Forum: Ruby-core [ruby-trunk - Bug #7627][Open] Instance variables of a Struct object are marshaled even when marshal

Issue #7627 has been reported by KL-7 (Kirill Lashuk).

----------------------------------------
Bug #7627: Instance variables of a Struct object are marshaled even when 
marshal_dump method is defined.
https://bugs.ruby-lang.org/issues/7627

Author: KL-7 (Kirill Lashuk)
Status: Open
Priority: Normal
Assignee:
Category:
Target version:
ruby -v: ruby 2.0.0dev (2012-11-01 trunk 37411) [x86_64-darwin12.2.0]


=begin
If (({marshal_dump})) method is defined in a class and you pass an 
instance of that class into (({Marshal.dump})), instead of dumping 
instance variables (as it happens by default, when (({marshal_dump})) is 
not defined) Ruby dumps only data returned from this method.

Though, if the class is a (({Struct})), MRI 1.8.7 behaves the same way 
as for a regular class, but both MRI 1.9.3 and MRI 2.0.0 in addition to 
the data returned from (({marshal_dump})) method automatically dump 
instance variables of the object.

Code example:

 # marshal.rb

 class A
   attr_accessor :foo

   def marshal_dump; 'dump'; end  # fake dump data
   def marshal_load(*); end       # do nothing
 end

 class B < Struct.new(:bar)
   attr_accessor :foo

   def marshal_dump; 'dump'; end  # fake dump data
   def marshal_load(*); end       # do nothing
 end

 def test(obj)
   obj.foo = 12

   dump   = Marshal.dump(obj)
   loaded = Marshal.load(dump)

   p obj, dump, loaded, loaded.foo
   puts
 end

 test(A.new)
 test(B.new)

MRI 1.8.7 (instance variable of the Struct object ((*is not dumped*))):

 $ ruby-1.8.7-p371 marshal.rb
 #<A:0x10ff5e0a0 @foo=12>
 "\004\bU:\006A\"\tdump"
 #<A:0x10ff5df38>
 nil

 #<struct B bar=nil>
 "\004\bIU:\006B\"\tdump\000"
 #<struct B bar=nil>
 nil

MRI 1.9.3 (instance variable of the Struct object ((*is dumped*))):

 $ ruby-1.9.3-p327 marshal.rb
 #<A:0x007ff79d80e6b0 @foo=12>
 "\x04\bU:\x06AI\"\tdump\x06:\x06EF"
 #<A:0x007ff79d80e5c0>
 nil

 #<struct B bar=nil>
 "\x04\bIU:\x06BI\"\tdump\x06:\x06EF\x06:\t@fooi\x11"
 #<struct B bar=nil>
 12

MRI 2.0.0-preview1 (instance variable of the Struct object ((*is 
dumped*))):

 $ ruby-2.0.0-preview1 marshal.rb
 #<A:0x007fb46908bf68 @foo=12>
 "\x04\bU:\x06AI\"\tdump\x06:\x06EF"
 #<A:0x007fb46908be00>
 nil

 #<struct B bar=nil>
 "\x04\bIU:\x06BI\"\tdump\x06:\x06EF\x06:\t@fooi\x11"
 #<struct B bar=nil>
 12

I'm not sure if it's expected behavior, but it doesn't look like one to 
me. Probably, it has something to do with the implementation of 
(({has_ivars})) macro that for some reason is used 
((<here|URL:https://github.com/ruby/ruby/blob/trunk/marshal.c#L661>)) 
while dumping an object with defined (({marshal_dump})) method, but I 
might be wrong, because I didn't manage to track down the problem 
further.

=end

Issue #7627 has been updated by ko1 (Koichi Sasada).


I'm not sure about this issue, but is it high-priority trunk issue?
or not important issue?

----------------------------------------
Bug #7627: Instance variables of a Struct object are marshaled even when 
marshal_dump method is defined.
https://bugs.ruby-lang.org/issues/7627#change-35614

Author: KL-7 (Kirill Lashuk)
Status: Open
Priority: Normal
Assignee:
Category:
Target version:
ruby -v: ruby 2.0.0dev (2012-11-01 trunk 37411) [x86_64-darwin12.2.0]


=begin
If (({marshal_dump})) method is defined in a class and you pass an 
instance of that class into (({Marshal.dump})), instead of dumping 
instance variables (as it happens by default, when (({marshal_dump})) is 
not defined) Ruby dumps only data returned from this method.

Though, if the class is a (({Struct})), MRI 1.8.7 behaves the same way 
as for a regular class, but both MRI 1.9.3 and MRI 2.0.0 in addition to 
the data returned from (({marshal_dump})) method automatically dump 
instance variables of the object.

Code example:

 # marshal.rb

 class A
   attr_accessor :foo

   def marshal_dump; 'dump'; end  # fake dump data
   def marshal_load(*); end       # do nothing
 end

 class B < Struct.new(:bar)
   attr_accessor :foo

   def marshal_dump; 'dump'; end  # fake dump data
   def marshal_load(*); end       # do nothing
 end

 def test(obj)
   obj.foo = 12

   dump   = Marshal.dump(obj)
   loaded = Marshal.load(dump)

   p obj, dump, loaded, loaded.foo
   puts
 end

 test(A.new)
 test(B.new)

MRI 1.8.7 (instance variable of the Struct object ((*is not dumped*))):

 $ ruby-1.8.7-p371 marshal.rb
 #<A:0x10ff5e0a0 @foo=12>
 "\004\bU:\006A\"\tdump"
 #<A:0x10ff5df38>
 nil

 #<struct B bar=nil>
 "\004\bIU:\006B\"\tdump\000"
 #<struct B bar=nil>
 nil

MRI 1.9.3 (instance variable of the Struct object ((*is dumped*))):

 $ ruby-1.9.3-p327 marshal.rb
 #<A:0x007ff79d80e6b0 @foo=12>
 "\x04\bU:\x06AI\"\tdump\x06:\x06EF"
 #<A:0x007ff79d80e5c0>
 nil

 #<struct B bar=nil>
 "\x04\bIU:\x06BI\"\tdump\x06:\x06EF\x06:\t@fooi\x11"
 #<struct B bar=nil>
 12

MRI 2.0.0-preview1 (instance variable of the Struct object ((*is 
dumped*))):

 $ ruby-2.0.0-preview1 marshal.rb
 #<A:0x007fb46908bf68 @foo=12>
 "\x04\bU:\x06AI\"\tdump\x06:\x06EF"
 #<A:0x007fb46908be00>
 nil

 #<struct B bar=nil>
 "\x04\bIU:\x06BI\"\tdump\x06:\x06EF\x06:\t@fooi\x11"
 #<struct B bar=nil>
 12

I'm not sure if it's expected behavior, but it doesn't look like one to 
me. Probably, it has something to do with the implementation of 
(({has_ivars})) macro that for some reason is used 
((<here|URL:https://github.com/ruby/ruby/blob/trunk/marshal.c#L661>)) 
while dumping an object with defined (({marshal_dump})) method, but I 
might be wrong, because I didn't manage to track down the problem 
further.

=end