Lua GC 之 Finalizer

lua 的 __gc元方法，被称为 finalizer 。这里根据其特点结合源码进行分析：

1. 设置 __gc 元方法

要想对象的 __gc 元方法生效，必须在设置对象的元方法的时候元表已经存在 __gc 方法，（即调用 setmetatable 时，元表已经存在__gc方法）。以后即使对元表增加了 __gc 元方法，也不会生效。见下面的代码

mt = {}
obj = setmetatable({}, mt)
mt.__gc = function print "do gc" end
-- obj对象在内存释放前并不会执行__gc元方法。虽然查看obj的元表可以看到__gc元方法

setmetatable(obj, mt)
-- obj对象的__gc元方法，产生作用。会再内存释放前被调用

  原因在于仅仅只有的 lua_setmetatable 才会调用 luaC_checkfinalizer 函数。luaC_checkfinalizer 函数检查表中是否有 __gc 方法，如果有，则将对象从 allgc 链中移到了 finobj 链中（lua 5.2 中 Table 和 Userdata 类型均支持__gc元方法，但 lua5.1 仅 Userdata 类型支持）。其实现如下：

LUA_API int lua_setmetatable (lua_State *L, int objindex) {
  TValue *obj;
  Table *mt;
  lua_lock(L);
  api_checknelems(L, 1);
  obj = index2addr(L, objindex);
  api_checkvalidindex(L, obj);
  if (ttisnil(L->top - 1))
    mt = NULL;
  else {
    api_check(L, ttistable(L->top - 1), "table expected");
    mt = hvalue(L->top - 1);
  }
  switch (ttypenv(obj)) {
    case LUA_TTABLE: {
      hvalue(obj)->metatable = mt;
      if (mt)
        luaC_objbarrierback(L, gcvalue(obj), mt);
        luaC_checkfinalizer(L, gcvalue(obj), mt);
      break;
    }
    case LUA_TUSERDATA: {
      uvalue(obj)->metatable = mt;
      if (mt) {
        luaC_objbarrier(L, rawuvalue(obj), mt);
        luaC_checkfinalizer(L, gcvalue(obj), mt);
      }
      break;
    }
    default: {
      G(L)->mt[ttypenv(obj)] = mt;
      break;
    }
  }
  L->top--;
  lua_unlock(L);
  return 1;
}

/*
** if object 'o' has a finalizer, remove it from 'allgc' list (must
** search the list to find it) and link it in 'finobj' list.
*/
void luaC_checkfinalizer (lua_State *L, GCObject *o, Table *mt) {
  global_State *g = G(L);
  if (testbit(gch(o)->marked, SEPARATED) || /* obj. is already separated... */
      isfinalized(o) ||                           /* ... or is finalized... */
      gfasttm(g, mt, TM_GC) == NULL)                /* or has no finalizer? */
    return;  /* nothing to be done */
  else {  /* move 'o' to 'finobj' list */
    GCObject **p;
    GCheader *ho = gch(o);
    if (g->sweepgc == &ho->next) {  /* avoid removing current sweep object */
      lua_assert(issweepphase(g));
      g->sweepgc = sweeptolive(L, g->sweepgc, NULL);
    }
    /* search for pointer pointing to 'o' */
    for (p = &g->allgc; *p != o; p = &gch(*p)->next) { /* empty */ }
    *p = ho->next;  /* remove 'o' from root list */
    ho->next = g->finobj;  /* link it in list 'finobj' */
    g->finobj = o;
    l_setbit(ho->marked, SEPARATED);  /* mark it as such */
    if (!keepinvariant(g))  /* not keeping invariant? */
      makewhite(g, o);  /* "sweep" object */
    else
      resetoldbit(o);  /* see MOVE OLD rule */
  }
}

 也就是说你单独对 metatable 的修改并不会触发 luaC_checkfinalizer 函数的对象分类操作。

为什么会这样？原因很简单，性能。表的操作太频繁了，所以不能在对表的赋值操作中插入 luaC_checkfinalizer 函数。

2. __gc 元方法执行

__gc 元方法执行的时候，弱引用特点:

     弱键强值表可访问（以对象自身为键）

     弱值表不可访问（以对象自身为值）

见下面代码：

mt = {__gc = function(self)
                print("self as key:", tk[self])
                print("self as value:", tv[self.i])
                print("self as key and value:", tkv[self])
            end
        }

tk = setmetatable({}, {__mode='k'})
tv = setmetatable({}, {__mode='v'})
tkv = setmetatable({}, {__mode='kv'})

obj = setmetatable({}, mt)
obj.i = 1


tk[obj] = 'obj value'
tv[obj.i] = obj
tkv[obj] = obj

collectgarbage()

print '----'

obj = nil


collectgarbage()
print '----'

