Hello
Why does GC uses finalizer in the sweep phase? Also what is the
final_list?
Thanks
Tridib
Tridib B. писал 17.09.2012 19:56:
Hello
Why does GC uses finalizer in the sweep phase? Also what is the
final_list?Thanks
Tridib
Judging from my telepathic abilities, you are talking about
Ruby MRI 1.9.x.
GC invokes finalizers in the sweep phase because before the
end of the mark phase you don’t know which objects will remain alive
and which will be collected.
Ruby uses a (somewhat contrived) packed format for storing objects
in memory. The basis of it is the struct RVALUE, defined here:
http://rxr.whitequark.org/mri/source/gc.c#146
The first VALUE-sized field in any Ruby object is always the flags;
these flags signify the type of the object and explain how to decode
the following data. If flags == 0, then the object is already freed and
the second VALUE-sized field points to the next freed object, thus
forming a singly linked list.
It also happens that all of the non-RBasic types extend RBasic, and
the as.free.next field is located at the same place as RBasic’s klass
field:
http://rxr.whitequark.org/mri/source/include/ruby/ruby.h#621
Finalizers are actually run as part of the rb_objspace_call_finalizer
routine:
http://rxr.whitequark.org/mri/source/gc.c#1466
This routine effectively references two different chains of finalizable
objects. First chain, called deferred_final_list is a global one (it’s
expanded to a field in the objspace global object), and the second is
a local variable final_list. They have the same semantics, through.
The first chain is collected via chain_finalized_object (#1427) and
contains plain Ruby objects with finalizers, and the second one is
collected at #1518-1529.
… There could be a nice explanation, but I caught myself on the fact
that I don’t understand why does MRI’s GC have two separate finalizer
list, or why does it repeatedly try to finalize all objects in hopes
that the finalizable list will shrink to zero elements, or why does it
“force-finalize” the rest of objects anyway after rebuilding the list
as a hash table.
I’m probably too stupid, or my telepathy skill is not good enough, or
the comment /* XXX: this loop will make no sense */ precisely describes
the situation and all of that doesn’t make sense.
Why does GC uses finalizer in the sweep phase? Also what is the
final_list?
I think historically it allowed to run finalizers “later” i.e. “after GC
is done” so that code isn’t running in the middle of a GC, when objects
are in a state of being “marked and not” IIRC.