Running a simple ruby script that reads a file, parses through it, and generates a new file (same name) caused the latest commit to produce an assertion failure. I have a tarball of the script along with the sample file that caused it. I'll try to attach them here somehow... otherwise a followup message will contain pointers to gists containing the data.

A segfault results if the VM encounters a (set|push)_local_depth when the scope nesting does not match the depth given.

For example:

class Hash
  def self.allocate
    hash = super()
    Rubinius.privately { hash.setup }
    hash
  end
end

If the compiler does not fix up the local access for hash in hash.setup in Rubinius.privately, the push_local_depth insn will attempt call_frame->scope->parent() and parent is 0x0.

ERROR in temp_server, 0
An exception occurred evaluating command line code
unbalanced stack at line 81 (151): 1 != 2 (RuntimeError)

Backtrace:
Compiler::StackDepthCalculator#run_from at kernel/compiler/stack.rb:48 (2 times)
Compiler::StackDepthCalculator#run at kernel/compiler/stack.rb:13
    Compiler::Generator#to_cmethod at kernel/compiler/generator.rb:185
Compiler::MethodDescription#to_cmethod at kernel/compiler/bytecode.rb:34
Compiler::Generator#encode_literals {} at kernel/compiler/generator.rb:142
                        Array#each at kernel/bootstrap/array.rb:152
Compiler::Generator#encode_literals at kernel/compiler/generator.rb:140
    Compiler::Generator#to_cmethod at kernel/compiler/generator.rb:193
Compiler::MethodDescription#to_cmethod at kernel/compiler/bytecode.rb:34
Compiler::Generator#encode_literals {} at kernel/compiler/generator.rb:142
                        Array#each at kernel/bootstrap/array.rb:152
Compiler::Generator#encode_literals at kernel/compiler/generator.rb:140
    Compiler::Generator#to_cmethod at kernel/compiler/generator.rb:193
Compiler::MethodDescription#to_cmethod at kernel/compiler/bytecode.rb:34
Compiler::Generator#encode_literals {} at kernel/compiler/generator.rb:142
                        Array#each at kernel/bootstrap/array.rb:152
Compiler::Generator#encode_literals at kernel/compiler/generator.rb:140
    Compiler::Generator#to_cmethod at kernel/compiler/generator.rb:193
Compiler::MethodDescription#to_cmethod at kernel/compiler/bytecode.rb:34
             Compiler.compile_file at kernel/compiler/compiler.rb:56
      Compiler::Utils.compile_file at kernel/compiler/compile.rb:36
    Compiler::Utils.single_load {} at kernel/compiler/compile.rb:199
   Compiler::Utils.compile_feature at kernel/compiler/compile.rb:142
       Compiler::Utils.single_load at kernel/compiler/compile.rb:198
   Compiler::Utils.unified_load {} at kernel/compiler/compile.rb:131
                        Array#each at kernel/bootstrap/array.rb:152
      Compiler::Utils.unified_load at kernel/compiler/compile.rb:95
            Kernel(Object)#require at kernel/common/kernel.rb:711
                   Object#__eval__ at (eval):1
     Kernel(Rubinius::Loader)#eval at kernel/common/eval.rb:85
         Rubinius::Loader#evals {} at kernel/loader.rb:304
                        Array#each at kernel/bootstrap/array.rb:152
            Rubinius::Loader#evals at kernel/loader.rb:303
             Rubinius::Loader#main at kernel/loader.rb:403
                 Object#__script__ at kernel/loader.rb:452

Since @subclasses stores the list of classes that are inherited, the class definition will not be garbage collected due to being referenced in this list.

i tried installing the rjb gem without installing rubinius (that is, running rbx directly after 'rake'. i didn't install since that did not succeed, see another issue)

bin/rbx gem install rjb-1.1.6.gem --no-rdoc --no-ri Building native extensions. This could take a while...
ERROR: Error installing rjb-1.1.6.gem:

ERROR: Failed to build gem native extension.

/work/research/rubinius/bin/rbx extconf.rb checking for jni.h... yes
checking for dl.h... no
checking for ruby/dl.h... no
extconf.rb failed

Assigning to already defined constants gives unexpected behaviour or even segmentation faults. I have built from latest git. Examples:

String = nil
puts "foo"
=> Segmentation Fault

String = nil
class String; end
=> Segmentation Fault

Array = nil
puts "foo"
=> (Nothing gets printed)

Array = nil
class Array; end
=> (Nothing happens)

Fixnum = nil
puts 42
=> 42

Fixnum = nil
class Fixnum; end
=> Fixnum is not a class (TypeError) with backtrace: Exception occurred during top-level exception output! (THIS IS BAD)

In all of this, Fixnum seems to behave most correctly. I haven't tried any other constants.

