Bug #17379: Refinement with modules redefinition bug

• Status: Open
• Priority: Normal
• Assignee: ko1 (Koichi Sasada)
• Target version: -
• ruby -v: ruby 3.0.0dev (2020-12-05T10:40:00Z master 9dbb2bfd73) [x86_64-darwin18]
• Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN

[marcandre (Marc-Andre Lafortune)]

Depending on the circumstance, a refinement can be modified even after being used:

def foo
  [:base]
end

module M
  def foo
    super << :M
  end
end

module Ext
  refine Object do
    include M
  end
end

using Ext

p 'asd'.foo unless ENV['SKIP'] # => [:base, :M] (ok)

module M
  def foo
    super << :new_ref
  end
end

p 'asd'.foo # => depends (not ok)

Running this gives:

$ ruby refinement.rb
[:base, :M]
[:base, :M] # => ok
$ SKIP=t ruby refinement.rb
[:base, :new_ref]  # => should be [:base, :M]

[marcandre (Marc-Andre Lafortune)]

• Subject changed from Refinement with modules redefinition issues to Refinement with modules redefinition bug

[shugo (Shugo Maeda)]

• Assignee changed from shugo (Shugo Maeda) to matz (Yukihiro Matsumoto)

It's an inline method cache issue and hard to solve without performance regression.

If the behavior is not acceptable as a limitation of Refinements, it may be better to prohibit module inclusion in Refinements.

[shugo (Shugo Maeda)]

shugo (Shugo Maeda) wrote in #note-2:

If the behavior is not acceptable as a limitation of Refinements, it may be better to prohibit module inclusion in Refinements.

I think there is no need to prohibit inclusion of frozen modules, at least for this issue.

[shugo (Shugo Maeda)]

• Assignee changed from matz (Yukihiro Matsumoto) to shugo (Shugo Maeda)

shugo (Shugo Maeda) wrote in #note-2:

It's an inline method cache issue and hard to solve without performance regression.

I might be wrong, and will investigate it further.

[marcandre (Marc-Andre Lafortune)]

Probably same bug, without using, found by Daniel DeLorme:

class Foo
  def foo
    p :hello
  end
end

module Code
  def foo
    p :A
  end
end

module Extension
  refine Foo do
    prepend Code
  end
end

Foo.new.foo unless ENV['SKIP'] # => :hello (ok)
Foo.prepend Code
Foo.new.foo     # => depends (not ok)

gives:

$ ruby refinement.rb
:hello
:hello
$ SKIP=t ruby refinement.rb
:A

[shugo (Shugo Maeda)]

marcandre (Marc-Andre Lafortune) wrote in #note-5:

Probably same bug, without using, found by Daniel DeLorme:

It seems a different issue, so I've filed #17386.

[shugo (Shugo Maeda)]

• Assignee changed from shugo (Shugo Maeda) to ko1 (Koichi Sasada)

It seems that callable method entry cache caused the problem.
The problem doesn't occur with the following patch:

diff --git a/vm_method.c b/vm_method.c
index a0ccdb8a51..d3a3926780 100644
--- a/vm_method.c
+++ b/vm_method.c
@@ -1023,7 +1023,7 @@ prepare_callable_method_entry(VALUE defined_class, ID id, const rb_method_entry_
                     mtbl = RCLASS_EXT(defined_class)->callable_m_tbl = rb_id_table_create(0);
                 }
                 cme = rb_method_entry_complement_defined_class(me, me->called_id, defined_class);
-                rb_id_table_insert(mtbl, id, (VALUE)cme);
+                // rb_id_table_insert(mtbl, id, (VALUE)cme);
                 RB_OBJ_WRITTEN(defined_class, Qundef, (VALUE)cme);
                 VM_ASSERT(callable_method_entry_p(cme));
             }
@@ -1122,7 +1122,8 @@ callable_method_entry(VALUE klass, ID mid, VALUE *defined_class_ptr)
     VM_ASSERT(RB_TYPE_P(klass, T_CLASS) || RB_TYPE_P(klass, T_ICLASS));
     RB_VM_LOCK_ENTER();
     {
-        cme = cached_callable_method_entry(klass, mid);
+        // cme = cached_callable_method_entry(klass, mid);
+        cme = NULL;

         if (cme) {
             if (defined_class_ptr != NULL) *defined_class_ptr = cme->defined_class;
@@ -1139,7 +1140,7 @@ callable_method_entry(VALUE klass, ID mid, VALUE *defined_class_ptr)
                 cme = negative_cme(mid);
             }

-            cache_callable_method_entry(klass, mid, cme);
+            // cache_callable_method_entry(klass, mid, cme);
         }
     }
     RB_VM_LOCK_LEAVE();

(The fix of prepare_callable_method_entry() is for Kernel#method)

ko1 (Koichi Sasada), is there any way to fix the problem without performance regression?

[jeremyevans0 (Jeremy Evans)]

• Related to Bug #17429: Prohibit include/prepend in refinement modules added