Bug #3841: RubyVM::InstructionSequence.to_a() and disasm() do not work properly for "for <var> in <list>"
http://redmine.ruby-lang.org/issues/show/3841

Author: Sean O'Neil
Status: Open, Priority: Normal
Category: core
ruby -v: ruby 1.9.2p0 (2010-08-18 revision 29036) [i386-mswin32_90]

I have been playing with the concept of caching compiled Ruby instruction sequences and reloading them later. For now the simplest way to do it is to use RubyVM::InstructionSequence.to_a() and Marshal.dump() to convert the resulting array to a smaller binary byte sequence. While testing this, I found that some Ruby scripts worked fine when an instruction sequence was reloaded, but others broke.

As a test, I decided to try compiling irb.rb and its dependencies. When I try to reload the compiled sequences, the first file that breaks is irb/extend-command.rb. I tracked it down to the way "for <vars> in <list>" code block and parameters are handled, but I am not sure how to fix it. When I converted each "for <vars> in <list>" loop to "<list>.each {|<vars>|", it started working perfectly. Here's an example of how the disassembly differs between the two loops:

def self.install_extend_commands
  @EXTEND_COMMANDS.each {|args|
    def_extend_command(*args)
  }
end
== disasm: <RubyVM::InstructionSequence:block in install_extend_commands@irb/extend-command.rb>
== catch table
| catch type: redo   st: 0000 ed: 0012 sp: 0000 cont: 0000
| catch type: next   st: 0000 ed: 0012 sp: 0000 cont: 0012
|------------------------------------------------------------------------
local table (size: 2, argc: 1 [opts: 0, rest: -1, post: 0, block: -1] s3)
[ 2] args<Arg>  
0000 trace            1                                               ( 110)
0002 putnil           
0003 getdynamic       args, 0
0006 send             :def_extend_command, 1, nil, 10, <ic:0>
0012 leave            

def self.install_extend_commands
  for args in @EXTEND_COMMANDS
    def_extend_command(*args)
  end
end
== disasm: <RubyVM::InstructionSequence:block in install_extend_commands@irb/extend-command.rb>
== catch table
| catch type: redo   st: 0005 ed: 0016 sp: 0000 cont: 0005
| catch type: next   st: 0005 ed: 0016 sp: 0000 cont: 0016
|------------------------------------------------------------------------
local table (size: 2, argc: 1 [opts: 0, rest: -1, post: 0, block: -1] s3)
[ 2] ?<Arg>     
0000 getdynamic       *, 0                                            ( 206)
0003 setlocal         args                                            ( 204)
0005 trace            1                                               ( 205)
0007 putnil           
0008 getlocal         args
0010 send             :def_extend_command, 1, nil, 10, <ic:0>
0016 leave            

In this example, rb_id2name() returns NULL for the name of the argument passed to the code block. That is why it puts a "?" in the local table and a "*" in the getdynamic command. The to_a() method has a similar problem, and it appears to put a number there instead. When I use the load() method to convert the array back into an instruction sequence and try to execute it, the code block ends up receiving a number as a parameter instead of the entry in the list it is supposed to get.

I imagine this is not a high priority for the Ruby core team, and I don't mind putting in some work to fix it. However, it looks like it would take me a while to figure out, so I would appreciate any tips you could give me to point me in the right direction.

Thank you,
Sean O'Neil


----------------------------------------
http://redmine.ruby-lang.org