mgaunard a day ago

There are several advantages to intrusive node-based data structures that I haven't seen stated in the article or the comments:

- the same object can move between containers with no allocation and no need for a dedicated complex API

- the same object can be part of multiple containers at once; particularly useful for intrusive binary trees, for indexing data with different criteria.

- the container can be fully polymorphic, no need for all elements to be the same dynamic type.

- no need for complex allocators, you can just store the objects as you see fit.

  • suspended_state a day ago

    I think the main advantage of the intrusive definition is that you can use it to implement the non intrusive one easily, whereas the reverse isn't possible.

    • NobodyNada a day ago

      With a non-intrusive linked list type, can't you define the intrusive version as `Node<void>`?

  • boomlinde a day ago

    > the same object can move between containers with no allocation and no need for a dedicated complex API

    This is true also of the previous design.

    > no need for complex allocators, you can just store the objects as you see fit.

    Same here; the previous design leaves it up to the user.

  • paraboul a day ago

    > the same object can be part of multiple containers at once

    I'm not sure I understand this one. Since the object contains the reference to where it belongs inside a container (e.g. object.node.next) how can it be re-used in multiple containers. Conversely, in a non-intrusive data structure, multiple containers can hold a ref to the same object through an intermediate node object

    • gnubison a day ago

      You add multiple next variables. buffer.next, buffer.next_displayed, etc

      • paraboul a day ago

        That's not an advantage of intrusive data structures then. That's precisely an advantage of non-intrusive data structure : object can be inserted in an arbitrary number of containers

        • mikepurvis a day ago

          But for the intrusive one you have the visibility into its membership in the other structures, since you're holding a reference to the object and its intrusive fields.

          In the non-intrusive case, all you know is how you got there, you don't have return info to the other structures— particularly annoying if you're trying to destroy an object and have to speculatively look for references to it in the other lists in order to get rid of them.

          • paraboul 10 hours ago

            Of course. But I wouldn’t qualify "multi-container use" to be an advantage of intrusive data structures per se.

    • Zambyte a day ago

      The object can contain multiple intrusive node fields.

  • MrCroxx a day ago

    Agreed. Intrusive data structures are good for implementing multi-container data structures with lower allocation overhead. And they are widely used than expected. E.g. LRU is a classic multi-container problem.

  • grandempire a day ago

    Is there any generic implementation which is not intrusive? I expect C++ forward_list to look like

    struct Node<T> { Node<T> *next; T x; }

    • _huayra_ a day ago

      At least in C++ land, that is not quite what is referred to as intrusive lists. It's basically the opposite / inside-out of what you wrote:

      ```C++ struct MyType : Node<MyType> { Node<MyType> next, prev; // rest of data }; ```

      I usually reach for Boost.Intrusive when I need this [0].

      [0] https://www.boost.org/doc/libs/1_88_0/doc/html/intrusive/usa...

      • grandempire a day ago

        I see. I am noticing the main difference is that forward_list manages the lifetime and allocation of nodes and that having a pointer to an object is equivalent to a pointer to the node (could be implemented as a helper).

        The layouts look the same?

        • _huayra_ a day ago

          At the byte level, it's quite possible the layouts are the same. However, an "intrusive data structure" means that the nodes themselves are the data.

          In other words, intrusive is like this:

          struct T : NodeType<T> { // T data NodeType<T> next, prev; };

          whereas non-intrusive data structures the T's are not the nodes themselves

          struct NodeType<T> { NodeType<T> next, prev; T t; };

          Doing non-intrusive means you need to own the lifecycle of the nodes (and make them copyable, move-constructible, or some combo thereof). Intrusive means that the caller can manage the nodes' lifecycles and just say "here's a node, I promise it'll live while the list is still alive" and have that node be on the stack, heap, static, etc.

  • cheezebubba a day ago

    How can you have elements of different types? I don't understand how you would know which type a given node was in?

    • throwawaymaths a day ago

      it was maybe easier to understand when @fieldParentPtr took the type as an argument, but if you look at this:

         var x: *T = @fieldParentPtr("node", node_ptr);
      
      @fieldParentPtr knows that its result must have type pointer T; and then the builtin function "looks" inside of T and notices there is a "node" field. Then it can backtrack the pointer of T given the pointer to the node field, since the layout of T is well-defined at compile time.

      So you could have a mixed-type linked list, and at the callsite zig can at compile-time back out the parent type. There obviously needs to be outside logic (and an outside mechanism to track) this with branches for each type that you could be putting into the linked list. And yes, this is more than a little bit type-unsafe because of the potential for type erasure.

      You could probably pretty trivially write a typesafe wrapper for this, as suggested in one of the comments here.

      • cheezebubba a day ago

        Thanks. I still don't understand where the bits live that tell you which type T is for a given node.

        Is this something like "list starts with type A, and that will tell you if the next node is type A or B"?

        Is there a way to stuff those bits in with the node somehow so you always know the type you expect back?

        • throwawaymaths a day ago

          > Thanks. I still don't understand where the bits live that tell you which type T is for a given node.

          Yup. You'd have to set that up yourself. It could be anywhere. It could be in a separate array, it could be in the previous node-data, etc.

          you wouldn't put it in the Node (capital N) datastructure though, that is privileged.

          Personally, I would say doing something like this is "bad idea bears" unless you are absolutely certain you need it and you should explore using, say, data payload type with a union field instead, if possible (though know that that comes with a potential memory penalty)

  • rowanG077 a day ago

    All of these sound like grievous sources of bugs to me. Sure it's nice that it's possible but please don't do those things unless there isn't a safer way to do it.

    • Diggsey a day ago

      They are definitely more error-prone usage patterns, and coming from Rust the amount of hidden unsafety is quite unsettling (in addition to the obvious potential for user errors, depending on the aliasing model of your language/compiler, which I suspect Zig hasn't figured out yet, this parent pointer trick could even lead to miscompilations...)

      Having said that, intrusive linked lists are quite an important data structure in code which is required to be allocation-free. Kernel code, signal handlers and code which requires consistent performance (eg. audio processing) need to be able to move objects between containers or have objects in multiple containers without allocating.

      • bananabiscuit a day ago

        Do you mean you will likely make an error when implementing a standard textbook linked list? Or that a standard textbook linked list has non-obvious errors in its design? If so, would be curious to learn more.

