Merge pull request #220 from ThakeeNathees/leak-detect

Memory leak detection implemented

scripts/leak_detect.py

@@ -0,0 +1,75 @@
+##
+## Copyright (c) 2020-2022 Thakee Nathees
+## Copyright (c) 2021-2022 Pocketlang Contributors
+## Distributed Under The MIT License
+##
+
+## A quick script to detect memory leaks, from the trace report.
+## To get the trace report redefine TRACE_MEMORY as 1 at the
+## pk_internal.h and compile pocketlang.
+
+import sys
+
+def detect_leak():
+  trace_log_id = "[memory trace]"
+  
+  total_bytes = 0 ## Totally allocated bytes.
+  mem = dict()    ## key = address, value = bytes.
+
+  trace_log_file_path = sys.argv[1]
+
+  with open(trace_log_file_path, 'r') as f:
+    for line in f.readlines():
+      if line.startswith(trace_log_id):
+        line = line[len(trace_log_id) + 1:].strip()
+        type, data = split_and_strip(line, ':')
+        
+        addr, bytes = split_and_strip(data, '=')
+        bytes = bytes.replace(' bytes', '')
+        
+        if type == "malloc":
+          bytes = int(bytes)
+          assert(bytes >= 0); total_bytes += bytes
+          mem[addr] = bytes
+          
+        elif type == "free":
+          bytes = int(bytes)
+          assert(bytes <= 0); total_bytes += bytes
+          mem.pop(addr)
+        
+        elif type == "realloc":
+          oldp, newp = split_and_strip(addr, '->')
+          olds, news = split_and_strip(bytes, '->')
+          olds = int(olds); news = int(news)
+          total_bytes += (news - olds)
+          assert(mem[oldp] == olds)
+          mem.pop(oldp)
+          mem[newp] = news
+
+  success = True
+  if total_bytes != 0:
+    print_err(f"Memory leak detected - {total_bytes} bytes were never freed.")
+    success = False
+    
+  if len(mem) != 0:
+    print_err("Memory leak detected - some addresses were never freed.")
+    for addr in mem:
+      print(f"  {addr} : {mem[addr]} bytes")
+    success = False
+  
+  if success:
+    print("No leaks were detected.")
+  else:
+    sys.exit(1)
+
+
+def split_and_strip(string, delim):  
+  return map(lambda s: s.strip(), string.split(delim))
+
+
+def print_err(msg):
+  print("Error:", msg, file=sys.stderr)
+
+  
+if __name__ == "__main__":
+  detect_leak()

src/pk_debug.c

@@ -179,7 +179,7 @@ static void _reportStackFrame(PKVM* vm, CallFrame* frame) {
   // reducing it by 1. But stack overflows are occure before executing
   // any instruction of that function, so the instruction_index possibly
   // be -1 (set it to zero in that case).
-  int instruction_index = frame->ip - fn->fn->opcodes.data - 1;
+  int instruction_index = (int) (frame->ip - fn->fn->opcodes.data) - 1;
   if (instruction_index == -1) instruction_index++;
 
   int line = fn->fn->oplines.data[instruction_index];

src/pk_internal.h

@@ -50,6 +50,10 @@
 // Dump the stack values and the globals.
 #define DUMP_STACK 0
 
+// Trace memory allocations. Enable this and redirect the trace dump to a file
+// and run the script leak_detect.py to check for memory leaks.
+#define TRACE_MEMORY 0
+
 // Nan-Tagging could be disable for debugging/portability purposes. See "var.h"
 // header for more information on Nan-tagging.
 #define VAR_NAN_TAGGING 1
@@ -91,8 +95,18 @@
   ((type*)vmRealloc(vm, NULL, 0, sizeof(type) * (count)))
 
