module ObjectSpace

static VALUE
allocation_class_path(VALUE self, VALUE obj)
{
    struct allocation_info *info = lookup_allocation_info(obj);

    if (info && info->class_path) {
        return rb_str_new2(info->class_path);
    }
    else {
        return Qnil;
    }
}

Returns the class for the given object.

class A
  def foo
    ObjectSpace::trace_object_allocations do
      obj = Object.new
      p "#{ObjectSpace::allocation_class_path(obj)}"
    end
  end
end

A.new.foo #=> "Class"

See ::trace_object_allocations for more information and examples.

static VALUE
allocation_generation(VALUE self, VALUE obj)
{
    struct allocation_info *info = lookup_allocation_info(obj);
    if (info) {
        return SIZET2NUM(info->generation);
    }
    else {
        return Qnil;
    }
}

Returns garbage collector generation for the given object.

class B
  include ObjectSpace

  def foo
    trace_object_allocations do
      obj = Object.new
      p "Generation is #{allocation_generation(obj)}"
    end
  end
end

B.new.foo #=> "Generation is 3"

See ::trace_object_allocations for more information and examples.

static VALUE
allocation_method_id(VALUE self, VALUE obj)
{
    struct allocation_info *info = lookup_allocation_info(obj);
    if (info) {
        return info->mid;
    }
    else {
        return Qnil;
    }
}

Returns the method identifier for the given object.

class A
  include ObjectSpace

  def foo
    trace_object_allocations do
      obj = Object.new
      p "#{allocation_class_path(obj)}##{allocation_method_id(obj)}"
    end
  end
end

A.new.foo #=> "Class#new"

See ::trace_object_allocations for more information and examples.

static VALUE
allocation_sourcefile(VALUE self, VALUE obj)
{
    struct allocation_info *info = lookup_allocation_info(obj);

    if (info && info->path) {
        return rb_str_new2(info->path);
    }
    else {
        return Qnil;
    }
}

Returns the source file origin from the given object.

See ::trace_object_allocations for more information and examples.

static VALUE
allocation_sourceline(VALUE self, VALUE obj)
{
    struct allocation_info *info = lookup_allocation_info(obj);

    if (info) {
        return INT2FIX(info->line);
    }
    else {
        return Qnil;
    }
}

Returns the original line from source for from the given object.

See ::trace_object_allocations for more information and examples.

static VALUE
count_imemo_objects(int argc, VALUE *argv, VALUE self)
{
    VALUE hash = setup_hash(argc, argv);

    if (imemo_type_ids[0] == 0) {
#define INIT_IMEMO_TYPE_ID(n) (imemo_type_ids[n] = rb_intern_const(#n))
        INIT_IMEMO_TYPE_ID(imemo_env);
        INIT_IMEMO_TYPE_ID(imemo_cref);
        INIT_IMEMO_TYPE_ID(imemo_svar);
        INIT_IMEMO_TYPE_ID(imemo_throw_data);
        INIT_IMEMO_TYPE_ID(imemo_ifunc);
        INIT_IMEMO_TYPE_ID(imemo_memo);
        INIT_IMEMO_TYPE_ID(imemo_ment);
        INIT_IMEMO_TYPE_ID(imemo_iseq);
        INIT_IMEMO_TYPE_ID(imemo_tmpbuf);
        INIT_IMEMO_TYPE_ID(imemo_ast);
        INIT_IMEMO_TYPE_ID(imemo_parser_strterm);
        INIT_IMEMO_TYPE_ID(imemo_callinfo);
        INIT_IMEMO_TYPE_ID(imemo_callcache);
        INIT_IMEMO_TYPE_ID(imemo_constcache);
#undef INIT_IMEMO_TYPE_ID
    }

    each_object_with_flags(count_imemo_objects_i, (void *)hash);

    return hash;
}

Counts objects for each T_IMEMO type.

This method is only for MRI developers interested in performance and memory usage of Ruby programs.

It returns a hash as:

{:imemo_ifunc=>8,
 :imemo_svar=>7,
 :imemo_cref=>509,
 :imemo_memo=>1,
 :imemo_throw_data=>1}

If the optional argument, result_hash, is given, it is overwritten and returned. This is intended to avoid probe effect.

The contents of the returned hash is implementation specific and may change in the future.

