ruby19 and defined?

[Joel VanderWerf]

This seems to be a change in 1.9 that's breaking some code (it was
reported by a user of the bit-struct library). Here's the test case:

class C; def self.foo; end; end
class D<C
p superclass
p defined?(superclass.foo)
p defined?(superclass.fooz)
end

$ ruby -v bug.rb
ruby 1.8.6 (2007-09-24 patchlevel 111) [i686-linux]
C
"method"
nil
$ ruby19 -v bug.rb
ruby 1.9.0 (2007-12-25 revision 14709) [i686-linux]
C
true
true
$ ruby19svn -v bug.rb
ruby 1.9.0 (2008-03-21 revision 15824) [i686-linux]
C
true
true

I don't see anything quite relevant in
http://eigenclass.org/hiki/Changes+in+Ruby+1.9.

 

[Roger Pack]

I know there was an RCR related to this once. Maybe it was accepted.
GL.
-R

$ ruby -v bug.rb
ruby 1.8.6 (2007-09-24 patchlevel 111) [i686-linux]
C
"method"
nil
$ ruby19 -v bug.rb
ruby 1.9.0 (2007-12-25 revision 14709) [i686-linux]
C
true
true

 

[Sebastian Hungerecker]

I know there was an RCR related to this once. =C2=A0Maybe it was accepted.

You mean an RCR proposing that defined? return true, right?
The problem here is not that defined? returns true instead of "method" thou=
gh.
The problem is that it returns true instead of nil for the method that does=
n't
exist. I doubt that that behaviour was proposed in an RCR.

=2D-=20
Jabber: (e-mail address removed)
ICQ: 205544826

 

[Rick DeNatale]

You mean an RCR proposing that defined? return true, right?
The problem here is not that defined? returns true instead of "method" though.
The problem is that it returns true instead of nil for the method that doesn't
exist. I doubt that that behaviour was proposed in an RCR.

Hmmmm, not so sure that that's actually a bug.

I've not found a good definition for what define? is supposed to do
when given an expression instead of a simple term. It certainly never
evaluated the expression. I don't know that I've ever used it with an
expression, but my impression was that it works syntactically more
than semantically, so whether or not a given expression would raise an
exception when evaluated wouldn't affect whether the expression was
defined. For example, I'd expect that

defined?(1/0)

would not indicate that the expression was undefined just because it
represents a divide by zero.

Having just looked at the latest beta of Pickaxe 3, I notice that Dave
still has define? returning strings instead of true/nil, and the page
has the red header indicating that it's been updated for 1.9.

Personally, I'm unhappy if 1.9 is no longer returning a description of
the syntactical class of the argument to defined? seems like a loss of
valuable information, and I'd hope that it was the returning of nil
rather than 'method' which is a bug.

 

[ts]

Hmmmm, not so sure that that's actually a bug.

It's a bug

Here the bytecode

vgs% ./ruby -ve 'puts
VM::InstructionSequence.compile("defined?(a.b)").disasm'
ruby 1.9.0 (2008-03-21 revision 15825) [i686-linux]
<compiled>:1: warning: useless use of defined? in void context
== disasm: <ISeq:<compiled>@<compiled>>=================================
0000 putself ( 1)
0001 defined 21, :a, false
0005 leave
0006 pop
0007 branchif 12
0009 putnil
0010 leave
0011 pop
0012 putnil
0013 send :a, 0, nil, 24, <ic>
0019 defined 13, :b, true
0023 leave
vgs%

Now if I add 'p'

vgs% ./ruby -ve 'puts VM::InstructionSequence.compile("p
defined?(a.b)").disasm'
ruby 1.9.0 (2008-03-21 revision 15825) [i686-linux]
== disasm: <ISeq:<compiled>@<compiled>>=================================
0000 putnil ( 1)
0001 putself
0002 defined 21, :a, false
0006 jump 24
0008 branchif 13
0010 putnil
0011 jump 24
0013 putnil
0014 send :a, 0, nil, 24, <ic>
0020 defined 13, :b, true
0024 send , 1, nil, 8, <ic>
0030 leave
vgs%



See the bytecode 'jump 24' in 0006


Guy Decoux