Reported by wycats as Lighthouse issue 781.

module Annotations
  def self.extend(klass)
    klass.class_eval { include Annotation }
  end
  
  def annotate(module_name, &block)    
    new_module = Module.new do
      @@execute_method = Annotations.const_get(module_name).method(:execute)
      
      def self.method_added(name)
        original_method = instance_method(name)
        remove_method(name)
        
        method_module = Module.new do
          define_method(name, original_method)
        end
        
        execute_module = Module.new do
          define_method(name, @@execute_method)
        end
        
        include method_module
        include execute_module
      end
    end
    
    new_module.module_eval(&block)
    
    include new_module
  end
end

module Annotations
  module Logging
    def self.execute(*args, &block)
      puts "ARGS: #{args}"
      super
    end
  end
end

class Foo
  extend Annotations
  
  annotate :Logging do
    def hello
      puts "HELLO"
    end
  end
end

puts Foo.new.hello

Result in MRI:

annotations.rb:52:in `hello': singleton method bound for a different object (TypeError)
    from annotations.rb:52

Result in Rubinius

ARGS: 
An exception occurred running annotations.rb
    No method 'execute' on Annotations::Logging (Module) (NoMethodError)

Backtrace:
  Kernel(Module)#execute (method_missing) at kernel/delta/kernel.rb:45
          Annotations::Logging.execute at annotations.rb:37
                           Method#call at kernel/common/method.rb:70
        Rubinius::DelegatedMethod#call at kernel/common/delegated_method.rb:17
                       main.__script__ at annotations.rb:52
    Rubinius::CompiledMethod#as_script at kernel/common/compiled_method.rb:216
           Compiler::Utils.single_load at kernel/compiler/compile.rb:244
   Compiler::Utils.load_from_extension at kernel/compiler/compile.rb:329
               Rubinius::Loader#script at kernel/loader.rb:326
                 Rubinius::Loader#main at kernel/loader.rb:406
                     Object#__script__ at kernel/loader.rb:454

Reported by shugo, LH 779.

Module#append_features does not detect cyclic include.

$ cat t.rb          
module A
end

module B
  include A
end

begin
  A.send(:append_features, A)
rescue ArgumentError => e
  p e
end
begin
  B.send(:append_features, A)
rescue ArgumentError => e
  p e
end
p A.ancestors
p B.ancestors
$ ruby -v t.rb 
ruby 1.8.7 (2008-08-11 patchlevel 72) [i486-linux]
#<ArgumentError: cyclic include detected>
#<ArgumentError: cyclic include detected>
[A, B]
[B, A]
$ rbx -v t.rb
rubinius 0.11.0-dev (1.8.6 8679f8dd 12/31/2009) [i686-pc-linux-gnu]
[A, B, A]
[B, A]

LH 765 by shugo.

Rubinius handles the following script differently from MRI. Is it intensional?

$ cat t.rb
x = Object.new
def x.to_s
  return 1
end
p "#{x}"
$ ruby-1_8_6 -v t.rb
ruby 1.8.6 (2008-08-08 patchlevel 286) [i686-linux]
"#<Object:0xb7dedbd8>"
$ rbx -v t.rb     
rubinius 0.11.0-dev (ruby 1.8.6) (eabf5ec7d 12/31/2009) [i686-pc-linux-gnu]
An exception has occurred:
      Tried to use non-reference value 0x3 as type String (48) (TypeError)

Backtrace:
            Compiler::Utils.single_load at kernel/compiler/compile.rb:245
    Compiler::Utils.load_from_extension at kernel/compiler/compile.rb:323
                      Object#__script__ at kernel/loader.rb:240

LH 717 reported by Charles L

gauss:rubinius brian$ bin/rbx 
irb(main):001:0> a = 1; a()
=> 1
irb(main):002:0> b = 1
=> 1
irb(main):003:0> b()
=> 1
irb(main):004:0>

gauss:rubinius brian$ irb
>> a = 1; a()
NoMethodError: undefined method `a' for main:Object
        from (irb):1
>> b = 1
=> 1
>> b()
NoMethodError: undefined method `b' for main:Object
        from (irb):3
>>

Evan's comment:

I investigated this a bit more today. The fix is to, when vcall is normalized, remember that it was a vcall in some way.