        • simonask a day ago

          Most programmers can implement a linked list.

          Extremely few can use one safely.

          They're an elegant weapon for a more civilized age - namely one without multithreading, aliasing-based optimization, and a host of other nice things.

          • tauoverpi a day ago

            Multithreading is exactly where linked lists are useful as in the implementation of MPSC, MPMC, SPMC, and so on it allows one to safely construct objects out of vire of other threads then use a CAS to append the new item to the list making it visible for other threads to consume even without needing to allocate additional memory.

            They're also pretty useful for free-list structures which overwrite entries to be removed (they're scheduled for deletion thus assumptiong here is that the object is now dead) with linked list nodes that would fit in that space thus never needing to allocate any additional as you can reuse existing space.

            • simonask 16 minutes ago

              Yeah, wait-free data structures often contain linked lists. That doesn’t cancel out all of the other problems.

              Anecdotally, I’ve seen wait-free structures used incorrectly many more times than I’ve seen them used correctly. Some people think they are magically faster because “acquiring a mutex lock is expensive”, but that’s far from always the case.

        • im3w1l a day ago

          Implementing anything is more error prone than using something battle-tested.

      • throwawaymaths a day ago

        Maybe advocate for wrappers in the stdlib to implement typesafe. LL, typesafe extrusive LL, etc. Using this as a primitive?

      • rowanG077 a day ago

        Use a separate data structure and then index into it with a linked list like API. No need to do something so dangerous if you simply need an allocation less linked list interface.

        But yes there is a place for these data structures. It just shouldn't be common.

        • Diggsey a day ago

          If you just mean using array indexes instead of pointers, that can work in some cases, but can run into concurrency issues when the container needs to be resized. With pointers to nodes, new nodes can be allocated and old nodes freed on a separate thread without interfering with the performance sensitive thread, whereas resizing an array requires all access to stop while the backing memory is re-allocated.

          I don't think you mean this, but elaborating on why a non-intrusive linked list is not very useful, that's because it takes linear time to find an object in a non-intrusive linked list, but constant time in an intrusive linked list.

          For example:

              struct Node {
                  Node* prev_by_last_update;
                  Node* next_by_last_update;
                  Node* prev_with_owner;
                  Node* next_with_owner;
                  Payload data;
              }
          
          This intrusive structure allows reassigning and updating nodes, whilst being able to efficiently find eg. the next most recently updated node, or to iterate over nodes with the same owner.
    • tauoverpi a day ago

      You don't need everything to be generic to have safety as the API provided by the object using the linked list can maintain safety by making it difficult to do the wrong thing and being correct by construction. Of course that doesn't stop a user of a library ignoring the safe way to handle types which use intrusive linked list nodes but that same user is likely going to shoot themselves in the foot regardless even when given a "safer" interface.

      Trying to write a safer version of the above without losing the performance / advantages takes you into the realm of ATS2 https://www.cs.bu.edu/~hwxi/atslangweb/ which quickly becomes more difficult to work with than practical (and no, rust can't solve this with it's implementation of resource types as far as I'm aware).

      So yes, doing the above makes sense when the problem demands it as allocating memory you don't need when there's a clear way to represent a graph without it just wastes cycles on memory management you didn't even need to do.

    • immibis a day ago

      bugs* ironic

      It depends on whether you approach programming from a perspective of writing piles of obviously correct abstraction layers (the total effect of which is to leak so much the solution ends up being wrong anyway), or from the direct unabstracted perspective of actually making the computer do what you want it to do (which does not preclude the solution from using abstractions).

      I call them the Uncle Bob vs Casey Muratori school, after Casey's lectures such as "Clean Code, Horrible Performance": https://www.youtube.com/watch?v=tD5NrevFtbU

      • rowanG077 a day ago

        Well uncle bob gives good advice. If you do exactly the opposite of what he tells you to do. so I don't think this comparison is apt at all.

        • immibis a day ago

          Can you elaborate on why he's a hack and perhaps how this relates to the idea of forcing yourself to use highly constrained inefficient "safe" data structures?

          • rowanG077 a day ago

            Because the excessive OOP and TDD he recommends is just bad. He also hates static typing and seems to think testing can somehow replace it.

            I'm not for forcing inefficient safe data structures. You are pretending that this is a dichotomy when the vast majority of the time it isn't. You are completely misinterpreting what I said. I'm for safe data structures. And the data structures described are huge footguns. I made no point about efficiency.

            • Ygg2 a day ago

              In programming and life, both extremes can be equally wrong. I'm not saying Uncle Bob is the shining beacon to steer your boat towards at all cost, but neither is Casey Muratori.

              > He also hates static typing

              I mean, in OOP there are many reasons to dislike it. It's harder to extend and test. Calls to such functions within other methods are a nuisance to remove. E.g. `Instant.now()`.

              • rowanG077 a day ago

                I'm sorry I'm not familiar with Casey Muratori. So I have no clue what he says.

                • Ygg2 a day ago

                  He's very data and performance oriented. He gives advice on how to write games, high-performance systems. His advice is solid and great even, but not when your source of latency is a cloud DB.

                  • immibis a day ago

                    I beg to differ even on that. When your source of latency is a cloud DB, learning about CPU caches won't help you, but learning to think about end-to-end performance absolutely will. "What sequence of queries achieves my goal with the shortest critical path?" is very similar to "What sequence of vector instructions achieves my goal with the shortest loop-carried dependency?"

                    • Ygg2 11 hours ago

                      > but learning to think about end-to-end performance absolutely will.

                      In a cloud DB webservice what will save you is having a good tracing/debugging story. From it, it will be easy to see E2E performance. Sure, some things will translate, but going SIMD optimized layout on a class, that will have to talk to DB across the ocean is a fool's errands.

                      To optimize you must have a good target and make sure you are measuring correctly.

      • Ygg2 a day ago

        > do what you want it to do

        Until a colleague or future you, or someone else calling the code, forgets.

        Which leads to "I WANT TO KILL WHOEVER WROTE THIS" debugging sessions.

        Encapsulation exists for a reason.

mastax a day ago

> The new version isn't generic. Rather, you embed the linked list node with your data. This is known as an intrusive linked list and tends to perform better and require fewer allocations.

