Talk:Flyweight pattern

Bug in Java example?
Note: I just fixed the bug described below. —Preceding unsigned comment added by 74.68.108.87 (talk) 21:59, 8 April 2010 (UTC)

I believe there's an intermittent bug in the Java example. Here's the part I'm concerned about, in the create method:

Here's the problem: Suppose you call the create method with a certain set of parameters, for the first time. The method constructs a new FontData object (call it d0), and the local variable data refers to it. So, d0 is added to FLY_WEIGHT_DATA, and the method returns d0. No problem so far.

Now imagine that the return value of create is discarded, i.e., there are no more references to d0 except for the weak references in FLY_WEIGHT_DATA. Suppose that the garbage collector has not yet run, and that you call create again, with the same parameters that gave you the original d0. The lines quoted above construct a new FontData object that equals d0, but it's not identical to d0. FLY_WEIGHT_DATA.containsKey(data) returns true because data.equals(d0) is true. Now suppose that the garbage collector kicks in after this check, but before the return statement. Now FLY_WEIGHT_DATA.get(data) will return null because the key d0 has disappeared, and so FLY_WEIGHT_DATA.get(data).get will fail with a NullPointerException.

Another kind of failure occurs if the garbage collecture runs after get(data) but before get. In that case, get(data) will return the desired weak reference, but then get will return null, presumably leading to a NullPointerException somewhere down the road.

By the way, the documentation of WeakHashMap warns of exactly this kind of problem (see the paragraph on checking for equality with ==, as well as the paragraph after that). —Preceding unsigned comment added by 74.68.108.87 (talk) 15:58, 4 March 2010 (UTC)

comment
Flyweight is a software design pattern.

The sentence above is how the articles title "builder pattern", "adapter pattern", etc. should begin. Someone who is ignorant of computer software looking at those articles should be told right at the beginning that that is what the article is about. -- Mike Hardy

Chnged the method printAtPosition to a static one as the point of a flyweight object was to save memory. -- Jari Mustonen

each GraphicChar takes 10 bytes of heap space (8 overhead + 2 char + 4 string pointer for font) 8 + 2 + 4 = 14, not 10 --195.6.224.137 15:17, 18 October 2006 (UTC)

I'm removing the two paragraphs that start "Often the desire to use a flyweight pattern indicates another design problem" because they seem way off-topic, they contain errors (see comment above regarding addition error), and they sound a lot like original research. CMosher01 17:16, 24 October 2006 (UTC)

As the author said that this is structural pattern, According to me its not structural or object oriented pattern, it's a hybrid pattern which consist both kind of pattern.--- Sanket Patel

Thread safety issue
The article should outline the importance of being thread-safe when the Flyweight is static. The Java example should also be thread-safe, some one who doesnt know this pattern may fall in this trap. —Preceding unsigned comment added by 81.252.131.1 (talk) 09:11, 15 April 2009 (UTC)

Memory issue in the Java example
The Java example adds WeakReferences into the flyweightData map but at no point removes them. This could in time lead to a very large map, whose values would mostly end up being null-valued WeakReferences.

Actually it doesn't because the map used is a WeakHashMap and the keys are the same as the referenced value. This approach works nicely when object creation cost is low and the object itself can be used as the key.

When object creation is high and needs to be avoided as well as keeping memory usage down some sort of other key needs to be used. If the object's hashCode can be replicated with the same parameters given in the constructor then the hashCode is a good choice. But now you definitely have the problem of the map potentially growing large with null-valued WeakReferences. The solution is to attach a ReferenceQueue that removes the entry from the map.

org.apache.commons.collections.map.ReferenceMap provides a basic solution for this, but there exists a superior solution, also open sourced, here http://sesat.no/projects/sesat-commons/commons-reference-map/

> This implementation improves over org.apache.commons.collections.map.ReferenceMap > in that the synchronisation and concurrency is determined through delegation to the > map supplied in the constructor as described above.

Michaelsembwever (talk) 21:48, 20 March 2008 (UTC)

VB.Net example: Is this a flyweight?
I removed this section from the article:

VB.NET
To me it doesn't look like a Flyweight. It may be a pattern to save memory, but to me it doesn't look anything like the Flyweight pattern. Am I wrong? --Teglsbo (talk) 12:50, 21 April 2008 (UTC)

That is the use of a proxy pattern. The Flyweight pattern is there to prevent needing a cartesian product number of classes, not to prevent accessing a huge class. Loek Bergman (talk) 10:56, 27 June 2009 (UTC)

Illustration
The illustration seems to illustrate an object pool, not a flyweight setup. While object pools can be handy when dealing with flyeights, it isn't directly relevant to the design pattern. The illustration outright fails at communicating the concept correctly.--Henke37 (talk) 19:07, 8 July 2012 (UTC)


 * While we're on the subject, i would say that every code example on the page indicates to me that their authors did not understand this pattern. They do not illustrate the pattern as described at all. I am inclined to remove them. --Rob* (talk) 08:17, 14 November 2015 (UTC)

