Ruby Demystify Ruby Class Variables Lookup

zhengpd · 2020年03月19日 · 最后由 Qwaz1314 回复于 2020年11月07日 · 9528 次阅读


在社区看到一个类变量查询问题的帖子,做了些关于 Ruby 类变量查找的研究。如有错误,欢迎指正。

Demystify Ruby Class Variables Lookup

Zheng Piaodan @ 2020-03-19


A post caught my attention recently. It's at It's about class variables collecting and lookup in Ruby. And I decided to find out what's happening under the hood.

Issue Description

See how class_variables and class_variable_get behave differently in sample:

module Foo; @@foo = 'foo'; end
class Bar; extend Foo; end

# => "2.6.5"
# => []
# => [:@@foo]
# NameError (uninitialized class variable @@foo in #<Class:Bar>)
# => "foo"

Details Investigation

NOTE: all code analyses are based on Ruby 2.6.5 if ruby version not specified.

How extend method works

In short, ruby's extend puts module as direct ancestor to the singleton class of current class, while include puts module as direct ancestor to current class.

In long, I don't have the long story right now...I'll tell you all about it when I dig it again.

How class_variables method works

Let ST = Bar.singleton_class for convenience.

Ruby 2.6.5 behavior

For Ruby 2.6, when Bar.class_variables or ST.class_variablesis called, Ruby will collect class variables defined on:

  • receiver itself , Bar or ST

  • receivers' ancestors , Bar.ancestors or ST.ancestors

Normal class and singleton class behave the same. ST.ancestors includes Foo, so ST.class_variables collects [:@@foo] from Foo.

Ruby 2.7.0 behavior

After Evans' fix, things changed for singleton class. Ruby will collect class variables for ST.class_variables from:

  • singleton class itself, ST

  • instance class of singleton class, Bar, see cvar_front_klass below

  • ancestors of instance class of singleton class, Bar.ancestors

Note that Bar.ancestors doesn't include Foo, so ST.class_variables couldn't find @@foo and returns [].

Take a trip into Evans' code fix

Let's take a deeper look at Evans' fix (section with +):

/* file: variable.c */
  static void*
 mod_cvar_of(VALUE mod, void *data)
     VALUE tmp = mod;
+    if (FL_TEST(mod, FL_SINGLETON)) {
+        if (rb_namespace_p(rb_ivar_get(mod, id__attached__))) {
+            data = mod_cvar_at(tmp, data);
+            tmp = cvar_front_klass(tmp);
+        }
+    }
     for (;;) {
        data = mod_cvar_at(tmp, data);
        tmp = RCLASS_SUPER(tmp);
        if (!tmp) break;
     return data;

The if section was added to use cvar_front_klass of singleton class for later class variables lookup. Here's the cvar_front_klass:

/* file: variable.c */
static VALUE
cvar_front_klass(VALUE klass)
    if (FL_TEST(klass, FL_SINGLETON)) {
        VALUE obj = rb_ivar_get(klass, id__attached__);
        if (rb_namespace_p(obj)) {
            return obj;
    return RCLASS_SUPER(klass);

From the code we can see:

  • If klass is singleton class, cvar_front_klass would return the attached object of the singleton class.

  • For normal class, it just returns ancestor class.

Remember these two behaviors. Below is how Ruby attach object to singleton clss:

/* file: class.c */
 * Attach a object to a singleton class.
 * @pre \a klass is the singleton class of \a obj.
rb_singleton_class_attached(VALUE klass, VALUE obj)
    if (FL_TEST(klass, FL_SINGLETON)) {
        if (!RCLASS_IV_TBL(klass)) {
            RCLASS_IV_TBL(klass) = st_init_numtable();
        rb_class_ivar_set(klass, id_attached, obj);

From the code we can know that the attached object of a singleton class is actually a class itself. According to this, cvar_front_klass for Bar.singleton_class is Bar. That's why ST.class_variables in Ruby 2.7.0 collects class variables from Bar.ancestors not ST.ancestors

How class_variable_get method works

Let ST = Bar.singleton_class for convenience.

Class variable lookup workflow

Lookup steps for class_variable_get:

  1. Look up on receiver itself. Go to next step if not found.

  2. Look up on cvar_front_klass. Front class for Bar is Bar itself. Front class for ST is the object attached to singleton class, which is also Bar.

  3. Look up in ancestors of cvar_front_klass, which is Bar.ancestors.

From step 2, we can know that:

  • In our case ST.class_variable_get(:@@foo) is actually equal to Bar.class_variable_get(:@@foo).