I don’t get this. In the generic version the data is embedded with the linked list node, so there’s only one allocation per node. ‘T’ isn’t a pointer (or doesn’t have to be) there’s no indirection between data and node.

  • mppm a day ago

    It would probably have been more correct to say "requires fewer allocations in some cases". As you point out, in terms of layout, the old generic version is just as intrusive as the new version, and requires just as many allocations (one). However, the new version gives extra flexibility for you to move a long-lived object in and out of lists without copying or allocating, at the cost of having the pointer field baked into it permanently, rather than on demand.

    I think the reasoning behind this change is that (from Zig's perspective), if you are using linked lists you are probably doing something wrong, unless your application requires the above-mentioned kind of juggling, which favors explicitly intrusive LLs. In addition, this change de-generifys the lists's methods, like `prepend`, which reduces code size a little.

    At least that's my understanding.

    • DarkWiiPlayer a day ago

      > However, the new version gives extra flexibility for you to move a long-lived object in and out of lists without copying or allocating

      You could also do this with the entire node in the case of a generic implementation though, the API just needs to expose the node struct to you and allow you to detach it; but the same is true for this new implementation as well.

      In terms of memory, a detached node that embeds the payload struct isn't different from an object with an embedded detached node.

      What changes is that now, if you have an object class that you don't want to (or can't) extend to include a list node, you have to wrap it in a container struct that, again, looks the same in memory but now has a node and your object as its members. I'm not sure if this is really much of an improvement at the end of the day.

      Also, correct me if I'm wrong (I don't do zig), but shouldn't it be possible to create a list of void elements from the generic implementation and embed its node type inside your object, then proceed from there as if it were the new implementation?

      • mppm a day ago

        Yeah... with bitcasts and some creativity (and some boilerplate) both versions are ultimately equivalent, or nearly so. But the new one pushes you towards intrusive-data-structure thinking and away from container-of thinking.

        This, by the way, is a typical design consideration in Zig -- using "friction" to steer programmers away from doing the Wrong Thing (according to Andrew). In addition, Zig is really more of a fancy macro-assembler for outputting optimal code, and less a pragmatic general-purpose language, and makes design decisions accordingly. Taking both together, the linked-list change sort of makes sense, even though personally, I would have just added a separate intrusive list data structure.

  • torginus a day ago

    I remember reading an article about this technique - it was used in the original Starcraft. The idea here, is that the object needs to be allocated anyway by someone else, so you get the linked list for free.

    An object can also be part of multiple lists, each used by a different subsystem with its own concerns.

    Deleting an item involves unlinking it from all lists (though that requires a doubly-linked list)

    Edit: found the article

    https://www.codeofhonor.com/blog/tough-times-on-the-road-to-...

  • CBLT a day ago

    > only one allocation per node

    I believe the implication is there's fewer than one allocation per node with the new API. You allocate contiguous memory once, and use it to store n elements.

    • flowerthoughts a day ago

      No, that's not stated.

      > The new version isn't generic. Rather, you embed the linked list node with your data. This is known as an intrusive linked list and tends to perform better and require fewer allocations. Except in trivial examples, the data that we store in a linked list is typically stored on the heap. Because an intrusive linked list has the linked list node embedded in the data, it doesn't need its own allocation.

      Is simply a misunderstanding. The storage layout will be the same for the generic and the intrusive one.

      The benefit of intrusive linked lists is that each node can be a member of several linked lists with a single allocation per node. This is used heavily e.g. in the Linux kernel.

      The cost is that you need to keep track of the offset from object address to which linked list pointer you're currently dealing with. That's often a compile time constant, but makes the API more awkward. In this case it seems to be the string "node", which seems nice enough. C libraries often use the preprocessor to do something similar.

    • mastax a day ago

          new Node<TStruct>[16];
      
          new TStructContainingNode[16];
      
      What’s the difference?
      • ulbu a day ago

        there’s 16 contiguously stored pointers to 16 non-contiguously stored tstructs (they may be contiguously stored, but you can’t make this assumption from this type). there’s 16 contiguously stored tstructcontainingnode’s.

  • ajross a day ago

    > I don’t get this. In the generic version the data is embedded with the linked list node, so there’s only one allocation per node.

    The reason is that the object being placed "in" the list can have a longer lifetime than the list node, in fact that's generally the case. Like, I work on Zephyr and in Zephyr we have "threads", whose tracking structs are managed by the application code. But they can be in kernel-maintained lists like a scheduler queue, which is an ephemeral state that the essentially-perpetual thread will long outlive. There's no place to allocate a "generic" node at list insert time.

    (Also, more importantly, Zephyr is an RTOS and disallows heap use in the core parts of the system to permit use in environments that need to disallow dynamic allocation. But that's not really relevant to a generics discussion.)

    But you can trivially put a scheduler queue node right into the thread object, knowing that it's a behavior of the type. The semantics become different though: you only get the one node, it's not possible to have multiple "lists of threads" using this technique unless you know exactly what each list is for ahead of time.

steventhedev a day ago

This feels like a net negative result. It removes some of the complexity of using generics, but it couples between the data type and the collections it can be indexed by.

What are the benefits of this approach? Is it limited to data alignment, or is it out of a greater desire to remove generics?

  • whizzter a day ago

    Intrusive linked lists as this is called removes unnecessary allocations and traversal (main reason why linked lists have such a horrible reputation), say a hypothetical example where you have a temporary listener object installed that listens to broadcasts on channel X and times out after 5 minutes.

    Upon connecting the object is connected to the channel X linked list as well as some other list of objects that are killed at the timestamp 5 minutes in the future.

    With an intrusive linked list the link-node resides within the object, the only thing needed when installing is linking it into the 2 lists (this is a few move-operations), an external linked list would _require 2 extra allocations_ for the linked-list nodes.

    Broadcasting from X is almost the same since it's a simple traversal, albeit with double the cache pressure since the object and the linked-list node probably lives separately.

    The real HORROR comes when disconnecting, if channel X has millions of listeners it could become a real bottleneck to traverse the linked list to find the link-node that connects the actual object since there could be tons of jumping around memory. An intrusive linked list would just disconnect itself if it's doubly-linked.

    This is why hashsets/tables,vectors/arraylists,etc are often used in practice (because many "modern" OO languages never added good support for the machinery needed here) making linked lists look quite bad (there is other badness but using non-intrusive linked lists is almost always worse than using something else than a linked list altogether).

