Merge branch 'dev' into dev2

include/mimalloc.h

@@ -154,17 +154,21 @@ mi_decl_export void mi_stats_reset(void)      mi_attr_noexcept;
 mi_decl_export void mi_stats_merge(void)      mi_attr_noexcept;
 mi_decl_export void mi_stats_print(void* out) mi_attr_noexcept;  // backward compatibility: `out` is ignored and should be NULL
 mi_decl_export void mi_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept;
-mi_decl_export void mi_options_print(void)    mi_attr_noexcept;
-
-mi_decl_export void mi_process_init(void)     mi_attr_noexcept;
-mi_decl_export void mi_thread_init(void)      mi_attr_noexcept;
-mi_decl_export void mi_thread_done(void)      mi_attr_noexcept;
 mi_decl_export void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept;
+mi_decl_export void mi_options_print(void)    mi_attr_noexcept;
 
 mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, size_t* system_msecs,
                                     size_t* current_rss, size_t* peak_rss,
                                     size_t* current_commit, size_t* peak_commit, size_t* page_faults) mi_attr_noexcept;
 
+
+// Generally do not use the following as these are usually called automatically
+mi_decl_export void mi_process_init(void)     mi_attr_noexcept;
+mi_decl_export void mi_cdecl mi_process_done(void) mi_attr_noexcept;
+mi_decl_export void mi_thread_init(void)      mi_attr_noexcept;
+mi_decl_export void mi_thread_done(void)      mi_attr_noexcept;
+
+
 // -------------------------------------------------------------------------------------
 // Aligned allocation
 // Note that `alignment` always follows `size` for consistency with unaligned

include/mimalloc/internal.h

@@ -63,7 +63,7 @@ terms of the MIT license. A copy of the license can be found in the file
 #define mi_decl_noinline
 #define mi_decl_thread          __thread        // hope for the best :-)
 #define mi_decl_align(a)
-#define mi_decl_noreturn        
+#define mi_decl_noreturn
 #define mi_decl_weak
 #define mi_decl_hidden
 #define mi_decl_cold
@@ -135,8 +135,8 @@ static inline uintptr_t _mi_random_shuffle(uintptr_t x);
 // init.c
 extern mi_decl_hidden mi_decl_cache_align mi_stats_t       _mi_stats_main;
 extern mi_decl_hidden mi_decl_cache_align const mi_page_t  _mi_page_empty;
-void        _mi_process_load(void);
-void mi_cdecl _mi_process_done(void);
+void        _mi_auto_process_init(void);
+void mi_cdecl _mi_auto_process_done(void) mi_attr_noexcept;
 bool        _mi_is_redirected(void);
 bool        _mi_allocator_init(const char** message);
 void        _mi_allocator_done(void);

src/init.c

@@ -342,7 +342,7 @@ static mi_thread_data_t* mi_thread_data_zalloc(void) {
       return NULL;
     }
   }
-  td->memid = memid;  
+  td->memid = memid;
   return td;
 }
 
@@ -577,7 +577,7 @@ mi_decl_nodiscard bool mi_is_redirected(void) mi_attr_noexcept {
 }
 
 // Called once by the process loader from `src/prim/prim.c`
-void _mi_process_load(void) {
+void _mi_auto_process_init(void) {
   mi_heap_main_init();
   #if defined(__APPLE__) || defined(MI_TLS_RECURSE_GUARD)
   volatile mi_heap_t* dummy = _mi_heap_default; // access TLS to allocate it before setting tls_initialized to true;
@@ -664,8 +664,8 @@ void mi_process_init(void) mi_attr_noexcept {
   }
 }
 
-// Called when the process is done (through `at_exit`)
-void mi_cdecl _mi_process_done(void) {
+// Called when the process is done (cdecl as it is used with `at_exit` on some platforms)
+void mi_cdecl mi_process_done(void) mi_attr_noexcept {
   // only shutdown if we were initialized
   if (!_mi_process_is_initialized) return;
   // ensure we are called once
@@ -708,3 +708,7 @@ void mi_cdecl _mi_process_done(void) {
   os_preloading = true; // don't call the C runtime anymore
 }
 