  • And ST.singleton_class.class_variable_get(:@@foo) would look up @@foo on ST and ST's ancestors

Take a trip into class_variable_get method

Let's find out what's inside class_variable_get:

/* file: variable.c */
rb_cvar_get(VALUE klass, ID id)
    VALUE tmp, front = 0, target = 0;
    st_data_t value;

    tmp = klass;
    CVAR_LOOKUP(&value, {if (!front) front = klass; target = klass;});
    if (!target) {
        rb_name_err_raise("uninitialized class variable %1$s in %2$s",
                          tmp, ID2SYM(id));
    cvar_overtaken(front, target, id);
    return (VALUE)value;

The core is CVAR_LOOKUP, which is defined as:

/* file: variable.c */
#define CVAR_LOOKUP(v,r) do {\
    if (cvar_lookup_at(klass, id, (v))) {r;}\
    CVAR_FOREACH_ANCESTORS(klass, v, r);\
} while(0)

It first look up at klass itself, and then in each ancestor of klass. The cvar_lookup_at is a method to find class variable from klass's instance variables table. Think it as a black box with output of a class variable or 0. Right now it's not important in our short trip. Let's see what's inCVAR_FOREACH_ANCESTORS:

/* file: variable.c */
#define CVAR_FOREACH_ANCESTORS(klass, v, r) \
    for (klass = cvar_front_klass(klass); klass; klass = RCLASS_SUPER(klass)) { \
        if (cvar_lookup_at(klass, id, (v))) { \
            r; \
        } \

Got cvar_front_klass again! So if klass here is a singleton class, the for loop would start with instance class of the singleton class, and then iterate through the ancestors of the instance class. That means, when ST.class_variable_get(:@@foo) reaches here, it actually finds @@foo in Bar and Bar.ancestors.

How ST.singleton_class.class_variable_get(:@@foo) works

Let ST = Bar.singleton_class for convenience.

Lookup steps for ST.class_variable_get(:@@foo)

  1. @@foo is not found on ST, then it goes to its front class Bar.

  2. Bar doesn't define @@foo. Go to Bar.ancestors.

  3. Oops! Bar.ancestors doesn't contains Foo module, so we can never found @@foo!

  4. Nothing found. End of story.

Lookup steps for ST.singleton_class.class_variable_get(:@@foo)

  1. @@foo is not found on ST.singleton_class, then it goes to its front class ST.

  2. ST doesn't define @@foo. Go to ST.ancestors.

  3. @@foo is found on Foo, one of ST.ancestors.

  4. Cool! Let's show 'foo', the value of @@foo!

Final Words

The key to the mystery of original issue is the collecting and lookup workflows of class variables. Mystery is no mystery once you know how it happens.

At last, the following code works at Ruby 2.7.0:

# => []  # <= This is what fixed since 2.7.0

# => "foo"  # <= This does not change.



zhengpd 类 extend module 后,类变量的查询问题 提及了此话题。 03月19日 01:30

专业!👍 👍 👍 明白了,2.7 中,单例类的类变量的查询不在 receivers' ancestors 上了,而以前版本的查询又有 bug。



😅 绕到我头疼。一个类 extend 一个模块就相当于它的单例类继承了这个模块

[30] pry(main)> Bar.singleton_class.ancestors
=> [#<Class:Bar>,

所以这个类所对应的单例类的祖先链中会包含 Foo.

在版本 2.6.x 中,通过

[31] pry(main)> Bar.singleton_class.class_variables 
=> [:@@foo]

可以寻找到类变量是因为,系统先从Bar.singleton_class身上找这个变量不得,接着从 Bar.singleton_class的祖先链中去寻找,刚好在Foo身上找到。

而在 2.7.x 版本中修复了这个问题

irb(main):014:0> Bar.singleton_class.class_variables
=> []

系统先在Bar.singleton_class 身上寻找类变量不得,然后跑到Bar去找,依然不得,接下来便寻找Bar的祖先链条。通过 include 可以引入类变量

irb(main):015:1* module A
irb(main):016:1*   @@m = 1000
irb(main):017:0> end
=> 1000
irb(main):018:0> Bar.include(A)
=> Bar
irb(main):019:0> Bar.singleton_class.class_variables
=> [:@@m]

在 2.6.x 中这种方式无效。

lanzhiheng 回复

概括起来其实就三点 (ST = Bar.singleton_class):

  • Bar extend FooFoo 放在了 Bar.singleton_class.ancestors
  • ST.class_variables 在 2.6 遍历 STST.ancestors
  • ST.class_variables 在 2.7 遍历 BarBar.ancestors
jasl 将本帖设为了精华贴。 03月22日 00:33

@Rei 这种评论能删掉吗?😲

9 楼 已删除
10 楼 已删除
11 楼 已删除
需要 登录 后方可回复, 如果你还没有账号请 注册新账号