In this version, keys are symbol objects.

This method is only expected to work with C Ruby.

static VALUE
count_nodes(int argc, VALUE *argv, VALUE os)
{
    return setup_hash(argc, argv);
}

Counts nodes for each node type.

This method is only for MRI developers interested in performance and memory usage of Ruby programs.

It returns a hash as:

{:NODE_METHOD=>2027, :NODE_FBODY=>1927, :NODE_CFUNC=>1798, ...}

If the optional argument, result_hash, is given, it is overwritten and returned. This is intended to avoid probe effect.

Note: The contents of the returned hash is implementation defined. It may be changed in future.

This method is only expected to work with C Ruby.

static VALUE
count_objects(int argc, VALUE *argv, VALUE os)
{
    struct count_objects_data data = { 0 };
    VALUE hash = Qnil;

    if (rb_check_arity(argc, 0, 1) == 1) {
        hash = argv[0];
        if (!RB_TYPE_P(hash, T_HASH))
            rb_raise(rb_eTypeError, "non-hash given");
    }

    rb_gc_impl_each_object(rb_gc_get_objspace(), count_objects_i, &data);

    if (NIL_P(hash)) {
        hash = rb_hash_new();
    }
    else if (!RHASH_EMPTY_P(hash)) {
        rb_hash_stlike_foreach(hash, set_zero, hash);
    }
    rb_hash_aset(hash, ID2SYM(rb_intern("TOTAL")), SIZET2NUM(data.total));
    rb_hash_aset(hash, ID2SYM(rb_intern("FREE")), SIZET2NUM(data.freed));

    for (size_t i = 0; i <= T_MASK; i++) {
        VALUE type = type_sym(i);
        if (data.counts[i])
            rb_hash_aset(hash, type, SIZET2NUM(data.counts[i]));
    }

    return hash;
}

Counts all objects grouped by type.

It returns a hash, such as:

{
  :TOTAL=>10000,
  :FREE=>3011,
  :T_OBJECT=>6,
  :T_CLASS=>404,
  # ...
}

The contents of the returned hash are implementation specific. It may be changed in future.

The keys starting with :T_ means live objects. For example, :T_ARRAY is the number of arrays. :FREE means object slots which is not used now. :TOTAL means sum of above.

If the optional argument result_hash is given, it is overwritten and returned. This is intended to avoid probe effect.

h = {}
ObjectSpace.count_objects(h)
puts h
# => { :TOTAL=>10000, :T_CLASS=>158280, :T_MODULE=>20672, :T_STRING=>527249 }

This method is only expected to work on C Ruby.

static VALUE
count_objects_size(int argc, VALUE *argv, VALUE os)
{
    size_t counts[T_MASK+1];
    size_t total = 0;
    enum ruby_value_type i;
    VALUE hash = setup_hash(argc, argv);

    for (i = 0; i <= T_MASK; i++) {
        counts[i] = 0;
    }

    each_object_with_flags(cos_i, &counts[0]);

    for (i = 0; i <= T_MASK; i++) {
        if (counts[i]) {
            VALUE type = type2sym(i);
            total += counts[i];
            rb_hash_aset(hash, type, SIZET2NUM(counts[i]));
        }
    }
    rb_hash_aset(hash, ID2SYM(rb_intern("TOTAL")), SIZET2NUM(total));
    return hash;
}

Counts objects size (in bytes) for each type.

Note that this information is incomplete. You need to deal with this information as only a HINT. Especially, total size of T_DATA may be wrong.

It returns a hash as:

{:TOTAL=>1461154, :T_CLASS=>158280, :T_MODULE=>20672, :T_STRING=>527249, ...}

If the optional argument, result_hash, is given, it is overwritten and returned. This is intended to avoid probe effect.

The contents of the returned hash is implementation defined. It may be changed in future.