That way, we can only check for a local variable override where a vcall was used. This is tricky right now, because adding things to the sexp form blows up the compiler, since it's using .last for a number of things related to inspecting a call sexp (and expecting an s(:arglist...))

I tried to add a @style instance variable to the sexp, but it never makes it all the way through the pipeline, I'm assuming due to the sexp being pulled apart and rebuilt, which looses the ivars.

LH 614 reported by Sam Aaron

Run this code: http://gist.github.com/204542

ruby 1.8.6 (2008-08-11 patchlevel 287) [universal-darwin9.0]
A.cheese: edam
B.cheese: stilton
a.cheese: shropshire blue
b.cheese: cheddar
b_with_meta.cheese: cheshire
b_with_meta.metaclass.cheese: stilton
b_with_meta_meta.cheese: brie
b_with_meta_meta.metaclass.cheese: gouda
b_with_meta_meta.metaclass.metaclass.cheese: wensleydale

gauss:rubinius brian$ bin/rbx -v meta.rb
rubinius 0.13.0-dev (1.8.7 d300514e 2009-11-06 JI) [i686-apple-darwin9.8.0]
A.cheese: edam
B.cheese: stilton
a.cheese: shropshire blue
b.cheese: cheddar
b_with_meta.cheese: cheshire
An exception occurred running meta.rb
    No method 'cheese' on #> (MetaClass) (NoMethodError)

Backtrace:
  Kernel(MetaClass)#cheese (method_missing) at kernel/delta/kernel.rb:45
                       main.__script__ at meta.rb:57
    Rubinius::CompiledMethod#as_script at kernel/common/compiled_method.rb:221
           Compiler::Utils.single_load at kernel/compiler/compile.rb:244
   Compiler::Utils.load_from_extension at kernel/compiler/compile.rb:330
               Rubinius::Loader#script at kernel/loader.rb:327
                 Rubinius::Loader#main at kernel/loader.rb:406
                     Object#__script__ at kernel/loader.rb:454

LH 522 reported by manveru.

I found that there is a different behaviour in rubinius for local variables assigned in a block.

# Ruby 1.8:
10.times{ bar = 1 }; bar
#=> NameError: undefined local variable or method `bar' for main:Object

# Rubinius
10.times{ bar = 1 }; bar
# nil

According to evan, the following sexp has to change.

pp "10.times { bar = 1 }; bar".to_sexp
[:block,
 [:newline,
  1,
  "(eval)",
  [:iter,
   [:call, [:lit, 10], :times],
   nil,
   [:block,
    [:dasgn_curr, :bar],
    [:newline, 1, "(eval)", [:lasgn, :bar, [:lit, 1]]]]]],
 [:newline, 1, "(eval)", [:lvar, :bar, 0]]]

that should read

pp "10.times { bar = 1 }; bar".to_sexp
[:block,
 [:newline,
  1,
  "(eval)",
  [:iter,
   [:call, [:lit, 10], :times],
   nil,
   [:block,
    [:dasgn_curr, :bar],
    [:newline, 1, "(eval)", [:lasgn, :bar, [:lit, 1]]]]]],
 [:newline, 1, "(eval)", [:vcall, :bar, 0]]]

so the :lvar on the last line becomes :vcall

The libxml-ruby gem builds under the rbx C-API except for this error. Send a patch to not use st_foreach:

  case T_HASH:
    st_foreach(RHASH_TBL(nslist), iterate_ns_hash, self);
    break;
 

ruby_xml_xpath_context.c: In function ‘rxml_xpath_context_register_namespaces’:
ruby_xml_xpath_context.c:223: warning: implicit declaration of function ‘assert’
ruby_xml_xpath_context.c:223: error: syntax error before ‘?’ token
ruby_xml_xpath_context.c:223: error: syntax error before ‘)’ token

Reported by pluskid LH 310

The second optional argument of Regexp.new can be used to indicate the language/encoding.

The default behavior:

re = Regexp.new(".")
str = "中文" # or "\344\270\255\346\226\207" in utf-8 encoding
re.match(str)[0] # => "\344"

When specifying the language/encoding:

re = Regexp.new(".", nil, 'u')
str = "中文" # or "\344\270\255\346\226\207" in utf-8 encoding
re.match(str)[0] # => "\344\270\255"

However, there's a global variable $KCODE that indicate the current language/encoding. When set, the regexp should behavior according to this, thus:

$KCODE = 'u'
re = Regexp.new(".")
str = "中文" # or "\344\270\255\346\226\207" in utf-8 encoding
re.match(str)[0] # => "中" or  "\344\270\255" in utf-8 encoding

Those are Ruby 1.8 behavior. Since Ruby 1.9 gains full Unicode support, the global variable $KCODE is no longer used. I think Rubinius is currently making capability mainly to Ruby 1.8, so this should be considered.

