From: Yehuda Katz <wycats@...>
Date: 2010-02-28T04:38:30+09:00
Subject: [ruby-core:28364] Small 1.8.7 bug involving proc_alloc

--00c09f88cf33975b3704809a28d4
Content-Type: text/plain; charset=ISO-8859-1

I have found a small 1.8.7 bug that prevents proc_alloc from retrieving the
old Proc object if the Proc has a metaclass. Here's an example:

def foo
  yield
end

def bar(&block)
  puts Proc.new.object_id
  def block.foo() "FOO" end
  puts Proc.new.object_id
  baz(&block)
end

def baz(&block)
  puts Proc.new.object_id
  yield
end

foo { puts "HELLO" }
bar { puts "HELLO" }

Ruby 1.9, Rubinius and JRuby all return the same Proc object in all three
cases. Ruby 1.8.7 returns a different Proc object in each case, because of
the use of CLASS_OF in proc_alloc:

    if (!proc && ruby_block->block_obj && CLASS_OF(ruby_block->block_obj) ==
klass) {
return ruby_block->block_obj;
    }

It seems this code is trying to make sure that MyProc.new doesn't return a
Proc object. The solution is a tiny patch to use rb_class_real:

index 8847ee5..54a0f46 100644
--- a/eval.c
+++ b/eval.c
@@ -8725,7 +8725,7 @@ proc_alloc(klass, proc)
        rb_warn("tried to create Proc object without a block");
     }

-    if (!proc && ruby_block->block_obj && CLASS_OF(ruby_block->block_obj)
== klass) {
+    if (!proc && ruby_block->block_obj &&
rb_class_real(CLASS_OF(ruby_block->block_obj)) == klass) {
        return ruby_block->block_obj;
     }
     block = Data_Make_Struct(klass, struct BLOCK, blk_mark, blk_free,
data);


Yehuda Katz
Developer | Engine Yard
(ph) 718.877.1325

--00c09f88cf33975b3704809a28d4
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

<div>I have found a small 1.8.7 bug that prevents=A0proc_alloc from retriev=
ing the old Proc object if the Proc has a metaclass. Here&#39;s an example:=
</div><div><br></div><div><div>def foo</div><div>=A0=A0yield</div><div>end<=
/div>

<div>=A0</div><div>def bar(&amp;block)</div><div>=A0=A0puts Proc.new.object=
_id</div><div>=A0=A0def block.foo() &quot;FOO&quot; end</div><div>=A0=A0put=
s Proc.new.object_id</div><div>=A0=A0baz(&amp;block)</div><div>end</div><di=
v>=A0</div><div>

def baz(&amp;block)</div><div>=A0=A0puts Proc.new.object_id</div><div>=A0=
=A0yield</div><div>end</div><div>=A0</div><div>foo { puts &quot;HELLO&quot;=
 }</div><div>bar { puts &quot;HELLO&quot; }</div><div><br></div><div>Ruby 1=
.9, Rubinius and JRuby all return the same Proc object in all three cases. =
Ruby 1.8.7 returns a different Proc object in each case, because of the use=
 of CLASS_OF in proc_alloc:</div>

<div><br></div><div><div>=A0=A0 =A0if (!proc &amp;&amp; ruby_block-&gt;bloc=
k_obj &amp;&amp; CLASS_OF(ruby_block-&gt;block_obj) =3D=3D klass) {</div><d=
iv><span class=3D"Apple-tab-span" style=3D"white-space:pre">	</span>return =
ruby_block-&gt;block_obj;</div>

<div>=A0=A0 =A0}</div><div><br></div><div>It seems this code is trying to m=
ake sure that MyProc.new doesn&#39;t return a Proc object. The solution is =
a tiny patch to use rb_class_real:</div><div><br></div><div><div>index 8847=
ee5..54a0f46 100644</div>

<div>--- a/eval.c</div><div>+++ b/eval.c</div><div>@@ -8725,7 +8725,7 @@ pr=
oc_alloc(klass, proc)</div><div>=A0=A0 =A0 =A0 =A0rb_warn(&quot;tried to cr=
eate Proc object without a block&quot;);</div><div>=A0=A0 =A0 }</div><div>=
=A0</div><div>
- =A0 =A0if (!proc &amp;&amp; ruby_block-&gt;block_obj &amp;&amp; CLASS_OF(=
ruby_block-&gt;block_obj) =3D=3D klass) {</div>
<div>+ =A0 =A0if (!proc &amp;&amp; ruby_block-&gt;block_obj &amp;&amp; rb_c=
lass_real(CLASS_OF(ruby_block-&gt;block_obj)) =3D=3D klass) {</div><div>=A0=
=A0 =A0 =A0 =A0return ruby_block-&gt;block_obj;</div><div>=A0=A0 =A0 }</div=
><div>=A0=A0 =A0 block =3D Data_Make_Struct(klass, struct BLOCK, blk_mark, =
blk_free, data);</div>

</div><div><br></div></div></div><br clear=3D"all">Yehuda Katz<br>Developer=
 | Engine Yard<br>(ph) 718.877.1325<br>

--00c09f88cf33975b3704809a28d4--