Re: Are Singletons Worth Using?

From: Ilja Preuß (preuss_at_disy.net)
Date: 07/19/04 

Date: Mon, 19 Jul 2004 19:11:18 +0200

Bruno Desthuilliers wrote:
> Ilja Preuß wrote:
>> Bruno Desthuilliers wrote:
>>
>>
>>> Firstly, this seems quite specific to a certain category of
>>> languages... In hi-level languages, you can have 'transparent'
>>> Singletons. Look here for a Python exemple :
>>> http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/52558
>>
>>
>>> We're in the 21th century now, it would be time for some people
>>> around
>>> to realize that OO does'nt imply Java...
>>
>>
>> Why couldn't the same be implemented in Java??? :confused:
>
> That's a good question. Rereading the code, I'm not sure it couldn't.
> Ask a Java expert...

Well, I am a Java expert. ;)

If I understand the code correctly, it could be implemented in Java in the
same way. But unfortunately I am no Python expert, so I could be missing a
subtle point...

> Anyway the current Java idiom for Singletons seems to be this
> class::getInstance() stuff.

Well, that's what we are used to because of the GoF book, I guess. The
MonoState pattern simply is much less well known, probably because every
book and course on patterns typically just is another take on GoF...
:rolleyes:

>>>> The
>>>> question is whether the extra coupling is a price worth paying for
>>>> enforcing the one and only one instance rule?
>>>
>>> Now this 'extra coupling' is not a Singleton-or-not-singleton issue,
>>> it's a general design problem about concrete class knowledge's
>>> responsability. This issue arise whenever an object instanciate
>>> another one, be it a singleton or not. And then again, there are
>>> patterns to
>>> help solve this (AbstractFactory etc...).
>>
>>
>> The Singleton pattern is not about instanciation, but about a global
>> access point to a single instance.
>
> And ? please re-read more carefully !-)

I did, but it didn't help... ;)

Ah, doing it the fourth time helped, possibly. :)

Yes, the same coupling happens when you call a constructor instead of the
getInstance method (so your phyton solution actually doesn't solve *this*
problem ;). Therefore in some circumstances I would prefer to pass around
the one instance instead of providing a global access point.

Does that compute?

> What this paper mainly says is that
> """
> singleton are bad because
> 1/ the client class has to know that it is using a singleton because
> in
> Java you [have to | usually] use some kind of class::getInstance()
> syntax 2/ the client class knows the concrete class of the singleton
> """
>
> (Well, that's at least what I understood...)
>
> So I was pointing out that
> 1/ may be a Java limitation but has nothing to do with the Singleton
> pattern itself

I don't think this is true - the implementation presented in the above
recipe simply *is not* a canonical Singleton, but some enhanced variation.
The Singleton pattern, as defined in the GoF book, inherently has this
limitation. The Python recipe actually looks much more like a MonoState than
a Singleton. They both solve similar problems, but in different ways.

http://c2.com/cgi/wiki?MonostatePattern

> 2/ is a not a Singleton issue, but an instanciation issue

It's a "global access point" issue, which constructors are a subset of. And
it's amplified by the Singleton pattern, as there is only one instance of
the class (that is, only one single instanciation involved), but typically
dozens to hundreds of calls to the getInstance method (the globale access
point).

Cheers, Ilja

Relevant Pages

  • Re: Do I overuse class methods?
    ... In java I use instance methods for two purposes - maintaining object state ... As a compromise Michael could use module Singleton and thus define ... time instances need state. ...
    (comp.lang.ruby)
  • Re: singleton vs static
    ... > Disadvantages of Singleton ... pointer for the static instance. ... Are you referring to threading issues with getInstance? ... a surprize to many C++ and Java programmers that the standard Singleton ...
    (comp.object)
  • Re: Java Singleton Constraint?
    ... it would seem that Java would not ... >allow a class to be derived from a Singleton. ... >that you must never have a need to extend the Singleton class? ... "The aim of science is not to open the door to infinite wisdom, ...
    (comp.object)
  • Re: Java Singleton Constraint?
    ... I have a Singleton question that is specific to Java ... However, since this is a design pattern question, I ...
    (comp.object)
  • Re: Are Singletons Worth Using?
    ... Looks like a mix of Singleton and MonoState, ... Why couldn't the same be implemented in Java??? ... >> question is whether the extra coupling is a price worth paying for ... The Singleton pattern is not about instanciation, ...
    (comp.object)