I'm trying to get my head around mutable vs immutable objects. Using mutable objects gets a lot of bad press (e.g. returning an array of strings from a method) but I'm having trouble understanding what the negative impacts are of this. What are the best practices around using mutable objects? Should you avoid them whenever possible?
Well, there are a few aspects to this.
Mutable objects without reference-identity can cause bugs at odd times. For example, consider a Person bean with a value-based equals method:
Map<Person, String> map = ...
Person p = new Person();
map.put(p, "Hey, there!");
p.setName("Daniel");
map.get(p); // => null
The Person instance gets "lost" in the map when used as a key because its hashCode and equality were based upon mutable values. Those values changed outside the map and all of the hashing became obsolete. Theorists like to harp on this point, but in practice I haven't found it to be too much of an issue.
Another aspect is the logical "reasonability" of your code. This is a hard term to define, encompassing everything from readability to flow. Generically, you should be able to look at a piece of code and easily understand what it does. But more important than that, you should be able to convince yourself that it does what it does correctly. When objects can change independently across different code "domains", it sometimes becomes difficult to keep track of what is where and why ("spooky action at a distance"). This is a more difficult concept to exemplify, but it's something that is often faced in larger, more complex architectures.
Finally, mutable objects are killer in concurrent situations. Whenever you access a mutable object from separate threads, you have to deal with locking. This reduces throughput and makes your code dramatically more difficult to maintain. A sufficiently complicated system blows this problem so far out of proportion that it becomes nearly impossible to maintain (even for concurrency experts).
Immutable objects (and more particularly, immutable collections) avoid all of these problems. Once you get your mind around how they work, your code will develop into something which is easier to read, easier to maintain and less likely to fail in odd and unpredictable ways. Immutable objects are even easier to test, due not only to their easy mockability, but also the code patterns they tend to enforce. In short, they're good practice all around!
With that said, I'm hardly a zealot in this matter. Some problems just don't model nicely when everything is immutable. But I do think that you should try to push as much of your code in that direction as possible, assuming of course that you're using a language which makes this a tenable opinion (C/C++ makes this very difficult, as does Java). In short: the advantages depend somewhat on your problem, but I would tend to prefer immutability.
Immutable Objects vs. Immutable Collections
One of the finer points in the debate over mutable vs. immutable objects is the possibility of extending the concept of immutability to collections. An immutable object is an object that often represents a single logical structure of data (for example an immutable string). When you have a reference to an immutable object, the contents of the object will not change.
An immutable collection is a collection that never changes.
When I perform an operation on a mutable collection, then I change the collection in place, and all entities that have references to the collection will see the change.
When I perform an operation on an immutable collection, a reference is returned to a new collection reflecting the change. All entities that have references to previous versions of the collection will not see the change.
Clever implementations do not necessarily need to copy (clone) the entire collection in order to provide that immutability. The simplest example is the stack implemented as a singly linked list and the push/pop operations. You can reuse all of the nodes from the previous collection in the new collection, adding only a single node for the push, and cloning no nodes for the pop. The push_tail operation on a singly linked list, on the other hand, is not so simple or efficient.
Immutable vs. Mutable variables/references
Some functional languages take the concept of immutability to object references themselves, allowing only a single reference assignment.
In Erlang this is true for all "variables". I can only assign objects to a reference once. If I were to operate on a collection, I would not be able to reassign the new collection to the old reference (variable name).
Scala also builds this into the language with all references being declared with var or val, vals only being single assignment and promoting a functional style, but vars allowing a more C-like or Java-like program structure.
The var/val declaration is required, while many traditional languages use optional modifiers such as final in java and const in C.
Ease of Development vs. Performance
Almost always the reason to use an immutable object is to promote side effect free programming and simple reasoning about the code (especially in a highly concurrent/parallel environment). You don't have to worry about the underlying data being changed by another entity if the object is immutable.
The main drawback is performance. Here is a write-up on a simple test I did in Java comparing some immutable vs. mutable objects in a toy problem.
The performance issues are moot in many applications, but not all, which is why many large numerical packages, such as the Numpy Array class in Python, allow for In-Place updates of large arrays. This would be important for application areas that make use of large matrix and vector operations. This large data-parallel and computationally intensive problems achieve a great speed-up by operating in place.
Immutable objects are a very powerful concept. They take away a lot of the burden of trying to keep objects/variables consistent for all clients.
You can use them for low level, non-polymorphic objects - like a CPoint class - that are used mostly with value semantics.
Or you can use them for high level, polymorphic interfaces - like an IFunction representing a mathematical function - that is used exclusively with object semantics.