    • steventhedev a day ago

      The generic version in TFA puts the data type allocated alongside the next pointer - no additional allocation needed. The only functional difference is if the zig compiler is not sufficiently advanced to understand it can reorder the fields (hence the alignment question).

      The removal scenario is merely specifying that you are passing in ConnectionListNode instead of a Connection. Although maybe it's a good idea to think about how they compose comparatively.

  • flohofwoe a day ago

    It's not at all unusual for intrusive linked lists though?

    On AmigaOS (which is entirely built out of intrusive doubly linked list) the list node is placed at the start of a struct, so the pointer to the node is also the pointer to the linked item. There's also no 'coupling' because the list manipulation functions take pointers to the list node structs, but they don't need to know the 'outer' item struct.

    Zig's @fieldParentPtr is more flexible since the node struct can be anywhere in the parent struct and you still can get from the node pointer to the item base pointer (and more importantly, it makes it trivial to link the same item into multiple linked lists).

    • Joker_vD a day ago

      It's not unusual at all, it's just... should it be exposed? I personally prefer having "Node struct with next pointer + inlined generic data" design when it can be encapsulated away since, well, it can be encapsulated away, and the result data layout is the same. And when you expose this design, well, you end with all sorts of disasters, including OOP inheritance (no, really: [0]).

      [0] https://catern.com/inheritance.html

      • flohofwoe a day ago

        If the node embeds the data item you can't link the item into multiple linked lists, but if the data item embeds the node(s) suddenly that's trivial.

        I don't know if this was the motiviation for the design change though, but IMHO intrusive linked list are the 'proper' linked list design, and C++-style "nodes with payload" are a weird outlier.

        Another reason might be that generic code leads to a lot of code duplication in the binary (which may or may not be optimized away by the compiler though - but at the cost of increased build times though).

        • lelanthran a day ago

          > If the node embeds the data item you can't link the item into multiple linked lists, but if the data item embeds the node(s) suddenly that's trivial.

          Perhaps I'm missing something, but isn't it the other way around? In C, this is what I'd expect for the intrusive and extrusive:

              // Intrusive
              struct user_data_t {
                 struct user_data_t *next;
                 struct user_data_t *prev;
                 // The object's fields
              };
          
              // Extrusive
              struct node_t {
                 struct node_t *next;
                 struct node_t *prev;
                 struct user_data_t *payload;
              };
          
          The intrusive one obviously can't be shared between lists.
          • flohofwoe a day ago

            No, intrusive would be (note: no pointers in user_data_t):

                typedef struct { node_t* next } node_t;
            
                typedef struct { ... } payload_t;
            
                typedef struct {
                    node_t node;
                    payload_t payload;
                } user_data_t;
            
            ...and if you want user_data_t to be included into multiple lists:

                typedef struct {
                    node_t list1_node;
                    node_t list2_node;
                    node_t list3_node;
                    payload_t payload;
                } user_data_t;
            
            ...of course in C now it gets tricky to get to the start of user_data_t given a pointer to `list3_node`, but that's where @fieldParentPtr comes in.

            The advantage versus your extrusive example is that the payload doesn't need to be referenced through a pointer, which drastically simplifies lifetime tracking / memory management.

            • lelanthran a day ago

              > No, intrusive would be (note: no pointers in user_data_t):

                  typedef struct { node_t* next } node_t;
              
                  typedef struct { ... } payload_t;
              
                  typedef struct {
                      node_t node;
                      payload_t payload;
                  } user_data_t;
              
              
              That's exactly the same as my intrusive structure, no?

              > typedef struct { node_t list1_node; node_t list2_node; node_t list3_node; payload_t payload; } user_data_t;

              In C, at any rate, that doesn't give you "inclusion into multiple lists". It gives you "inclusion into at most 3 lists. The extrusive example I posted gives "inclusion into multiple lists".

              So, yeah, I'm still not seeing your point.

              • Zambyte 19 hours ago

                > That's exactly the same as my intrusive structure, no?

                It's not, for the reason they put in parentheses:

                > note: no pointers in user_data_t

                An allocation of user_data_t also allocates space for payload_t, rather than just allocating space for a pointer to payload_t. Your structure requires an additional allocation to point the payload_t* at something. The fact that they hid the next node_t* in a struct doesn't matter though.

                • lelanthran 16 hours ago

                  > It's not, for the reason they put in parentheses:

                  >> note: no pointers in user_data_t

                  I feel like I am taking crazy pills: where, in my intrusive example, are pointers in the `user_data_t`?

                  My intrusive code sample had no pointers for the user_data_t. It's exactly the same as GP's intrusive example.

                  • Zambyte 10 hours ago

                    It is not you that is crazy, I have been exhausted. My bad.

  • reissbaker a day ago

    Zig has no desire to remove comptime AFAIK (comptime being the larger Zig feature by which people implement generics in the language) — comptime is pretty much the reason to use Zig.

    The benefits of intrusive linked lists are higher performance; you can use comptime to still have decoupled code.

    • surajrmal a day ago

      The generic approach should be the same performance. This approach just lets you place your data in multiple lists without needing multiple allocations.

      • reissbaker 21 hours ago

        No, the generic approach requires your data to be spaced out further in memory (or to be heap allocated), which causes CPU cache misses and is slower. The entire reason for intrusive linked lists is performance. Standard linked lists are notoriously slow compared to almost any other similar data structure, which is why they are hardly ever used in real code (ArrayLists and vectors are much more common).

  • GolDDranks a day ago

    I think it matches really well for the Zig ethos of "artisanal, minimal code". More than often, a type is strongly tied to how it is being used in the code base. In that sense, having it be tightly coupled to a data structure isn't much of a problem. The opposite isn't true, and the data structure is independent of the embedding parent data. The implementation of that data structure itself can still be presented as a library, and the "generic" parts carefully tested.

  • Someone a day ago

    The only logical explanation I can see is that these are two decisions:

    - linked lists aren’t useful on modern systems because traversing them causes to many cache misses. Therefore, we shouldn’t provide such a simple implementation.

    - but we need one in low level OS code, and there, intrusive lists are preferred. Their limitation that you cannot store an object in multiple lists isn’t a problem, and the extra speed and the fact that you can move items between lists without allocations is desired.

