The Rubinius Debugger

When I read the posting about Cylon, the Ruby debugger for Visual Studio by SapphireSteel, I immediately thought about the Rubinius debugger highlighted in InfoQ recently. I decided to give it a try and see what I came up with. I wasn’t able to replicate Dermot’s tests exactly (I don’t know where he set his breakpoints, and Rubinius doesn’t support everything he tested), but here are my initial results:

Line Based

Rubinius no debugger	Rubinius w/debugger no breakpoints	Ruby
2.039 seconds	2.399 seconds	0.898 seconds

Call Based

Rubinius no debugger	Rubinius w/debugger no breakpoints	Ruby
13.441 second	13.999 seconds	4.171 seconds

Rubinius still has a long performance road ahead of them in terms of general execution, but it’s pretty exciting to see that their debugger is so fast. The cylon debugger (which smoked the other options in their testing) was 25% slower for line based debugging, and 306% slower for call based. By comparison, Rubinius’ debugger only added 17% for line based and 4% for call based (actually, I’m guessing that had the line based run through more iterations, the Rubinius debugger would have done better, I think most of its time is spent in set up).

Using the Rubinius debugger requires that you add a call to the debugger to the program:

debugger

def fac(n)
  lvar = n
  n == 1 ? 1 : n * fac(n-1)
  return n
end

count = 0

tstart = Time.new
0.upto(1000000) {fac(5)}               # 1ine based
#0.upto(30000000) {count += 1}         # call based
tend = Time.new  
puts "%10.3f" % tstart
puts "%10.3f" % tend.to_f
diff = tend - tstart
puts "%10.3f" % diff.to_f

When you run this, you’ll get something like this:

$ ../shotgun/rubinius debug_fac.rb
[Debugger activated]

debug_fac.rb:1 [IP:5]

rbx:debug> c
[Resuming program]
1201801369.694
1201801372.001
     2.307

There are several commands available:

command	explanation
h	get a listing of commands
b	list breakpoints
b	set a breakpoint at the start of the method
n	Step to the next line
ni	step to the next VM instruction
c	continue execution until the next breakpoint
l	list source code around the current breakpoint
d	decode VM bytecode around the current breakpoint
v	display local variables and their values
vs	display the contents of the VM stack

Some of these are pretty cool, take a look at these examples:

rbx:debug> vs
     VM stack [0-5]:
  sp => 0: TrueClass  true
  fp => 1: Class      Rubinius::VM
        2: String     "" 
        3: String     "debug_fac.rb" 
        4: String     "debug_fac.rbc" 
        5: NilClass   nil
           ...

or:

rbx:debug> d
   Bytecode instructions [5-25] in compiled method script:
           ...
           # line 1
     0005: pop
           # line 3
     0006: push_literal  #
     0008: push_literal  :fac
     0010: push_self
     0011: send_stack    :add_method, 2
     0014: pop
           # line 9
  => 0015: meta_push_0
     0016: set_local     :count
     0018: pop
           # line 11
     0019: push_const    :Time
     0021: send_method   :new
     0023: set_local     :tstart
     0025: pop
           ...

which compares to:

rbx:debug> l
   Source lines [1-18] in debug_fac.rb:
      1: debugger
      2:
      3: def fac(n)
      4:   lvar = n
      5:   n == 1 ? 1 : n * fac(n-1)
      6:   return n
      7: end
      8:
  =>  9: count = 0
     10:
     11: tstart = Time.new
     12: 0.upto(1000000) {fac(5)}               # $-1òølineòù based
     13: #0.upto(30000000) {count += 1}     #$-1òùcallòù based
     14: tend = Time.new
     15: puts "%10.3f" % tstart
     16: puts "%10.3f" % tend.to_f
     17: diff = tend - tstart
     18: puts "%10.3f" % diff.to_f

Like a lot of things in Rubinius, the debugger isn’t quite ready for primetime but it sure shows a lot of promise.