Greatest advantage: immutability + object semantics + smart pointers make object ownership a non-issue, all clients of the object have their own private copy by default. Implicitly this also means deterministic behavior in the presence of concurrency.
Disadvantage: when used with objects containing lots of data, memory consumption can become an issue. A solution to this could be to keep operations on an object symbolic and do a lazy evaluation. However, this can then lead to chains of symbolic calculations, that may negatively influence performance if the interface is not designed to accommodate symbolic operations. Something to definitely avoid in this case is returning huge chunks of memory from a method. In combination with chained symbolic operations, this could lead to massive memory consumption and performance degradation.
So immutable objects are definitely my primary way of thinking about object-oriented design, but they are not a dogma.
They solve a lot of problems for clients of objects, but also create many, especially for the implementers.
Check this blog post: http://www.yegor256.com/2014/06/09/objects-should-be-immutable.html. It explains why immutable objects are better than mutable. In short:
immutable objects are simpler to construct, test, and use
truly immutable objects are always thread-safe
they help to avoid temporal coupling
their usage is side-effect free (no defensive copies)
identity mutability problem is avoided
they always have failure atomicity
they are much easier to cache
You should specify what language you're talking about. For low-level languages like C or C++, I prefer to use mutable objects to conserve space and reduce memory churn. In higher-level languages, immutable objects make it easier to reason about the behavior of the code (especially multi-threaded code) because there's no "spooky action at a distance".
A mutable object is simply an object that can be modified after it's created/instantiated, vs an immutable object that cannot be modified (see the Wikipedia page on the subject). An example of this in a programming language is Pythons lists and tuples. Lists can be modified (e.g., new items can be added after it's created) whereas tuples cannot.
I don't really think there's a clearcut answer as to which one is better for all situations. They both have their places.
Shortly:
Mutable instance is passed by reference.
Immutable instance is passed by value.
Abstract example. Lets suppose that there exists a file named txtfile on my HDD. Now, when you are asking me to give you the txtfile file, I can do it in the following two modes:
I can create a shortcut to the txtfile and pass shortcut to you, or
I can do a full copy of the txtfile file and pass copied file to you.
In the first mode, the returned file represents a mutable file, because any change into the shortcut file will be reflected into the original one as well, and vice versa.
In the second mode, the returned file represents an immutable file, because any change into the copied file will not be reflected into the original one, and vice versa.
If a class type is mutable, a variable of that class type can have a number of different meanings. For example, suppose an object foo has a field int[] arr, and it holds a reference to a int[3] holding the numbers {5, 7, 9}. Even though the type of the field is known, there are at least four different things it can represent:
A potentially-shared reference, all of whose holders care only that it encapsulates the values 5, 7, and 9. If foo wants arr to encapsulate different values, it must replace it with a different array that contains the desired values. If one wants to make a copy of foo, one may give the copy either a reference to arr or a new array holding the values {1,2,3}, whichever is more convenient.
The only reference, anywhere in the universe, to an array which encapsulates the values 5, 7, and 9. set of three storage locations which at the moment hold the values 5, 7, and 9; if foo wants it to encapsulate the values 5, 8, and 9, it may either change the second item in that array or create a new array holding the values 5, 8, and 9 and abandon the old one. Note that if one wanted to make a copy of foo, one must in the copy replace arr with a reference to a new array in order for foo.arr to remain as the only reference to that array anywhere in the universe.
A reference to an array which is owned by some other object that has exposed it to foo for some reason (e.g. perhaps it wants foo to store some data there). In this scenario, arr doesn't encapsulate the contents of the array, but rather its identity. Because replacing arr with a reference to a new array would totally change its meaning, a copy of foo should hold a reference to the same array.
A reference to an array of which foo is the sole owner, but to which references are held by other object for some reason (e.g. it wants to have the other object to store data there--the flipside of the previous case). In this scenario, arr encapsulates both the identity of the array and its contents. Replacing arr with a reference to a new array would totally change its meaning, but having a clone's arr refer to foo.arr would violate the assumption that foo is the sole owner. There is thus no way to copy foo.
In theory, int[] should be a nice simple well-defined type, but it has four very different meanings. By contrast, a reference to an immutable object (e.g. String) generally only has one meaning. Much of the "power" of immutable objects stems from that fact.
Mutable collections are in general faster than their immutable counterparts when used for in-place
operations.
However, mutability comes at a cost: you need to be much more careful sharing them between
different parts of your program.
It is easy to create bugs where a shared mutable collection is updated
unexpectedly, forcing you to hunt down which line in a large codebase is performing the unwanted update.