    I don’t see why the intrusive implementation has to be so basic, though. Can’t you, in Zig, express that a generic type T used in a generic list class has to have a nullable next field that points to a T?

  • yccs27 a day ago

    You could create a comptime function that adds the Node field to any type. I guess then you've arrived back at the previous generic version.

lightingthedark a day ago

Can someone explain how the claim of higher performance works here? In C, which lacks generics, an intrusive list is preferred because otherwise you end up with each node having a void pointer to the data it holds. The previous Zig version was a generic, so the data type could easily be a value type. Given that the compiler is allowed to rearrange the layout of both structs unless data is packed or extern (in which case it almost certainly won't want to have a linked list node in the middle anyway) isn't the resulting type exactly the same in memory unless you intentionally made T a pointer type?

  • ashdnazg a day ago

    I don't understand the higher performance either. What I know as the significant advantage is that you can have one object in multiple lists.

    • messe a day ago

      > What I know as the significant advantage is that you can have one object in multiple lists.

      Another advantage is smaller code size, as the compiler doesn't need to generate code for SinglyLinkedList(i32), SinglyLinkedList(User), and SinglyLinkedList(PointCloud). This could have a performance impact by making it more likely that code remains in the cache.

    • lightingthedark a day ago

      Technically that should be possible the other way by using a Node<T> so the type of the second list ends up being a Node<Node<T>> but I can see why an intrusive list would be preferred to that, and also the linked list API might prevent that pattern.

      Usually if I have multiple lists holding something I have one that's the 'owner' and then the secondary data structures would have a non-owning pointer to it. Is that the case where the performance would be better with an intrusive list? My intuition would be that having multiple Node members would pollute the cache and not actually be a performance win but maybe it is still better off because it's all colocated? Seems like the kind of thing I'd have to benchmark to know since it would depend on the number of lists and the size of the actual data.

      • lightingthedark a day ago

        Okay so in the multiple list case performance would actually be worse because you'd have a double pointer dereference. I was thinking you'd have the list nodes contiguous in memory so the first dereference would always hit cache but that's a bad assumption for a linked list.

        Since you shouldn't reach for a linked list as a default data structure modern hardware anyway, I actually do see how this change makes sense for Zig. Neat!

  • grayhatter a day ago

    > Can someone explain how the claim of higher performance works here? In C, which lacks generics, an intrus

    I can only give a handwavey answer because I've yet to see any data, and if an engineer tells you something is better but doesn't provide any data, they're not a very good engineer. So grain of salt and all that. But the answer I got was because cache performance. Writing code this way your CPU will spend less time waiting for main memory, and the branch predictor will have better luck. The argument makes sense, but like I said,I've yet to see real world data.

    > isn't the resulting type exactly the same in memory unless you intentionally made T a pointer type?

    Yes and no. If I understand what you mean, the bit layout will be the 'same'. But I think your confusion is more about how what a compiler means by pointer type, and what a human means. If you pull away enough layers of abstraction, the compiler doesn't see *Type it'll only see *anyopaque, phrased completely differently; according to the compiler, all pointers are the same and are exactly memory_address_size() big. *Type doesn't really exist.

    Writing it this way, imagine using just the LinkedList type, without a container of any kind. node to node to node, without any data. While it would be pointless, that would (might) be faster to walk that list, right? There's no extra data loads for the whole struct? That's what this is. Using it this way it gets complicated, but translating theory to asm is always messy. Even more messy when you try to account for speculative execution.

    • Zambyte 19 hours ago

      > But I think your confusion is more about how what a compiler means by pointer type, and what a human means. If you pull away enough layers of abstraction, the compiler doesn't see *Type it'll only see *anyopaque, phrased completely differently; according to the compiler, all pointers are the same and are exactly memory_address_size() big. *Type doesn't really exist.

      Check out the implementation of SinglyLinkedList in the latest release (before the change in the post)[0]. You'll notice the argument for SinglyLinkedList is (comptime T: type), which is used in the internal Node struct for the data field, which is of type T. Notably, the data field is not a *T.

      In Zig, when you call the function SinglyLinkedList with the argument i32 (like SinglyLinkedList(i32)) to return a type for a list of integers, the i32 is used in the place of T, and a Node struct that is unique for i32 is defined and used internally in the returned type. Similarly, if you had a struct like Person with fields like name and age, and you created a list of Persons like SinglyLinkedList(Person), a new Node type would be internally defined for the new struct type returned for Person lists. This internal Node struct would instead use Person in place of T. The memory for the Node type used internally in SinglyLinkedList(Person) actually embeds the memory for the Person type, rather than just containing a pointer to a Person.

      These types are very much known to the compiler, as the layout of a Node for a SinglyLinkedList(i32) is not the same as the layout of a Node for a SinglyLinkedList(Person), because the argument T is not used as a pointer. Unless, as the gp mentioned, T is explicitly made to be a pointer (like SinglyLinkedList(*Person)).

      [0] https://ziglang.org/documentation/0.14.0/std/#src/std/linked...

  • anarazel a day ago

    Intrusive lists are often used to enqueued pre-existing structures onto lists. And often the same object can be in different lists at different times.

    That's not realistically dealt with by the compiler re-organizing the struct layout.

ralferoo a day ago

I don't use Zig, but one advantage of having a genericised intrusive linked list where the next pointer doesn't have to be the first thing in the structure is when you want to use larger types, such as 128-bit fields. Sticking a pointer at the beginning would mean the compiler would have to insert alignment padding after the pointer or break the default alignment.

  • jwmerrill a day ago

    The next pointer doesn’t have to go first in the structure here. It can go anywhere, and you can use @fieldParentPtr to go back from a reference to the embedded node to a reference to the structure.

    • n42 19 hours ago

      I think that's the point the parent was making. I read it as them describing a benefit of Zig's approach.

      • ralferoo 10 hours ago

        Yeah, that's what I was trying to say, but obviously not clearly enough.

        • jwmerrill 6 hours ago

          My mistake! It seems clear in hindsight…

spiffyk a day ago

For better or worse, the Zig community has long been embracing the `@fieldParentPtr` builtin as the primary way to do things like OOP-ish inheritance, so this change feels pretty in-character for the language to me.