 // Deallocate a pointer allocated by vmRealloc before.
-#define DEALLOCATE(vm, pointer) \
+#define DEALLOCATE(vm, pointer, type) \
-    vmRealloc(vm, pointer, 0, 0)
+  vmRealloc(vm, pointer, sizeof(type), 0)
+
+// Deallocate object of [type] which has a dynamic tail array of type
+// [tail_type] with [count] entries.
+#define DEALLOCATE_DYNAMIC(vm, pointer, type, count, tail_type) \
+  ((type*)vmRealloc(vm, pointer,                                \
+    sizeof(type) + sizeof(tail_type) * (count), 0))
+
+// Deallocate [count] amount of object of [type] array.
+#define DEALLOCATE_ARRAY(vm, pointer, type, count) \
+  ((type*)vmRealloc(vm, pointer, sizeof(type) * (count), 0))
 
 /*****************************************************************************/
 /* REUSABLE INTERNAL MACROS                                                  */

src/pk_public.c

@@ -129,7 +129,7 @@ void pkFreeVM(PKVM* vm) {
   // before freeing the VM.
   ASSERT(vm->handles == NULL, "Not all handles were released.");
 
-  DEALLOCATE(vm, vm);
+  DEALLOCATE(vm, vm, PKVM);
 }
 
 void* pkGetUserData(const PKVM* vm) {
@@ -247,7 +247,7 @@ void pkReleaseHandle(PKVM* vm, PkHandle* handle) {
   if (handle->prev) handle->prev->next = handle->next;
 
   // Free the handle.
-  DEALLOCATE(vm, handle);
+  DEALLOCATE(vm, handle, PkHandle);
 }
 
 PkResult pkCompileModule(PKVM* vm, PkHandle* module_handle, PkStringPtr source,

src/pk_value.c

@@ -847,7 +847,7 @@ static void _mapResize(PKVM* vm, Map* self, uint32_t capacity) {
     _mapInsertEntry(self, old_entries[i].key, old_entries[i].value);
   }
 
-  DEALLOCATE(vm, old_entries);
+  DEALLOCATE_ARRAY(vm, old_entries, MapEntry, old_capacity);
 }
 
 Var mapGet(Map* self, Var key) {
@@ -871,7 +871,7 @@ void mapSet(PKVM* vm, Map* self, Var key, Var value) {
 }
 
 void mapClear(PKVM* vm, Map* self) {
-  DEALLOCATE(vm, self->entries);
+  DEALLOCATE_ARRAY(vm, self->entries, MapEntry, self->capacity);
   self->entries = NULL;
   self->capacity = 0;
   self->count = 0;
@@ -930,60 +930,87 @@ void freeObject(PKVM* vm, Object* self) {
   // will won't be freed here instead they haven't marked at all, and will be
   // removed at the sweeping phase of the garbage collection.
   switch (self->type) {
-    case OBJ_STRING:
+    case OBJ_STRING: {
-      break;
+      String* str = (String*) self;
+      DEALLOCATE_DYNAMIC(vm, str, String, str->capacity, char);
+      return;
+    };
 
-    case OBJ_LIST:
+    case OBJ_LIST: {
       pkVarBufferClear(&(((List*)self)->elements), vm);
-      break;
+      DEALLOCATE(vm, self, List);
+      return;
+    }
 
-    case OBJ_MAP:
+    case OBJ_MAP: {
-      DEALLOCATE(vm, ((Map*)self)->entries);
+      Map* map = (Map*)self;
-      break;
+      DEALLOCATE_ARRAY(vm, map->entries, MapEntry, map->capacity);
+      DEALLOCATE(vm, self, Map);
+      return;
+    }
 
-    case OBJ_RANGE:
+    case OBJ_RANGE: {
-      break;
+      DEALLOCATE(vm, self, Range);
+      return;
+    }
 
     case OBJ_MODULE: {
       Module* module = (Module*)self;
       pkVarBufferClear(&module->globals, vm);
       pkUintBufferClear(&module->global_names, vm);
       pkVarBufferClear(&module->constants, vm);
-    } break;
+      DEALLOCATE(vm, self, Module);
+      return;
+    }
 
     case OBJ_FUNC: {
       Function* func = (Function*)self;
       if (!func->is_native) {
         pkByteBufferClear(&func->fn->opcodes, vm);
         pkUintBufferClear(&func->fn->oplines, vm);
-        DEALLOCATE(vm, func->fn);
+        DEALLOCATE(vm, func->fn, Fn);
       }
-    } break;
+      DEALLOCATE(vm, self, Function);
+      return;
+    };
 