One way to fix this, I think, is to change the default value for the second optional argument (lang) of Regexp.new from "nil" to "$KCODE".

I don't know whether Rubinius and Ruby1.8 use the same regexp engine. But it seems that even though I set $KCODE to 'u' in Ruby1.8. The code

Regexp.new(".").inspect

will return "/./" but

Regexp.new(".", nil, "u").inspect

returns "/./u" . However, the "/./" can successfully match a multi-byte character when setting $KCODE to 'u', but fails in Rubinius. So I think maybe some better way is to patch the regexp engine to take care of the global variable instead of patch the Regexp.new method.

LH 542

Add hooks like rubyprof uses so that we can get rcov running.

Reported by zimbatm LH 279

Paraphrasing: "gettimeofday" is used in various places. This could possibly cause timing problems if the system date is changed between the uses.

(I don't know if this is an issue.)

Suggested fix:

I find that "get_clock" in libev's "ev.c" is a good implementation but it is not exported in "ev.h". It would be nice if libev would also export it's cross-platform timers independently.

Reported by Tony Arcieri LH 536

Multiplexing different types of events on Channels is presently pretty difficult, to the point that IO#select cannot be implemented.

Evan threw out the idea of event objects. These could hopefully store any data resulting from I/O completions, the source of the I/O event, or completely non-I/O related things as not all events are I/O related, such as timeout events.

Event objects could abstract interaction with scheduler directly, e.g.:

te = TimeoutEvent.perform chan, 5.0, value
re = ReadEvent.perform chan, io, 4096, value

(where value is an optional user-specified parameter)

when the event occurs, a corresponding Event object is written to the channel, which would include the event type through the nature of its class, and also provide accessors for the event source, any data that may have resulted from the event (e.g. data read from an IO object), and the user value, if it was specified.

This would make for easy case-based processing of IO events, e.g.:

ev = chan.receive
case ev
when ReadEvent then read_handler(ev)
when WriteEvent then write_handler(ev)
when TimeoutEvent then timeout_handler(ev)
... end

Reported by kwatch LH 319

$SAFE is Proc local in Ruby but not in Rubinius.

hoge.rb:

puts   "*** before:  $SAFE=#{$SAFE.inspect}"
proc {
  $SAFE = 2
  puts "*** in proc: $SAFE=#{$SAFE}"
}.call
puts   "*** after:   $SAFE=#{$SAFE.inspect}"

result:

$ ruby hoge.rb
*** before:  $SAFE=0
*** in proc: $SAFE=2
*** after:   $SAFE=0     # not changed
$ shotgun/rubinius hoge.rb
*** before:  $SAFE=0
*** in proc: $SAFE=2
*** after:   $SAFE=2     # CHANGED!

environment:

$ ruby -v
ruby 1.8.6 (2007-09-24 patchlevel 111) [i686-darwin8.11.1]
$ shotgun/rubinius -v
rubinius 0.8.0 (ruby 1.8.6 compatible) (2ffa558ee) (02/08/2008) [i686-apple-darwin8.11.1]

Reported by Wilson LH 378

The Process.setrlimit spec for RLIMIT_MEMLOCK needs revision on Linux.
It appears to fail because the lim/max return values are not necessarily valid arguments to setrlimit.
While this is retarded, it is real, and needs some work.

I am currently installing Ubuntu so that I can debug this mess.

Process.setrlimit and Process.getrlimit limit and get total size for mlock(2) (bytes) ERROR
Invalid argument: 
                   Errno.handle at kernel/core/errno.rb:19
              Process.setrlimit at kernel/core/process.rb:35
           Object#__script__ {} at ./spec/ruby/1.8/core/process/setrlimit_spec.rb:73

Reported by agardiner LH 497

A seek error results on a socket when an attempt is made to write to a socket on which there is also data still to be read. On MatzRuby, no error occurs.