  • Zambyte 18 hours ago

    Really? I did a quick grep in ghostty (160k lines of Zig) and found four total references to @fieldParentPtr. I'm still a Zig baby, but I have not seen very many references to @fieldParentPtr at all.

    • spiffyk 11 hours ago

      I would expect the inheritance pattern to be quite rare in performance-aware software design overall, since such constructs tend to be rather cache-unfriendly. So I would attribute the low number of references to inheritance being rare, rather than @fieldParentPtr not being used for inheritance.

MawKKe a day ago

Isn't the new one also horribly type-unsafe? Since you can put (parent) objects of any kind into the list without any mechanism to detect which is which?

kajika91 a day ago

Couldn't we have a function to return the data like this?

  pub const SinglyLinkedList(comptime T: type) type {
    return struct {
      first: ?*Node = null,

      pub const Node = struct {
        next: ?*Node = null,
      };

      const Self = @This(); 
      pub fn data(self: Self) *T {
        return @fieldParentPtr("node", self);
      }
  };
  • nh1996 a day ago

    This would require the SinglyLinkedList to be generic and would require that the data struct use "node" as the field name. Also, as some comments have pointed out, this type of linked list can be useful when a data struct needs to be in multiple lists, in which case there is no single "node" field.

      const some_data = struct {
        // Some data fields
        // ...
        bar_node: SinglyLinkedList.Node,
        baz_node: SinglyLinkedList.Node,
      };
    • throwawaymaths a day ago

      sure, just make T == void fall back to the existing implementation.

alexvitkov a day ago

Please just use a pointer. Use metaprogramming when it makes sense, not to obfuscate something that was figured out 60 years ago.

roetlich a day ago

This looks somewhat horrifying. How do I safely write a function that takes a linked list as a parameter?

  • the_mitsuhiko a day ago

    > How do I safely write a function that takes a linked list as a parameter?

    Zig does not have a lot of generic code. You would pass the user directly and then walk the list or you use comptime. The real answer is that "you don't write code like that in Zig".

  • AndyKelley a day ago

    Two points here:

    Linked lists are useful in unsafe code. Most recent use case I had for them was in an event loop with coroutines. It's not possible to implement such thing in memory safe languages. For example if you use Rust, you have to use unsafe [1].

    @fieldParentPtr does not yet have safety but it is a planned upcoming change to the language, with a fairly straightforward implementation [2].

    [1]: https://github.com/search?q=repo%3Atokio-rs%2Ftokio%20unsafe...

    [2]: https://github.com/ziglang/zig/issues/2414

    • roetlich a day ago

      Oh, that makes more sense. Thank you!

      Is this linked list type then mostly meant to be used as an internal implementation detail, and not be exposed in a public API? Because if I see a function that takes a non-generic list I won't really know how to call it. :)

      • AndyKelley a day ago

        Linked lists are sometimes the right tool for the job. They are available in the standard library for such cases. If you are struggling to understand how you would use this API then it probably means you have not yet been exposed to such a use case, and you are being quite intentionally steered away from using them.

  • reissbaker a day ago

    Assuming you mean "how would I safely write a function that takes a generic linked list and does something with the data," I'm pretty sure you would use comptime: your function would take the concrete type (say, User) as a comptime parameter, and then you would do your list stuff via accessing .node and use the fieldParentPtr to access the underlying data.

    Syntactically I don't think it's that weird, TBH. And it's typesafe: if you write invalid code the compiler will give you an error.

    • boomlinde a day ago

      @fieldParentPtr is not type safe. It assumes that the given pointer is to a field of the given name in the inferred type. Zig can detect at compile time whether the inferred type has a field of that name of the same type as the pointer child. So far, so good: type safe.

      The lack of safety stems from the fact it doesn't know that the assumption that the pointer has that parent is true. Here's a very simple illustration. It'll compile.

          const T = struct {
              field: i32 = 0,
          };
          
          pub fn main() void {
              var nonfield: i32 = 0;
          
              _ = @as(*T, @fieldParentPtr("field", &nonfield));
          }
      
      `nonfield` doesn't have a parent T. The use of @fieldParentPtr here causes what the official reference calls "unchecked illegal behavior"; it isn't checked even in the safe build modes.
      • grayhatter a day ago

        > The use of @fieldParentPtr here [...]

        the *hypothetical* use here...

        This simple example as written is actually correct, and not actually unchecked illegal behavior. But this is just a bit of pedantry, because it would be unchecked illegal if you use this in a more complicated example.

        I mention it because it's important to note, it can be done correctly, as you obviously just did it correctly, by accident. That said, I still agree, there're no guardrails on this pattern.

        • boomlinde a day ago

          > the hypothetical use here...

          No, it's not hypothetical. I just did it, which demonstrates that it's not type safe and that the compiler won't give you an error if you write invalid code, which is what the post I responded to claimed.

          > This simple example as written is actually correct

          No. The official reference is clear about this: "If field_ptr does not point to the field_name field of an instance of the result type, and the result type has ill-defined layout, invokes unchecked Illegal Behavior."

          Because there is no well-defined layout for plain structs in Zig, they satisfy the "ill-defined layout" requirement. Merely invoking the @fieldParentPtr on a pointer that doesn't satisfy the other requirement in that case invokes unchecked illegal behavior according to the language reference.

          > I mention it because it's important to note, it can be done correctly,

          Ignoring that you can't seem to get the facts straight, the claim I am responding to is that it's "type safe" and that "if you write invalid code the compiler will give you an error". This is not the case, regardless of whether my example invokes unchecked illegal behavior (which it does) or not.

          I'm very clearly not saying that it can't be done correctly and I don't think you can argue for that in good faith.

          • grayhatter a day ago

            > No, it's not hypothetical. I just did it, which demonstrates that it's not type safe and that the compiler won't give you an error if you write invalid code, which is what the post I responded to claimed.

            Describe explicitly how your example will trigger illegal behavior? Happy to discuss the behavior of the compiled asm if you'd like to be that pendantic?

            Not type safe, and contains an appreciable risk or defect are two distinct concerns. One matters less than the other.

            > Because there is no well-defined layout for plain structs in Zig, they satisfy the "ill-defined layout" requirement.

            No? Struts have a defined layout, they're not guaranteed between compilations, but they don't change within a compilation unit.

            > Merely invoking the @fieldParentPtr on a pointer that doesn't satisfy the other requirement in that case invokes unchecked illegal behavior according to the language reference.

