {
   mi_assert_internal(mi_out_default == NULL);
-  mi_out_buf_flush(&mi_out_stderr, false, NULL); // flush current contents to stderr
-  mi_out_default = &mi_out_buf_stderr;           // and add stderr to the delayed output
+  if (mi_out_default==NULL) {
+    mi_out_buf_flush(&mi_out_stderr, false, NULL); // flush current contents to stderr
+    mi_out_default = &mi_out_buf_stderr;           // and add stderr to the delayed output
+  }
 }
 
 // --------------------------------------------------------

src/stats.c

@@ -310,7 +310,7 @@ static void mi_cdecl mi_buffered_out(const char* msg, void* arg) {
 // Print statistics
 //------------------------------------------------------------
 
-static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0) mi_attr_noexcept {
+void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0) mi_attr_noexcept {
   // wrap the output function to be line buffered
   char buf[256]; _mi_memzero_var(buf);
   buffered_t buffer = { out0, arg0, NULL, 0, 255 };