This method is only expected to work with C Ruby.

static VALUE
count_symbols(int argc, VALUE *argv, VALUE os)
{
    struct dynamic_symbol_counts dynamic_counts = {0, 0};
    VALUE hash = setup_hash(argc, argv);

    size_t immortal_symbols = rb_sym_immortal_count();
    each_object_with_flags(cs_i, &dynamic_counts);

    rb_hash_aset(hash, ID2SYM(rb_intern("mortal_dynamic_symbol")),   SIZET2NUM(dynamic_counts.mortal));
    rb_hash_aset(hash, ID2SYM(rb_intern("immortal_dynamic_symbol")), SIZET2NUM(dynamic_counts.immortal));
    rb_hash_aset(hash, ID2SYM(rb_intern("immortal_static_symbol")),  SIZET2NUM(immortal_symbols - dynamic_counts.immortal));
    rb_hash_aset(hash, ID2SYM(rb_intern("immortal_symbol")),         SIZET2NUM(immortal_symbols));

    return hash;
}

Counts symbols for each Symbol type.

This method is only for MRI developers interested in performance and memory usage of Ruby programs.

If the optional argument, result_hash, is given, it is overwritten and returned. This is intended to avoid probe effect.

Note: The contents of the returned hash is implementation defined. It may be changed in future.

This method is only expected to work with C Ruby.

On this version of MRI, they have 3 types of Symbols (and 1 total counts).

* mortal_dynamic_symbol: GC target symbols (collected by GC)
* immortal_dynamic_symbol: Immortal symbols promoted from dynamic symbols (do not collected by GC)
* immortal_static_symbol: Immortal symbols (do not collected by GC)
* immortal_symbol: total immortal symbols (immortal_dynamic_symbol+immortal_static_symbol)

static VALUE
count_tdata_objects(int argc, VALUE *argv, VALUE self)
{
    VALUE hash = setup_hash(argc, argv);
    each_object_with_flags(cto_i, (void *)hash);
    return hash;
}

Counts objects for each T_DATA type.

This method is only for MRI developers interested in performance and memory usage of Ruby programs.

It returns a hash as:

{RubyVM::InstructionSequence=>504, :parser=>5, :barrier=>6,
 :mutex=>6, Proc=>60, RubyVM::Env=>57, Mutex=>1, Encoding=>99,
 ThreadGroup=>1, Binding=>1, Thread=>1, RubyVM=>1, :iseq=>1,
 Random=>1, ARGF.class=>1, Data=>1, :autoload=>3, Time=>2}
# T_DATA objects existing at startup on r32276.

If the optional argument, result_hash, is given, it is overwritten and returned. This is intended to avoid probe effect.

The contents of the returned hash is implementation specific and may change in the future.

In this version, keys are Class object or Symbol object.

If object is kind of normal (accessible) object, the key is Class object. If object is not a kind of normal (internal) object, the key is symbol name, registered by rb_data_type_struct.

This method is only expected to work with C Ruby.

static VALUE
define_final(int argc, VALUE *argv, VALUE os)
{
    VALUE obj, block;

    rb_scan_args(argc, argv, "11", &obj, &block);
    if (argc == 1) {
        block = rb_block_proc();
    }

    if (rb_callable_receiver(block) == obj) {
        rb_warn("finalizer references object to be finalized");
    }

    return rb_define_finalizer(obj, block);
}

Adds aProc as a finalizer, to be called after obj was destroyed. The object ID of the obj will be passed as an argument to aProc. If aProc is a lambda or method, make sure it can be called with a single argument.

The return value is an array [0, aProc].

The two recommended patterns are to either create the finaliser proc in a non-instance method where it can safely capture the needed state, or to use a custom callable object that stores the needed state explicitly as instance variables.

class Foo
  def initialize(data_needed_for_finalization)
    ObjectSpace.define_finalizer(self, self.class.create_finalizer(data_needed_for_finalization))
  end

  def self.create_finalizer(data_needed_for_finalization)
    proc {
      puts "finalizing #{data_needed_for_finalization}"
    }
  end
end

class Bar
 class Remover
    def initialize(data_needed_for_finalization)
      @data_needed_for_finalization = data_needed_for_finalization
    end

    def call(id)
      puts "finalizing #{@data_needed_for_finalization}"
    end
  end

  def initialize(data_needed_for_finalization)
    ObjectSpace.define_finalizer(self, Remover.new(data_needed_for_finalization))
  end
end

Note that if your finalizer references the object to be finalized it will never be run on GC, although it will still be run at exit. You will get a warning if you capture the object to be finalized as the receiver of the finalizer.

class CapturesSelf
  def initialize(name)
    ObjectSpace.define_finalizer(self, proc {
      # this finalizer will only be run on exit
      puts "finalizing #{name}"
    })
  end
end

