Marshal dump format error and inconsistencies with MRI
------------------------------------------------------
Key: JRUBY-5123
URL: http://jira.codehaus.org/browse/JRUBY-5123
Project: JRuby
Issue Type: Bug
Components: Ruby 1.8.7, Standard Library
Environment: jruby 1.5.3 (ruby 1.8.7 patchlevel 249) (2010-09-28
7ca06d7) (Java HotSpot(TM) 64-Bit Server VM 1.6.0_20) [x86_64-java]
Reporter: Loren Segal
There seems to be some different behaviour between JRuby and MRI when
unmarshalling user-marshalled objects. The following is an IRB session across
Ruby 1.8.7, 1.9.2, MacRuby 0.7 and JRuby 1.5.3 in unmarshalling some dumped
objects which include user-marshalled objects. This marshal dumped data is used
in YARD's test suite to test loading of Registries, it can be found at
http://github.com/lsegal/yard/tree/master/spec/serializers/data/serialized_yardoc/objects/Foo.dat
yard (master)$ ruby-switch-8
yard (master)$ ruby -v
ruby 1.8.7 (2009-06-12 patchlevel 174) [i686-darwin10.0.0]
yard (master)$ irb
>> require 'lib/yard';
Marshal.load(File.read_binary('spec/serializers/data/serialized_yardoc/objects/Foo.dat'))
Calling CodeObject("").root?()
=> #<yardoc class Foo>
>> exit
yard (master)$ ruby-switch-92
yard (master)$ ruby -v
ruby 1.9.2p0 (2010-08-18) [x86_64-darwin10.4.0]
yard (master)$ irb
>> require 'lib/yard';
Marshal.load(File.read_binary('spec/serializers/data/serialized_yardoc/objects/Foo.dat'))
Calling CodeObject("").root?()
=> #<yardoc class Foo>
>> exit
yard (master)$ ruby-switch-mac
yard (master)$ ruby -v
MacRuby 0.7 (ruby 1.9.2) [universal-darwin10.0, x86_64]
yard (master)$ irb
>> require 'lib/yard';
Marshal.load(File.read_binary('spec/serializers/data/serialized_yardoc/objects/Foo.dat'))
Calling CodeObject("").class()
Calling CodeObject("").root?()
=> #<yardoc class Foo>
>> exit
yard (master)$ ruby-switch-jruby
yard (master)$ ruby -v
jruby 1.5.3 (ruby 1.8.7 patchlevel 249) (2010-09-28 7ca06d7) (Java
HotSpot(TM) 64-Bit Server VM 1.6.0_20) [x86_64-java]
yard (master)$ irb
>> require 'lib/yard';
Marshal.load(File.read_binary('spec/serializers/data/serialized_yardoc/objects/Foo.dat'))
Calling CodeObject("Foo#baz").respond_to?(":to_str")
TypeError: #<YARD::StubProxy:0x2c42a818 @object=nil, @transient=false,
@path="Foo#baz"> is not a string
from (irb):1:in `load'
from (irb):1
This Marshal.load error turns into a "dump format error" when other objects are
unmarshalled:
>> YARD::Registry.load!('spec/serializers/data/serialized_yardoc')
ArgumentError: dump format error()
from
/Users/jinx/Development/Ruby/yard/lib/yard/serializers/yardoc_serializer.rb:72:in
`load'
from
/Users/jinx/Development/Ruby/yard/lib/yard/serializers/yardoc_serializer.rb:72:in
`deserialize'
from
/Users/jinx/Development/Ruby/yard/lib/yard/registry_store.rb:220:in `load_root'
from
/Users/jinx/Development/Ruby/yard/lib/yard/registry_store.rb:190:in
`load_yardoc'
from /Users/jinx/Development/Ruby/yard/lib/yard/registry_store.rb:90:in
`load'
from
/Users/jinx/Development/Ruby/yard/lib/yard/registry_store.rb:101:in `load!'
from /Users/jinx/Development/Ruby/yard/lib/yard/registry.rb:116:in
`load!'
from (irb):2
>>
The "Calling <...>" debugging statement was captured from method_missing inside
of the object that was being unmarshalled during a self._load() call. The
following is the user-marshalled class:
class StubProxy
instance_methods.each {|m| undef_method(m) unless m.to_s =~
/^__|^object_id$/ }
def _dump(depth) @path end
def self._load(str) new(str) end
def hash; @path.hash end
def initialize(path, transient = false)
@path = path
@transient = transient
end
def method_missing(meth, *args, &block)
return true if meth == :respond_to? && args.first == :_dump
puts "Calling
CodeObject(#[email protected]}).#{meth}(#{args.map{|x|x.inspect}.inspect[1...-1]})"
@object = nil if @transient
@object ||= Registry.at(@path)
@object.send(meth, *args, &block)
rescue NoMethodError => e
e.backtrace.delete_if {|l| l[0, __FILE__.size] == __FILE__ }
raise
end
end
The source (minus the debugging statement) can be found at
http://github.com/lsegal/yard/blob/master/lib/yard/serializers/yardoc_serializer.rb
Unfortunately, I was unable to create a minimal test case for this issue, it
seems to have something to do with the way YARD does things coupled with a
difference in the way JRuby does things. I can however confirm that if the
user-marshalled objects are taken out of the equation, the data is loaded just
fine, so I'm pretty sure it's related to user-marshalled objects. Again, I
should reiterate that most other Ruby implementations (even MacRuby) seem able
to handle loading/unloading of the dump file, so there is something wonky going
on with JRuby, even if YARD might be "helping".
Hopefully someone can help narrow this issue down-- as it stands, YARD is
unable to load the data it caches to disk between parsing, which is affecting
its ability to function properly. I'd be glad to answer any other questions and
help find the problem, but I spent (at least) a few hours with this and I'm at
a loss as to what it could be. I'd really like to see it fixed.
Regards,
Loren
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.codehaus.org/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