Python __slots__ example
The construct '__slots__' in Python does not, as the article states, 'automatically enable the Flyweight pattern for that class.' It simply removes the backing dictionary ('__dict__') in favor of a tuple. As the definition of Flyweight states at the beginning of the article, the requirement for this pattern is 'sharing as much data as possible with other similar objects.' Two slotted Python classes initialized to the same value will be different objects. 'arr=[foo(1) for i in range(1000)]' will contain 1000 different objects, not 1000 references to the same object. NO PYTHON OBJECT IS IMMUTABLE. For this reason, I don't believe this Python example is valid and should be removed. — Preceding unsigned comment added by 207.10.138.156 (talk) 22:38, 24 November 2013 (UTC)

A valid Python example would be getting the immutable attributes from a singleton, via __getattr__, and denying its override via __setattr__. But would be simpler to use class-level attributes. This ones resides on the class space and are not copied over instances. However this ones could be overwriten if not denied via __setattr__ — Preceding unsigned comment added by Alanjds (talk • contribs) 14:49, 31 January 2017 (UTC)

Java example is a joke
Can somebody please fix this? The complexity, BMW advertising and lack of clarity (say Vs C# original Coffee example) is astounding. The original Coffee example, pre BMW, whilst a lot shorter and simpler is also flawed, using ConcurrentHashMap but then also using synchronize everywhere. It's like nobody has read Concurrency In Practice/even bothered learning that those 2 APIs are intended to be (mostly) mutually exclusive replacements for each other. — Preceding unsigned comment added by 2A02:C7D:31E1:4600:41DD:FF41:6751:5860 (talk) 09:46, 14 October 2018 (UTC)

C# example is confusing
What's up with class ReducedMemoryFootprint in the C# example? It appears it does the same thing as MinimumMemoryFootprint only worse, and is harder to read and understand. The class names are confusing, as you'd expect MinimumMemoryFootprint to use less memory than ReducedMemoryFootprint, but it doesn't. There's no text to explain the code; not even comments. The private method AddFlavourToCache(object state) takes an object instead of a CoffeeFlavour for no discernible reason. I think the example should be removed, leaving just the "Simple Implementation" example. Am I missing something?

-

I agree strongly. Adding cache locks for concurrency seems particularly confusing for anyone who just wants to see what the pattern looks like in action. If nobody objects, I'll remove the large example soon. Telemachus12389 (talk) 11:30, 9 June 2021 (UTC)

Implementation details section is unsourced, complex, and basically unreadable
I've made some minor edits to remove language in this section that sounds like it came from a tutorial (lots of 'you should do this in your application...'). But in general I'm unsure most of the detail here is really relevant - can someone with deep knowledge on the topic highlight any sections that it would be fine to remove in the sake of concision and clarity?

This also seems to be common on other design pattern pages, but I'm not sure what's gained by having examples from nine different languages - especially when some of them, like Java and C#, are so similar. The examples take up over three-quarters of the page, making it feel more like a reference manual than an encyclopedia page. Can any examples by removed? Telemachus12389 (talk) 11:28, 9 June 2021 (UTC)

Large changes, feel free to revert if felt necessary
I've gone ahead and cut out big swathes of the implementation details section and several of the code examples. The page still doesn't have enough citations for the implementation details section, but hopefully it now reads more like an encyclopedia page and less like a design manual. Deciding which language examples to cut out was ultimately arbitrary, of course, but I decided to remove some of the slightly less popular languages (Scala, Crystal) and the Java example, since the language is similar to C#.

I'm aware these are big changes, so feel free to revert the edits. But if you do, I would appreciate hearing why you think the material was worth bringing back, as it would help contextualise its place in the article and improve the piece overall.

Telemachus12389 (talk) 11:31, 12 June 2021 (UTC)

PHP example - cache implementation using WeakReference
I think the way how WeakReference object is used in CoffeeFlavour::intern function does not make sense. The WeakReference object should be stored in the cache. In the example classical reference (not weak reference) is stored in the cache which means it will be preserved in the cache all the time. The function should also deal with situation when weak reference has already been forgotten and should refresh it in that case. For that reason, I think the function should return the object itself rather than WeakReference.

Here is an example fix of the function.

Secondly, we should also remove ->get from Order::of function, because ::intern function already returns the object itself now.

See also this example of WeakReference usage (example no. 2) which is similar to my fix.

Mk90461 (talk) 14:02, 20 July 2021 (UTC)

Python example - wrong and off topic?
The Python example seems to be based on an incorrect understanding of the pattern. It's simply inserting and retrieving from a shared dictionary, but a shared dictionary by itself is not a flyweight pattern. Besides it seems to be largely original research and contains a large off-topic discussion of immutability in Python's data model. I'm WP:BOLDLY removing the section as the other examples illustrate the concept better. 2A02:A44A:5C96:1:D66A:D92F:5729:7268 (talk) 12:14, 30 April 2023 (UTC)