-    case OBJ_CLOSURE:
+    case OBJ_CLOSURE: {
-    case OBJ_UPVALUE:
+      DEALLOCATE_DYNAMIC(vm, self, Closure,
-    break;
+        ((Closure*)self)->fn->upvalue_count, Upvalue*);
+      return;
+    }
+
+    case OBJ_UPVALUE: {
+      DEALLOCATE(vm, self, Upvalue);
+      return;
+    }
 
     case OBJ_FIBER: {
       Fiber* fiber = (Fiber*)self;
-      DEALLOCATE(vm, fiber->stack);
+      DEALLOCATE_ARRAY(vm, fiber->stack, Var, fiber->stack_size);
-      DEALLOCATE(vm, fiber->frames);
+      DEALLOCATE_ARRAY(vm, fiber->frames, CallFrame, fiber->frame_capacity);
-    } break;
+      DEALLOCATE(vm, fiber, Fiber);
+      return;
+    }
 
     case OBJ_CLASS: {
       Class* cls = (Class*)self;
       pkClosureBufferClear(&cls->methods, vm);
-    } break;
+      DEALLOCATE(vm, cls, Class);
+      return;
+    }
 
     case OBJ_INST: {
       Instance* inst = (Instance*)self;
       if (inst->cls->delete_fn != NULL) {
         inst->cls->delete_fn(inst->native);
       }
-    } break;
+      DEALLOCATE(vm, inst, Instance);
+      return;
+    }
   }
 
-  DEALLOCATE(vm, self);
+  UNREACHABLE();
 }
 
 uint32_t moduleAddConstant(PKVM* vm, Module* module, Var value) {

src/pk_vm.c

@@ -22,8 +22,6 @@ PkHandle* vmNewHandle(PKVM* vm, Var value) {
 
 void* vmRealloc(PKVM* vm, void* memory, size_t old_size, size_t new_size) {
 
-  // TODO: Debug trace allocations here.
-
   // Track the total allocated memory of the VM to trigger the GC.
   // if vmRealloc is called for freeing, the old_size would be 0 since
   // deallocated bytes are traced by garbage collector.
@@ -33,7 +31,39 @@ void* vmRealloc(PKVM* vm, void* memory, size_t old_size, size_t new_size) {
     vmCollectGarbage(vm);
   }
 
+#if TRACE_MEMORY
+
+  void* ptr = vm->config.realloc_fn(memory, new_size, vm->config.user_data);
+  do {
+    // Deallocation of the VM itself cannot be traced.
+    if (memory == vm) break;
+    if (memory == NULL && new_size == 0) { //< Just nothing.
+      ASSERT(old_size == 0, OOPS);
+      break;
+    }
+
+    if (old_size == 0 && new_size > 0) { // New allocation.
+        ASSERT(memory == NULL, OOPS);
+        printf("[memory trace] malloc  : %p = %+li bytes\n",
+               ptr, (long) new_size);
+      } else if (new_size == 0) { // Free.
+        ASSERT(memory != NULL && old_size != 0, OOPS);
+        printf("[memory trace] free    : %p = -%li bytes\n",
+               memory, (long) old_size);
+      } else { // Realloc.
+        ASSERT(old_size != 0 && new_size != 0 && memory != NULL, OOPS);
+        printf("[memory trace] realloc : %p -> %p = %+li -> %+li bytes\n",
+               memory, ptr, (long) (old_size), (long) (new_size));
+      }
+
+  } while (false);
+
+  return ptr;
+
+#else
   return vm->config.realloc_fn(memory, new_size, vm->config.user_data);
+#endif
+
 }
 
 void vmPushTempRef(PKVM* vm, Object* obj) {