The exception on RBX is as follows:

    Unable to seek (ESPIPE)

Backtrace:
             IO(TCPSocket)#seek at kernel/core/io.rb:507
            IO(TCPSocket)#write at kernel/core/io.rb:490
          IO(TCPSocket)#puts {} at kernel/core/io.rb:371
                     Array#each at kernel/core/array.rb:573
             IO(TCPSocket)#puts at kernel/core/io.rb:355
              Object#__script__ at /home/ads/test_client.rb:10
       CompiledMethod#as_script at kernel/core/compiled_method.rb:326
            Compile.single_load at kernel/core/compile.rb:238
    Compile.load_from_extension at kernel/core/compile.rb:310
              Object#__script__ at kernel/loader.rb:201

The following server and client code can be used to reproduce the problem:

Server code

require 'socket'

REMOTE_DEBUG_PORT = 1098
TEST_DATA = "  Here
  is
  some
  multi-line
  data
"

port = ARGV.shift || REMOTE_DEBUG_PORT
server = TCPServer.open(port)

puts "*** waiting for client..."
socket = server.accept

puts "Sending data"
TEST_DATA.each_line {|l| socket.puts l}
socket.flush

resp = socket.gets
socket.close

Client code

require 'socket'

host = ARGV[0] || 'localhost'
port = ARGV[1] || 1098

socket = TCPSocket.new(host, port)
puts "Connected to server on #{host}:#{port}"
while line = socket.gets
  puts line
  socket.puts line
end
socket.close

Reported by Lin Jen-Shin LH 729

Greetings, I sent below to rubinius-dev mailing list but it didn't show up in google group, so I re-post them here.

Sorry if it's wrong place to post here.

I am not (yet) digging into source code,
and I guess IncludedModule was created
for separating each included module.
For example:
http://gist.github.com/30606 (first one or below)

module M
 def m
 end
end

module K
end

class C
 include K
end

module K
 include M
end

C.new.m # => undefined method

class D
 include K
end

D.new.m # => nil
C.new.m # => undefined method

Things are different in Kernel, because there's no
IncludedModule for Kernel in superclass_chain:
http://gist.github.com/30606 (last one or below)

module M
 def m
 end
end

module Kernel
 include M
end

class C
end

C.new.m # => undefined method

class D
 include Kernel
end

D.new.m # => nil
C.new.m # => nil in Rubinius
# undefined method in MRI 1.8/1.9 and JRuby

Should there be (an) IncludedModule(s) for Kernel?

Thanks and cheers,

Reported by Jari Bakken LH 774

Method#inspect was broken when backed by a Rubinius::AccessVariable:

[].method(:length) #=> NoMethodError: No method 'name' on an instance of Rubinius::AccessVariable.

There are other problems here as well

[].method(:length).call # NoMethodError: No method 'activate' on an instance of Rubinius::AccessVariable.

[].method(:length).arity # NoMethodError: No method 'arity' on an instance of Rubinius::AccessVariable.

Patch: http://gist.github.com/204865

steps to reproduce:

$> cd rubinius
$>./configure --prefix=/usr/local
$> rake
...
VM Assertion:
set RBX_RUNTIME to runtime (or equiv)

2   rbx  0x00048673 rubinius::VMException::VMException(char const*, bool) + 99
3   rbx  0x00126082 main + 850
4   rbx  0x00002746 start + 54

Ruby backtrace:
Abort!
2   rbx  0x00048968 rubinius::abort() + 40
3   rbx  0x0012668e main + 2398
4   rbx  0x00002746 start + 54
Assertion failed: (0), function abort, file vm/exception.cpp, line 114.
rake aborted!
Command failed with status (): [./bin/rbx compile  -p -I/usr/local/src/rub...]

Which seems to indicate rbx can't find it's runtime dir.

Here is a transcript of it in action with a gdb trace.
http://gist.github.com/212795

Rubinius version: 590b40c

I can't reproduce the failure when running a file with the same code.

Script and transcript here:
http://gist.github.com/212963

Rubinius version: 590b40c

It requires large and complex data to trigger (visible in example above).

https://gist.github.com/b31d63373fe0608cb8dd

def a; define_method(:b) { super }; end

calls a in rubinius, and calls b in MRI

class A
        def b
                p :b
        end

        def self.a
                p :a
        end
end

class B < A
        def self.a
                define_method(:b) {
                        super()
                }
        end
end

B.a

B.new.b

MRI >> :b (super() from the block)
rbx >> :a (super() from the method)

This is an easy reproducable segfault, basically the only thing needed to do is install the rdoc with RI enabled.

For a backtrace and some more information, see: https://gist.github.com/7ced204c9a6cfffb1721