A common approach is to use mutable collections locally within a function or private to a class where there
is a performance bottleneck, but to use immutable collections elsewhere where speed is less of a concern.
That gives you the high performance of mutable collections where it matters most, while not sacrificing
the safety that immutable collections give you throughout the bulk of your application logic.
If you return references of an array or string, then outside world can modify the content in that object, and hence make it as mutable (modifiable) object.
Immutable means can't be changed, and mutable means you can change.
Objects are different than primitives in Java. Primitives are built in types (boolean, int, etc) and objects (classes) are user created types.
Primitives and objects can be mutable or immutable when defined as member variables within the implementation of a class.
A lot of people people think primitives and object variables having a final modifier infront of them are immutable, however, this isn't exactly true. So final almost doesn't mean immutable for variables. See example here
http://www.siteconsortium.com/h/D0000F.php.
General Mutable vs Immutable
Unmodifiable - is a wrapper around modifiable. It guarantees that it can not be changed directly(but it is possibly using backing object)
Immutable - state of which can not be changed after creation. Object is immutable when all its fields are immutable. It is a next step of Unmodifiable object
Thread safe
The main advantage of Immutable object is that it is a naturally for concurrent environment. The biggest problem in concurrency is shared resource which can be changed any of thread. But if an object is immutable it is read-only which is thread safe operation. Any modification of an original immutable object return a copy
source of truth, side-effects free
As a developer you are completely sure that immutable object's state can not be changed from any place(on purpose or not). For example if a consumer uses immutable object he is able to use an original immutable object
compile optimisation
Improve performance
Disadvantage:
Copying of object is more heavy operation than changing a mutable object, that is why it has some performance footprint
To create an immutable object you should use:
1. Language level
Each language contains tools to help you with it. For example:
Java has final and primitives
Swift has let and struct[About].
Language defines a type of variable. For example:
Java has primitive and reference type,
Swift has value and reference type[About].
For immutable object more convenient is primitives and value type which make a copy by default. As for reference type it is more difficult(because you are able to change object's state out of it) but possible. For example you can use clone pattern on a developer level to make a deep(instead of shallow) copy.
2. Developer level
As a developer you should not provide an interface for changing state
[Swift] and [Java] immutable collection
Say I have a value type Foo, and a method Bar which accepts a reference to a Foo. Most languages will allow me to allocate a new Foo on the stack, and will automatically box it when I try and pass it in to Bar. However, as far as I am aware, this involves copying the Foo value onto the heap, and then using that reference.
Is it possible for a language to include a way of allocating a garbage collected object on the stack? When the method ends, the runtime could check if the object is still in use, and only then would it need to allocate the object on the heap, and update the references.
I imagine this would improve performance for methods that do not keep the reference, and it would hinder performance for methods that do.
Yes, Graal's partial escape analysis does that. While regular EA can only stack-allocate (more precisely: decompose into fields, put fields onto stack) when the object doesn't escape partial EA can optimistically allocate on the stack and only reify the data into an object on uncommon cases where the object must exist.
Also note that garbage collection is not a binary choice. You can have environments that mix and match garbage-collection, ref-counting, arena or scope-based allocators with automatic deallocation and completely manual management. In such a case stack allocations could also be one of the latter things while some heap would be garbage-collected.
Summary of question and answers
Objects of a particular type, say
type Foo
a::A
b::B
end
can be stored in either of two ways:
Inlined (aka by value): in this case, the statement "variable foo::Foo is stored at location x" effectively means we have a variable foo.a::A at location x and a variable foo.b::B at location x + sizeof(A) (technically the addresses could be a bit more complicated, but that's irrelevant for our purposes).
Referenced (aka by reference): "foo::Foo is stored at location x" means the location x contains a pointer fooptr::Ptr{Foo} such that there is a variable foo.a::A at location fooptr and foo.b::B at location fooptr + sizeof(A).
Unlike other languages (I'm looking at you, C/C++), Julia decides by itself whether to store variables inlined or referenced, and it does so based on the properties of the type:
mutable types -> referenced,
immutable types -> referenced if at least one of its fields is referenced, inlined otherwise.
There are at least two reasons for this rule:
StefanKarpinski's answer: The garbage collector needs be able to find all pointers to heap-allocated objects on the stack. Currently, Julia ensures this by storing all such pointers on a separate "shadow stack", but if we allowed composite types containing pointers to be placed on the stack then such a neat separation would no longer be possible. Instead, the compiler would need to look for pointers among other variables which poses technical difficulties.
yuyichao's answer: Julia requires the inline/reference decision to be made on a per-type rather than per-object basis, which means a hypothetical type
immutable A
a::A
end
would have to be infinitely big if we insisted on inlining it. So we would either have to forbid such recursive immutable types, or we could at most allow non-recursive immutable types to be inlined.