 输出如下:

----
self as key:    obj value
self as value:  nil
self as key and value:  nil
----

 从上面的输出可以看到 __gc 元方法调用时，仅仅 tk 表中以自身为键的的值还可以访问。为什么仅仅是键可访问，而不是值？主要是基于这点：通常我们用表以该对象为弱键，来保存该对象的备注信息。可参考 PIL 中的 17.2 关联对象属性 和 lua 邮件组中的讨论。

 

筛选可释放对象，实现细节（有删减）：

static l_mem atomic (lua_State *L) {
  global_State *g = G(L);

// 弱表（弱值、弱键弱值）中清除不可达的值
  /* at this point, all strongly accessible objects are marked. */
  /* clear values from weak tables, before checking finalizers */
  clearvalues(g, g->weak, NULL);
  clearvalues(g, g->allweak, NULL);

// 将不可达对象从 finobj 链移入 tobefnz 链
  separatetobefnz(L, 0);  /* separate objects to be finalized */

// 将 tobefnz 链中所有对象及其引用的对象，都标记为可达对象
  markbeingfnz(g);  /* mark objects that will be finalized */
  propagateall(g);  /* remark, to propagate `preserveness' */

// 弱表（弱键、弱键弱值）中清理不可达的 键
  /* at this point, all resurrected objects are marked. */
  /* remove dead objects from weak tables */
  clearkeys(g, g->ephemeron, NULL);  /* clear keys from all ephemeron tables */
  clearkeys(g, g->allweak, NULL);  /* clear keys from all allweak tables */

  return work;  /* estimate of memory marked by 'atomic' */
}

 可以看到， 先清理弱表中的键，再复活待执行对象，这时才开始清理弱表中的值。

g->allweak、g->weak、g->ephemeron 这三个链表存放的是弱表，下篇文章会详细介绍弱表

执行 __gc 元方法，实现细节：

/*
** performs a basic GC step
*/
void luaC_forcestep (lua_State *L) {
  global_State *g = G(L);
  int i;
  if (isgenerational(g)) generationalcollection(L);
  else incstep(L);
// 一次最多执行 GCFINALIZENUM 个对象的 __gc 元方法
  /* run a few finalizers (or all of them at the end of a collect cycle) */ 
  for (i = 0; g->tobefnz && (i < GCFINALIZENUM || g->gcstate == GCSpause); i++)
    GCTM(L, 1);  /* call one finalizer */
}

static void GCTM (lua_State *L, int propagateerrors) {
  global_State *g = G(L);
  const TValue *tm;
  TValue v;
  setgcovalue(L, &v, udata2finalize(g)); // udata2finalize(g) 将对象从 tobefnz 链中移入 allgc 链
  tm = luaT_gettmbyobj(L, &v, TM_GC);
  if (tm != NULL && ttisfunction(tm)) {  /* is there a finalizer? */
    int status;
    lu_byte oldah = L->allowhook;
    int running  = g->gcrunning;
    L->allowhook = 0;  /* stop debug hooks during GC metamethod */
    g->gcrunning = 0;  /* avoid GC steps */
    setobj2s(L, L->top, tm);  /* push finalizer... */
    setobj2s(L, L->top + 1, &v);  /* ... and its argument */
    L->top += 2;  /* and (next line) call the finalizer */
    status = luaD_pcall(L, dothecall, NULL, savestack(L, L->top - 2), 0); // 执行 __gc 元方法
    L->allowhook = oldah;  /* restore hooks */
    g->gcrunning = running;  /* restore state */
    if (status != LUA_OK && propagateerrors) {  /* error while running __gc? */
      if (status == LUA_ERRRUN) {  /* is there an error object? */
        const char *msg = (ttisstring(L->top - 1))
                            ? svalue(L->top - 1)
                            : "no message";
        luaO_pushfstring(L, "error in __gc metamethod (%s)", msg);
        status = LUA_ERRGCMM;  /* error in __gc metamethod */
      }
      luaD_throw(L, status);  /* re-throw error */
    }
  }
}

 

 GCTM 函数会从 tobefnz 链中移出一个对象，并放入 allobj 链，然后执行该对象的 __gc 元方法。

 

 

最后总结下对象的流向：

             lua_setmetable                               separatetobefnz                               GCTM

   allgc --------------------------> finobj ------------------------> tobefnz ---------------------> allgc