123456789_123456789_123456789_123456789_123456789_

Module: ObjectSpace

Relationships & Source Files
Namespace Children
Classes:
Defined in: gc.c,
gc.rb,
weakmap.c

Overview

The ObjectSpace module contains a number of routines that interact with the garbage collection facility and allow you to traverse all living objects with an iterator.

ObjectSpace also provides support for object finalizers, procs that will be called when a specific object is about to be destroyed by garbage collection. See the documentation for .define_finalizer for important information on how to use this method correctly.

a = "A"
b = "B"

ObjectSpace.define_finalizer(a, proc {|id| puts "Finalizer one on #{id}" })
ObjectSpace.define_finalizer(b, proc {|id| puts "Finalizer two on #{id}" })

a = nil
b = nil

produces:

Finalizer two on 537763470
Finalizer one on 537763480

Class Method Summary

Class Method Details

._id2ref(objid) (mod_func)

This method is for internal use only.
[ GitHub ]

  
# File 'gc.c', line 4876

static VALUE
os_id2ref(VALUE os, VALUE objid)
{
    return id2ref(objid);
}

.count_objects([result_hash]) ⇒ Hash (mod_func)

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.

[ GitHub ]

  
# File 'gc.c', line 5239

static VALUE
count_objects(int argc, VALUE *argv, VALUE os)
{
    rb_objspace_t *objspace = &rb_objspace;
    size_t counts[T_MASK+1];
    size_t freed = 0;
    size_t total = 0;
    size_t i;
    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");
    }

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

    for (i = 0; i < heap_allocated_pages; i++) {
        struct heap_page *page = heap_pages_sorted[i];
        short stride = page->slot_size;

        uintptr_t p = (uintptr_t)page->start;
        uintptr_t pend = p + page->total_slots * stride;
        for (;p < pend; p += stride) {
            VALUE vp = (VALUE)p;
            GC_ASSERT((NUM_IN_PAGE(vp) * BASE_SLOT_SIZE) % page->slot_size == 0);

            void *poisoned = asan_unpoison_object_temporary(vp);
            if (RANY(p)->as.basic.flags) {
                counts[BUILTIN_TYPE(vp)]++;
            }
            else {
                freed++;
            }
            if (poisoned) {
                GC_ASSERT(BUILTIN_TYPE(vp) == T_NONE);
                asan_poison_object(vp);
            }
        }
        total += page->total_slots;
    }

    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(total));
    rb_hash_aset(hash, ID2SYM(rb_intern("FREE")), SIZET2NUM(freed));

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

    return hash;
}

.define_finalizer(obj, aProc = proc()) (mod_func)

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.

[ GitHub ]

  
# File 'gc.c', line 4363

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

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

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

    return rb_define_finalizer_no_check(obj, block);
}

.each_object([module]) {|obj| ... } ⇒ Integer (mod_func) .each_object([module]) ⇒ Enumerator

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
[ GitHub ]

  
# File 'gc.c', line 4208

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);
}

.garbage_collect(full_mark: true, immediate_mark: true, immediate_sweep: true) (mod_func)

Alias of GC.start

[ GitHub ]

  
# File 'gc.rb', line 327

def garbage_collect full_mark: true, immediate_mark: true, immediate_sweep: true
  Primitive.gc_start_internal full_mark, immediate_mark, immediate_sweep, false
end

.undefine_finalizer(obj) (mod_func)

Removes all finalizers for obj.

[ GitHub ]

  
# File 'gc.c', line 4226

static VALUE
undefine_final(VALUE os, VALUE obj)
{
    return rb_undefine_finalizer(obj);
}