Original question
My understanding of memory management in Julia is:
mutable types -> heap-allocated,
immutable types and tuples -> stack-allocated unless one of their fields is heap-allocated (i.e. mutable).
I don't quite understand the rationale for this behaviour, however. I've read somewhere that the problem with stack-allocating immutables with pointers to mutables is that then the garbage collector might consider the mutables unreachable and destroy them prematurely. On the other hand, if we place the immutable on the heap then there will still be a pointer to the mutables, so it might seem like we avoided the problem, but actually we just shifted it to making sure that now the immutable itself will not be destroyed.
Can anyone explain this to me who has only very superficial knowledge of how garbage collection works?
The problem with stack-allocation of objects which reference other objects is knowing that they need to be traced during garbage collection. The simplest way to do this is what Julia does: heap allocate the objects and "root" them using "shadow stack" which is pushed and popped in sync with the actual stack. This introduces a fair bit of overhead and forces these objects to be heap allocated.
A more sophisticated approach that avoids the overhead of a shadow stack and heap allocation is to stack allocated these objects and then scan the stack which doing garbage collection and follow references from objects in the stack to objects on the heap. However, this requires knowing which objects in the stack are pointers to objects on the heap – in general, non heap-allocated objects are not guaranteed to be kept intact or contiguous in registers or the stack. One approach to doing this is called "conservative stack scanning" which entails assuming during gc that any value on the stack which looks like it could be a pointer to an object on the heap actually is. That approach has been successfully used in applications like Safari's JavaScript engine, but it's not without it's challenges. We've contemplated using conservative stack scanning in Julia, and an initial effort to do so was started but the effort was never completed.
References:
https://github.com/JuliaLang/julia/issues/11714
https://github.com/JuliaLang/julia/pull/8134
There are multiple issues/concepts that are frequently mixed together whenever this is brought up.
mutable or non-pointerfree immutable doesn't necessarily mean heap allocation, we already have optimization passes to elide some of the optimizations and are working on improving them further.
The object layout ABI is an user visible behavior and not something an optimization pass can easily change (unless it can prove that the local optimization it wants to do does not escape). The current ABI is that only isbits immutable will be stored inline (and "stack allocated" when used as local variable). There's a fundamental limitation of lifting the requirement of pointerfree-ness for inlined object, i.e. the necessity to handle recursive types. It is impossible to make all types in a reference circle stored inline and the loop has to be broken somewhere if we want to make some of them inlined. I believe we do have a consistent and predictable model to do this though whether this is desireable is another issue.
This is somewhat related to performance but not always. Stored inline means more copy so it's hard to make sure there's no regression if we do the switch.
Edit: And I should also mention that pointer-free is a sufficient condition for cycle free and is easier to compute, which is partly why we are currently using it to break inlining cycles.
GC support. This is basically the easiest part. It's very easy to make GC recognize pointers on the stack. It just needs to be done if we decide to change the object layout ABI.
Edit: And I should add that "GC support" is needed because we currently only support a limited / simple stack layout for object reference (i.e. an array of pointers). It's this that needs to be improved.
I understand what weak tables are.
But I'd like to know where weak tables can be used practically?
The docs say
Weak tables are often used in situations where you wish to annotate
values without altering them.
I don't understand that. What does that mean?
Posted as an answer from comments...
Since Lua doesn't know what you consider garbage, it won't collect anything it isn't sure to be garbage. In some situations (one of which could be debugging) you want to specify a value for a variable without causing it to be considered "not trash" by Lua. From my understanding, weak tables allow you to do what you'd normally do with variables/objects/etc, but if they're weak referenced (or in a weak table), they will still be considered garbage by Lua and collected when the garbage collection function is called.
Example: Think about if you wanted to use an associative array, with key/value pairs in two separate private tables. If you only wanted to use the key table for one specific use, once you are done using it, it will be locked into existence in Lua. If you were to use a weak table, however, you'd be able to collect it as garbage as soon as you were done using it, freeing up the resources it was using.
To explain that one cryptic sentence about annotating, when you "alter" a variable, you lock it into existence and Lua no longer considers it to be garbage. To "annotate" a variable means to give it a name, number, or some other value. So, it means that you're allowed to give a variable a name/value without locking it into existence (so then Lua can garbage collect it).
Translation:
Weak tables are often used in situations where you wish to give a name to a value without locking the value into existence, which takes up memory.
Normally, storing a reference to an obect will prevent that object from being reclaimed when the object goes out of scope. Weak references do not prevent garbage collection.