Also note that finalization can be unpredictable and is never guaranteed to be run except on exit.

static VALUE
os_each_obj(int argc, VALUE *argv, VALUE os)
{
    VALUE of;

    of = (!rb_check_arity(argc, 0, 1) ? 0 : argv[0]);
    RETURN_ENUMERATOR(os, 1, &of);
    return os_obj_of(of);
}

Calls the block once for each living, nonimmediate object in this Ruby process. If module is specified, calls the block for only those classes or modules that match (or are a subclass of) module. Returns the number of objects found. Immediate objects (Fixnums, Symbols true, false, and nil) are never returned. In the example below, each_object returns both the numbers we defined and several constants defined in the Math module.

If no block is given, an enumerator is returned instead.

a = 102.7
b = 95       # Won't be returned
c = 12345678987654321
count = ObjectSpace.each_object(Numeric) {|x| p x }
puts "Total count: #{count}"

produces:

12345678987654321
102.7
2.71828182845905
3.14159265358979
2.22044604925031e-16
1.7976931348623157e+308
2.2250738585072e-308
Total count: 7

# File gc.rb, line 378
def garbage_collect full_mark: true, immediate_mark: true, immediate_sweep: true
  Primitive.gc_start_internal full_mark, immediate_mark, immediate_sweep, false
end

Alias of GC.start

static VALUE
objspace_internal_class_of(VALUE self, VALUE obj)
{
    VALUE klass;

    if (rb_typeddata_is_kind_of(obj, &iow_data_type)) {
        obj = (VALUE)DATA_PTR(obj);
    }

    if (RB_TYPE_P(obj, T_IMEMO)) {
        return Qnil;
    }
    else {
        klass = CLASS_OF(obj);
        return wrap_klass_iow(klass);
    }
}

MRI specific feature

Return internal class of obj.

obj can be an instance of InternalObjectWrapper.

Note that you should not use this method in your application.

static VALUE
objspace_internal_super_of(VALUE self, VALUE obj)
{
    VALUE super;

    if (rb_typeddata_is_kind_of(obj, &iow_data_type)) {
        obj = (VALUE)DATA_PTR(obj);
    }

    switch (OBJ_BUILTIN_TYPE(obj)) {
      case T_MODULE:
      case T_CLASS:
      case T_ICLASS:
        super = RCLASS_SUPER(obj);
        break;
      default:
        rb_raise(rb_eArgError, "class or module is expected");
    }

    return wrap_klass_iow(super);
}

MRI specific feature

Return internal super class of cls (Class or Module).

obj can be an instance of InternalObjectWrapper.

Note that you should not use this method in your application.

static VALUE
memsize_of_m(VALUE self, VALUE obj)
{
    return SIZET2NUM(rb_obj_memsize_of(obj));
}

Return consuming memory size of obj in bytes.

Note that the return size is incomplete. You need to deal with this information as only a HINT. Especially, the size of T_DATA may not be correct.

This method is only expected to work with C Ruby.

From Ruby 2.2, memsize_of(obj) returns a memory size includes sizeof(RVALUE).

static VALUE
memsize_of_all_m(int argc, VALUE *argv, VALUE self)
{
    struct total_data data = {0, 0};

    if (argc > 0) {
        rb_scan_args(argc, argv, "01", &data.klass);
    }

    each_object_with_flags(total_i, &data);
    return SIZET2NUM(data.total);
}

Return consuming memory size of all living objects in bytes.

If klass (should be Class object) is given, return the total memory size of instances of the given class.

Note that the returned size is incomplete. You need to deal with this information as only a HINT. Especially, the size of T_DATA may not be correct.

Note that this method does NOT return total malloc’ed memory size.

This method can be defined by the following Ruby code:

def memsize_of_all klass = false
  total = 0
  ObjectSpace.each_object{|e|
    total += ObjectSpace.memsize_of(e) if klass == false || e.kind_of?(klass)
  }
  total
end

This method is only expected to work with C Ruby.

static VALUE
reachable_objects_from(VALUE self, VALUE obj)
{
    if (!RB_SPECIAL_CONST_P(obj)) {
        struct rof_data data;

        if (rb_typeddata_is_kind_of(obj, &iow_data_type)) {
            obj = (VALUE)DATA_PTR(obj);
        }

        data.refs = rb_obj_hide(rb_ident_hash_new());
        data.values = rb_ary_new();

        rb_objspace_reachable_objects_from(obj, reachable_object_from_i, &data);

        return data.values;
    }
    else {
        return Qnil;
    }
}

MRI specific feature

Return all reachable objects from ‘obj’.