            And the failure mode is the compiler will decide to delete your OS?

            Unchecked illegal behavior isn't magic, doing something similar to illegal behavior doesn't mean the code will or won't do what you intended. uint rollover is unchecked illegal behavior (when not explict, in release fast) but just because the uint might roll over doesn't make that code invalid.

            > I'm very clearly not saying that it can't be done correctly and I don't think you can argue for that in good faith.

            I thought you were saying that it's guaranteed to be incorrect. In fact, I think that's what you said in your reply to me. Mostly because you used the word 'mearly'.

            The way that I understood what you meant was, the example you provided was guaranteed wrong. In fact it's actually correct. It will have the same semantics and behavior as what I assume was desired.

            > Ignoring that you can't seem to get the facts straight [...] and I don't think you can argue for that in good faith.

            I'm sorry if I was unclear, I was trying to discuss it in good faith. :(

            • boomlinde 10 hours ago

              > Describe explicitly how your example will trigger illegal behavior? Happy to discuss the behavior of the compiled asm if you'd like to be that pendantic?

              Through calling @fieldParentPtr with a field_ptr not pointing at a field of the given field name in the result type, when the result type has an ill-defined memory layout. I'm essentially just paraphrasing the documentation, which I've already quoted.

              Generated code is irrelevant to this end. Illegal behavior can coincidentally result in code that works as intended in practice as a side effect of the implementation. Similarly you may expect a signed integer to wrap around as it overflows in C because of the implementation of the compiler and the code it generates, but it's still undefined behavior.

              > No? Struts have a defined layout, they're not guaranteed between compilations, but they don't change within a compilation unit.

              This is not the sense in which the Zig language reference uses the term well-defined memory layout. Bare structs, error unions, slices and optionals don't have a well-defined memory layout in the sense the Zig language reference uses it.

              > And the failure mode is the compiler will decide to delete your OS?

              The failure mode is irrelevant. Unchecked illegal behavior is unchecked illegal behavior regardless of the failure mode. You are moving goalposts now.

              > I thought you were saying that it's guaranteed to be incorrect. In fact, I think that's what you said in your reply to me. Mostly because you used the word 'mearly'.

              The example I posted is guaranteed to be incorrect, which demonstrates that the use of @fieldParentPtr can result in unchecked illegal behavior, hence not type safe nor giving you an error when you write invalid code. Regarding the use of "merely", you should adopt the habit of reading complete sentences before you draw conclusions about what they imply.

              > The way that I understood what you meant was, the example you provided was guaranteed wrong.

              The example is guaranteed to cause unchecked illegal behavior.

              > I'm sorry if I was unclear, I was trying to discuss it in good faith. :(

              If you want to argue in good faith, you can start with a good faith reading of the language reference.

    • roetlich a day ago

      Okay, thanks for explaining more.

      > And it's typesafe: if you write invalid code the compiler will give you an error.

      One more question, if @fieldParentPtr("node", n) is typesafe and can get you a User, the compiler needs to know that the struct has the fields of a User, and that this pointer is indeed pointing at a node field, right? Then why do you need to specify "node" at all?

      I think I don't understand Zig at all :)

      • messe a day ago

        What if the User is in multiple linked lists?

            const User = struct {
                username: []const u8,
                id: u128,
                age: u7,
                sorted_by_age: std.SinglyLinkedList.Node,
                sorted_by_id: std.SinglyLinkedList.Node,
            };
    • fpoling a day ago

      But how does the generic code know the name by which the field is known in the struct that contains the field? Is it supposed to be passed as an extra parameter to the generic function?

      • spiffyk a day ago

        I think the more important question here is: What operations would you possibly want to do with a linked list that are generic and at the same time do something with the concrete data in it?

        To me these complaints sound like hypotheticals with no sound grounding in the real world. If there indeed is data that is common to the various types you would want to operate upon in such linked lists, you would nest. E.g. you would have some common Entity struct containing the common data and the LinkedList.Node; this Entity would then be inside your more concrete structs. The generic function would then take a pointer to Entity.

        • reissbaker a day ago

          I think a pretty good example would be: how do you write a map function? Or filter, find, etc.

          (You would do it with more comptime, but the question is legitimate!)

          • spiffyk a day ago

            > how do you write a map function? Or filter, find, etc.

            You just don't. Zig does not have lambdas anyway, there's no readability incentive to having such functions there. You do these things with plain old loops built into the language.

            • reissbaker a day ago

              Zig has first-class functions (you can pass functions as parameters); it just doesn't have closures. Map pretty rarely uses closures anyway IME; e.g. converting a list to JSON doesn't need to close over outside variables. And anyway, anything that's:

              1. Generic over lists, and

              2. Takes a function as a parameter

              Will want to know what the node field name is. Luckily, comptime provides a solution there.

              TBH I think "you just don't" is a pretty unsatisfying answer to "how do I use these features that are built into Zig" — especially when you can use them, very easily.

              • spiffyk a day ago

                I mean yes, you can pass functions as parameters in Zig, you can even define them inline if you wrap them in anonymous structs, but my point is that — in a language where this is supported somewhat properly — you would do this to improve readability with some sort of a fluent API. If you attempt to do it in status-quo Zig, readability will only suffer and you are still much better off doing it with for and/or while loops (depending on the data structure), especially when Zig has nice syntax sugar with payloads in both to support these.

                edit: formatting

            • Zambyte a day ago

              Zig definitely encourages an imperative style by design, but it does support function pointers, which can be used to implement a map function. I do think passing the field name of the node as a comptime []const u8 as an extra argument to map, filter, etc. would be the only way to implement a generic version of these functions.

              The fact that these functions are already designed to be cumbersome to implement, I think that's fine? I have also yet to use a linked list in Zig anyways. It's probably better to use an array, slice, std.BoundedArray, std.ArrayList, std.SegmentedList, or std.MultiArrayList unless there is a specific reason that a linked list is the best option.

      • reissbaker a day ago

        Yup, you could pass it in at comptime as well! And access via @field.

  • GolDDranks a day ago

    If you write a function that takes a _generic_ linked list as a parameter, you'd have a function that refers to just the linked list subrecord, and does only linked list operations to that and the linked nodes.

