Included in Kernel, yet won't show up in `C::ancestors'?

[Arlen Cuss]

[Note: parts of this message were removed to make it a legal post.]

Hi all,

I'm fairly sure I'm a long way down the path of madness with this one.
Please bare with me. I've been on this for some time now.

I've been trying to work out the order in which items appear in
"SomeClass::ancestors". I thought I had it before:

[SomeClass, SomeClass'sIncludedModules..., SuperClass,
SuperClass'sIncludedModules, SuperSuperClass, ...]

I constructed a test case:

$ ruby -e 'module M; end; module N; end; class A; include M; end; class B <
A; include N; end; p B.ancestors'
[B, N, A, M, Object, Kernel]

This is the expected result! But then I wondered.. what if we included
something in `M'? It shouldn't appear according to my theory here, since M
is not included by SomeClass itself. But, sure enough, it does appear.
(which makes sense!)

The reason I presumed this is because if I include a module in Kernel, it
won't appear in the list of any Class's ancestors, yet it will appear in
Kernel's:

$ ruby -e 'module A; end; module Kernel; include A; end; class C; end; p
C.ancestors; p Kernel.ancestors'
[C, Object, Kernel]
[Kernel, A]

This confused me slightly, since it goes directly against what we experience
with this:

$ ruby -e 'module A; end; module B; include A; end; class C; include B; end;
class D < C; end; p D.ancestors'
[D, C, B, A, Object, Kernel]

.. that is, the modules included by a superclass's included module appearing
in the ancestors list. In the previous example, Object (superclass of C)
includes Kernel, and we don't see Kernel's included A. Here, C (superclass
of D) includes B, yet we also see A. So, that proves that theory wrong.

In frustation, and in trying to work this out, my test case became larger
and larger.. I think this provides about all the useful data we could hope
for:

$ ruby -e 'module M; end; module N; end; module O; end; module P; end;
module Q; end; module R; include Q; end; module M; include O; end; module
Kernel; include P; end; class Object; include R; end; class A; include M;
end; class B < A; include N; end; p B::ancestors'
[B, N, A, M, O, Object, R, Q, Kernel]

modules M, N, O, P, Q, R.
M includes O
Object includes R, R includes Q
Kernel includes P

class A includes M
class B derives from A, includes N

To start, [B, N, A ..] shows that the included items are first, followed by
the superclass.
[.. A, M, O, Object ..] continues to demonstrate this, and also tells us
that included items also have *their* included items. [M's O] Object is A's
superclass..
[.. Object, R, Q, Kernel ..] agrees with the above results, showing that
Object's included R is there, and R's included Q, and the Object's included
Kernel ..
[.. Kernel] then ruins things little, since Kernel's included P never makes
it.

I have a feeling Kernel's already being included by something (or it
deriving from something-or-other) -- I have no idea, really -- has something
to do with this. What interests me also is that Kernel::ancestors in this
test *does* return [Kernel, P].


This ended up being quite long and drawn out, and I hope this interested
you, or that you may have some advice.

Cheers,
Arlen.

 

[Joel VanderWerf]

I'm fairly sure I'm a long way down the path of madness with this one.
Please bare with me.

You may be stark, raving mad, but there's nothing to be gained by group
streaking.

SCNR. Maybe a few lols will help you solve the problem (or not), but
that's all I can offer at this hour.

 

[Robert Klemme]

This ended up being quite long and drawn out, and I hope this interested
you, or that you may have some advice.

Relax. Don't worry to much. Take a nap.

Kernel is special so you cannot expect ordinary mechanisms to apply. If
there was not end to inclusion / inheritance chain there would be
endless recursion (e.g. during method lookup). So the chain has to be
broken at *some* point.

Cheers

robert

 

[Calamitas]

Hi all,

I'm fairly sure I'm a long way down the path of madness with this one.
Please bare with me. I've been on this for some time now.

I've been trying to work out the order in which items appear in
"SomeClass::ancestors". I thought I had it before:

[SomeClass, SomeClass'sIncludedModules..., SuperClass,
SuperClass'sIncludedModules, SuperSuperClass, ...]

I constructed a test case:

$ ruby -e 'module M; end; module N; end; class A; include M; end; class B <
A; include N; end; p B.ancestors'
[B, N, A, M, Object, Kernel]

This is the expected result! But then I wondered.. what if we included
something in `M'? It shouldn't appear according to my theory here, since M
is not included by SomeClass itself. But, sure enough, it does appear.
(which makes sense!)

The reason I presumed this is because if I include a module in Kernel, it
won't appear in the list of any Class's ancestors, yet it will appear in
Kernel's:

$ ruby -e 'module A; end; module Kernel; include A; end; class C; end; p
C.ancestors; p Kernel.ancestors'
[C, Object, Kernel]
[Kernel, A]

This confused me slightly, since it goes directly against what we experience
with this:

$ ruby -e 'module A; end; module B; include A; end; class C; include B; end;
class D < C; end; p D.ancestors'
[D, C, B, A, Object, Kernel]

.. that is, the modules included by a superclass's included module appearing
in the ancestors list. In the previous example, Object (superclass of C)
includes Kernel, and we don't see Kernel's included A. Here, C (superclass
of D) includes B, yet we also see A. So, that proves that theory wrong.

You've bumped here into what some people call the double inclusion
problem or the dynamic inclusion problem. With modules in modules, the
order of the inclusions matters. A slightly reordered version of your
last example is this:

$ ruby -e 'module A; end; module B; end; class C; include B; end;
module B ; include A ; end ; class D < C; end; p D.ancestors ; p
B.ancestors'
[D, C, B, Object, Kernel]
[B, A]

See how A disappeared from the ancestors even though it is still
included in A? This is what you experience with Kernel: Kernel was
included in Object before you included your own module in it. This is
because of the way module inclusion is implemented in Ruby. In
database terms, Ruby's internal representation is denormalized for
efficiency reasons. The danger is inconsistency, which is what happens
here.

Note that you can force inclusion of recursively included modules by
reincluding the top-level module after recursive inclusions, like
this:

ruby -e 'module A; end; module Kernel; include A; end; class Object ;
include Kernel ; end ; class C; end; p C.ancestors'
[C, Object, Kernel, A]

The reinclusion of Kernel in Object doesn't result in it being
included twice (Ruby tries to prevent that from happening), but it
does update the recursive inclusion.

So basically, Kernel is nothing special in this regard, except that by
its predefinition it resulted in a different inclusion order than your
other test cases.

HTH,

Peter

 

[Arlen Cuss]

[Note: parts of this message were removed to make it a legal post.]

Hi Peter,

You've bumped here into what some people call the double inclusion
problem or the dynamic inclusion problem. With modules in modules, the
order of the inclusions matters.


The reinclusion of Kernel in Object doesn't result in it being
included twice (Ruby tries to prevent that from happening), but it
does update the recursive inclusion.


So basically, Kernel is nothing special in this regard, except that by
its predefinition it resulted in a different inclusion order than your
other test cases.

Amazing!

Thank you very much! You answered my (rather unnecessarily difficultly
worded) question perfectly! It is somewhat comforting to know this is an
implementation detail.

HTH,

Peter

Warmest thanks and regards,
Arlen