This method returns all reachable objects from ‘obj’.

If ‘obj’ has two or more references to the same object ‘x’, then returned array only includes one ‘x’ object.

If ‘obj’ is a non-markable (non-heap management) object such as true, false, nil, symbols and Fixnums (and Flonum) then it simply returns nil.

If ‘obj’ has references to an internal object, then it returns instances of ObjectSpace::InternalObjectWrapper class. This object contains a reference to an internal object and you can check the type of internal object with ‘type’ method.

If ‘obj’ is instance of ObjectSpace::InternalObjectWrapper class, then this method returns all reachable object from an internal object, which is pointed by ‘obj’.

With this method, you can find memory leaks.

This method is only expected to work with C Ruby.

Example:

ObjectSpace.reachable_objects_from(['a', 'b', 'c'])
#=> [Array, 'a', 'b', 'c']

ObjectSpace.reachable_objects_from(['a', 'a', 'a'])
#=> [Array, 'a', 'a', 'a'] # all 'a' strings have different object id

ObjectSpace.reachable_objects_from([v = 'a', v, v])
#=> [Array, 'a']

ObjectSpace.reachable_objects_from(1)
#=> nil # 1 is not markable (heap managed) object

static VALUE
reachable_objects_from_root(VALUE self)
{
    struct rofr_data data;
    VALUE hash = data.categories = rb_ident_hash_new();
    data.last_category = 0;

    rb_objspace_reachable_objects_from_root(reachable_object_from_root_i, &data);
    rb_hash_foreach(hash, collect_values_of_values, hash);

    return hash;
}

MRI specific feature

Return all reachable objects from root.

static VALUE
trace_object_allocations(VALUE self)
{
    trace_object_allocations_start(self);
    return rb_ensure(rb_yield, Qnil, trace_object_allocations_stop, self);
}

Starts tracing object allocations from the ObjectSpace extension module.

For example:

require 'objspace'

class C
  include ObjectSpace

  def foo
    trace_object_allocations do
      obj = Object.new
      p "#{allocation_sourcefile(obj)}:#{allocation_sourceline(obj)}"
    end
  end
end

C.new.foo #=> "objtrace.rb:8"

This example has included the ObjectSpace module to make it easier to read, but you can also use the ::trace_object_allocations notation (recommended).

Note that this feature introduces a huge performance decrease and huge memory consumption.

static VALUE
trace_object_allocations_clear(VALUE self)
{
    struct traceobj_arg *arg = get_traceobj_arg();

    /* clear tables */
    st_foreach(arg->object_table, free_values_i, 0);
    st_clear(arg->object_table);
    st_foreach(arg->str_table, free_keys_i, 0);
    st_clear(arg->str_table);

    /* do not touch TracePoints */

    return Qnil;
}

Clear recorded tracing information.

static VALUE
trace_object_allocations_start(VALUE self)
{
    struct traceobj_arg *arg = get_traceobj_arg();

    if (arg->running++ > 0) {
        /* do nothing */
    }
    else {
        if (arg->newobj_trace == 0) {
            arg->newobj_trace = rb_tracepoint_new(0, RUBY_INTERNAL_EVENT_NEWOBJ, newobj_i, arg);
            arg->freeobj_trace = rb_tracepoint_new(0, RUBY_INTERNAL_EVENT_FREEOBJ, freeobj_i, arg);
        }
        rb_tracepoint_enable(arg->newobj_trace);
        rb_tracepoint_enable(arg->freeobj_trace);
    }

    return Qnil;
}

Starts tracing object allocations.

static VALUE
trace_object_allocations_stop(VALUE self)
{
    struct traceobj_arg *arg = get_traceobj_arg();

    if (arg->running > 0) {
        arg->running--;
    }

    if (arg->running == 0) {
        if (arg->newobj_trace != 0) {
            rb_tracepoint_disable(arg->newobj_trace);
        }
        if (arg->freeobj_trace != 0) {
            rb_tracepoint_disable(arg->freeobj_trace);
        }
    }

    return Qnil;
}

Stop tracing object allocations.