+void mi_cdecl _mi_auto_process_done(void) mi_attr_noexcept {
+  if (_mi_option_get_fast(mi_option_destroy_on_exit)>1) return;
+  mi_process_done();
+}

src/prim/prim.c

@@ -39,29 +39,29 @@ terms of the MIT license. A copy of the license can be found in the file
     #define mi_attr_destructor  __attribute__((destructor))
   #endif
   static void mi_attr_constructor mi_process_attach(void) {
-    _mi_process_load();
+    _mi_auto_process_init();
   }
   static void mi_attr_destructor mi_process_detach(void) {
-    _mi_process_done();
+    _mi_auto_process_done();
   }
 #elif defined(__cplusplus)
   // C++: use static initialization to detect process start/end
   // This is not guaranteed to be first/last but the best we can generally do?
   struct mi_init_done_t {
     mi_init_done_t() {
-      _mi_process_load();
+      _mi_auto_process_init();
     }
     ~mi_init_done_t() {
-      _mi_process_done();
+      _mi_auto_process_done();
     }
   };
   static mi_init_done_t mi_init_done;
  #else
-  #pragma message("define a way to call _mi_process_load/done on your platform")
+  #pragma message("define a way to call _mi_auto_process_init/done on your platform")
 #endif
 #endif
 
-// Generic allocator init/done callback 
+// Generic allocator init/done callback
 #ifndef MI_PRIM_HAS_ALLOCATOR_INIT
 bool _mi_is_redirected(void) {
   return false;

src/prim/windows/prim.c

@@ -633,7 +633,7 @@ bool _mi_prim_random_buf(void* buf, size_t buf_len) {
 //----------------------------------------------------------------
 
 #if MI_WIN_USE_FIXED_TLS==1
-mi_decl_cache_align size_t _mi_win_tls_offset = 0;  
+mi_decl_cache_align size_t _mi_win_tls_offset = 0;
 #endif
 
 //static void mi_debug_out(const char* s) {
@@ -654,14 +654,14 @@ static void mi_win_tls_init(DWORD reason) {
     #endif
     #if MI_HAS_TLS_SLOT >= 2  // we must initialize the TLS slot before any allocation
     if (mi_prim_get_default_heap() == NULL) {
-      _mi_heap_set_default_direct((mi_heap_t*)&_mi_heap_empty);  
+      _mi_heap_set_default_direct((mi_heap_t*)&_mi_heap_empty);
       #if MI_DEBUG && MI_WIN_USE_FIXED_TLS==1
       void* const p = TlsGetValue((DWORD)(_mi_win_tls_offset / sizeof(void*)));
       mi_assert_internal(p == (void*)&_mi_heap_empty);
-      #endif  
+      #endif
     }
-    #endif  
-  }  
+    #endif
+  }
 }
 
 static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
@@ -669,10 +669,10 @@ static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
   MI_UNUSED(module);
   mi_win_tls_init(reason);
   if (reason==DLL_PROCESS_ATTACH) {
-    _mi_process_load();
+    _mi_auto_process_init();
   }
   else if (reason==DLL_PROCESS_DETACH) {
-    _mi_process_done();
+    _mi_auto_process_done();
   }
   else if (reason==DLL_THREAD_DETACH && !_mi_is_redirected()) {
     _mi_thread_done(NULL);
@@ -684,7 +684,7 @@ static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
   #define MI_PRIM_HAS_PROCESS_ATTACH  1
 
   // Windows DLL: easy to hook into process_init and thread_done
-  BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) {    
+  BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) {
     mi_win_main((PVOID)inst,reason,reserved);
     return TRUE;
   }
@@ -762,7 +762,7 @@ static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
 
     static int mi_process_attach(void) {
       mi_win_main(NULL,DLL_PROCESS_ATTACH,NULL);
-      atexit(&_mi_process_done);
+      atexit(&_mi_auto_process_done);
       return 0;
     }
     typedef int(*mi_crt_callback_t)(void);