Note that if ::trace_object_allocations_start is called n-times, then tracing will stop after calling ::trace_object_allocations_stop n-times.

static VALUE
undefine_final(VALUE os, VALUE obj)
{
    rb_check_frozen(obj);

    rb_gc_impl_undefine_finalizer(rb_gc_get_objspace(), obj);

    return obj;
}

Removes all finalizers for obj.

# File ext/objspace/lib/objspace.rb, line 28
def dump(obj, output: :string)
  out = case output
  when :file, nil
    require 'tempfile'
    Tempfile.create(%w(rubyobj .json))
  when :stdout
    STDOUT
  when :string
    +''
  when IO
    output
  else
    raise ArgumentError, "wrong output option: #{output.inspect}"
  end

  ret = _dump(obj, out)
  return nil if output == :stdout
  ret
end

Dump the contents of a ruby object as JSON.

output can be one of: :stdout, :file, :string, or IO object.

• :file means dumping to a tempfile and returning corresponding File object;

• :stdout means printing the dump and returning nil;

• :string means returning a string with the dump;

• if an instance of IO object is provided, the output goes there, and the object is returned.

This method is only expected to work with C Ruby. This is an experimental method and is subject to change. In particular, the function signature and output format are not guaranteed to be compatible in future versions of ruby.

# File ext/objspace/lib/objspace.rb, line 84
def dump_all(output: :file, full: false, since: nil, shapes: true)
  out = case output
  when :file, nil
    require 'tempfile'
    Tempfile.create(%w(rubyheap .json))
  when :stdout
    STDOUT
  when :string
    +''
  when IO
    output
  else
    raise ArgumentError, "wrong output option: #{output.inspect}"
  end

  shapes = 0 if shapes == true
  ret = _dump_all(out, full, since, shapes)
  return nil if output == :stdout
  ret
end

Dump the contents of the ruby heap as JSON.

output argument is the same as for dump.

full must be a boolean. If true, all heap slots are dumped including the empty ones (T_NONE).

since must be a non-negative integer or nil.

If since is a positive integer, only objects of that generation and newer generations are dumped. The current generation can be accessed using GC::count. Objects that were allocated without object allocation tracing enabled are ignored. See ::trace_object_allocations for more information and examples.

If since is omitted or is nil, all objects are dumped.

shapes must be a boolean or a non-negative integer.

If shapes is a positive integer, only shapes newer than the provided shape id are dumped. The current shape_id can be accessed using RubyVM.stat(:next_shape_id).

If shapes is false, no shapes are dumped.

To only dump objects allocated past a certain point you can combine since and shapes:

ObjectSpace.trace_object_allocations
GC.start
gc_generation = GC.count
shape_generation = RubyVM.stat(:next_shape_id)
call_method_to_instrument
ObjectSpace.dump_all(since: gc_generation, shapes: shape_generation)

This method is only expected to work with C Ruby. This is an experimental method and is subject to change. In particular, the function signature and output format are not guaranteed to be compatible in future versions of ruby.

# File ext/objspace/lib/objspace.rb, line 116
def dump_shapes(output: :file, since: 0)
  out = case output
  when :file, nil
    require 'tempfile'
    Tempfile.create(%w(rubyshapes .json))
  when :stdout
    STDOUT
  when :string
    +''
  when IO
    output
  else
    raise ArgumentError, "wrong output option: #{output.inspect}"
  end

  ret = _dump_shapes(out, since)
  return nil if output == :stdout
  ret
end

Dump the contents of the ruby shape tree as JSON.

output argument is the same as for dump.

If since is a positive integer, only shapes newer than the provided shape id are dumped. The current shape_id can be accessed using RubyVM.stat(:next_shape_id).

This method is only expected to work with C Ruby. This is an experimental method and is subject to change. In particular, the function signature and output format are not guaranteed to be compatible in future versions of ruby.

# File gc.rb, line 378
def garbage_collect full_mark: true, immediate_mark: true, immediate_sweep: true
  Primitive.gc_start_internal full_mark, immediate_mark, immediate_sweep, false
end

Alias of GC.start