    If you want to write a function that takes a linked list of a specific type as a parameter, you just take in a value of that type. The linked list is baked in, so you can get to the other nodes, and because the type is known, you can get back to the parent type from the linked nodes with fieldParentPtr. How to do that _safely_? I don't think that Zig embraces any Rust-like safe/unsafe dichotomy, so you don't.

  • flohofwoe a day ago

    Since you don't need to care about the 'outer type' in that case you just pass a pointer to the linked list header or a linked list node and that's it.

    If you need to access the outer type, just pass a pointer to that type (since your functions need to know the outer type anyway I don't think there's a need to reach for generics).

    • skywal_l a day ago

      You have to pass the field name too ("node").

      • flohofwoe a day ago

        Not really. If you only want to manipulate the list, you only need a direct pointer to the nested node but don't need to know the 'outer type'.

        Only if the function takes a pointer to the outer type it needs to know how to get the pointer to the embedded node struct from the item pointer.

        ...I guess it makes a lot more sense if you ever wrote code for AmigaOS ;)

  • kllrnohj a day ago

    You don't & generally shouldn't be in the first place, in any language. Linked lists are a very niche data structure, so generic code should ~never be using them. So it's a moot question. It's kinda like the complaints about how hard a doubly linked list is in Rust - it's just not important because it's not something you should be using 99.999% of the time anyway.

    • grayhatter a day ago

      A linked list is just a simpler version of both a graph and a tree. What data structure would you suggest to represent both/either of those?

      Or are you asserting, because you've never used them they're not common? Because while maybe I agree, and I don't often reach for a linked list, I've built plenty of trees and graphs.

      • kllrnohj a day ago

        Trees and graphs serve useful roles, of course, but calling a linked list just a simpler version is a stretch. It'd be like calling an array a simplified B-tree. The simplification changes not just the implementation but also radically changes the applicability of the result.

        • grayhatter a day ago

          except a tree or graph with multiple pointers is similar to linked list with one pointer.

          where as a b-tree stored in an array without pointers probably shouldn't be called a b-tree.

          or am I missing something?

    • roetlich a day ago

      Well I personally almost never use linked lists, don't like them. But Zig seems to think otherwise, they have two implementations in the standard library.

      I don't know much Zig, I just wanted to ask how to use the type that the article talks about?

      Let's use the very simple example from the article. Let's say I want to extract this code into a function:

          while (node) |n| {
            const user: *User = @fieldParentPtr("node", n);
            std.debug.print("{any}\n", .{user});
            node = n.next;
          }
      
      1. How does that look, what's the type signature of this function? 2. What happens if I put in a list doesn't contain users? Do I get a simple to understand compile time error, or can I segfault because I'm accessing bad memory?

      And I don't think that would need any generics, since the list type isn't generic, right?

      • boomlinde a day ago

        1. If you mean @fieldParentPtr, the first argument is a compile time-known name of a field, the second argument is a pointer to a field of that name in the parent type, which is inferred from the left hand side of the assignment

        2. Then you're screwed. In Zig, using @fieldParentPtr on a field that doesn't have the given parent is unchecked illegal behavior, meaning that there are no checks that can catch it at compile time. This change basically guarantees that there will be a pretty serious foot gun every time you iterate over the items of a standard library list.

        • roetlich a day ago

          Thank you, that answers all my questions.

esjeon a day ago

Hmph, is there any rationale against a generic wrapper over it? Exposing the internal details can be irky in higher level application development. It would not hurt to have some inline functions.

DeathArrow a day ago

I wonder why should a language implement in its libraries a SingleLinkedList and a DoubleLinkedList.

I do get why you need an array-like list, a dictionary/hashmap, a stack and a queue. I got the impression that linked lists aren't used very often. Or maybe it's a different case with Zig?

  • messe a day ago

    Maybe because it's used elsewhere in the standard library?

    SinglyLinkedList is used in the standard library in std.Thread.Pool, and std.heap.ArenaAllocator. DoublyLinkedList is used in the http client's connection pool, as well as the ObjectCache for the package manager's git client.

  • boomlinde a day ago

    > I got the impression that linked lists aren't used very often.

    I did some cursory research on the software I've been using today and presented the results here: https://news.ycombinator.com/item?id=43684121

    I think there is a misconception around linked lists mostly stemming from the fact that they're not great for cache locality. Someone presents the idea that linked lists under common circumstances are unintuitively slow and therefore should be considered carefully for performance sensitive applications. A game of Chinese whispers immediately forms around that idea and what comes out the other end is a new idea: that linked lists are bad, shouldn't be used and aren't used. Meanwhile, they continue to be used extensively in system software.

    • bsder a day ago

      Linked lists get a bad reputation because they wound up being a default data structure--and that's almost always a mistake on modern computers.

      Back in days of yore, small memory and CPU meant that linked lists and arrays were really the only effective data structures. Allocating and computation were super expensive, so following a link or having an index is really the only thing that works.

      On today's computers, CPU is so much faster than memory that it is almost always better to recompute a value than try to look it up. Locality is so important that linear actions trump asymptotic complexity until you get quite a lot more elements than you might expect. Memory is so cheap that it is worth exponentially overallocating in order to amortize complexity. etc.

      At the end of the day with modern programming, you should almost always be reaching for something simpler than a linked list (array/vector) or you should be reaching for an abstraction that is more complex than a linked list (hash/dict, queue/deque, etc.).

      With modern CPUs, if you are using a linked list for anything other than building a more complex data structure, you're probably making an incorrect choice on almost all fronts.

      • Validark 19 hours ago

        > Locality is so important that linear actions trump asymptotic complexity until you get quite a lot more elements than you might expect.

        The reverse of this can be true. In the following paper, they found that William's Heap Construction (one by one, O(n log n) time) beat Floyd's Heap Construction (O(n) time) when the input is larger than the size of off-chip cache.

        https://doi.org/10.1145/351827.384257

  • bsder a day ago

    What do you think is the data structure underneath a hash, stack or queue? It's likely a linked list.

    So, you are completely correct. The issue is simply that you are at a different level of abstraction.

RcouF1uZ4gsC a day ago

Awesome! In my opinion, an intrusive linked-list is almost always what you really want when you reach for a linked list. An external linked list is often worse than a vector.

hoseja a day ago

Linked lists should be obscure niche data structures for when you absolutely need their unique characteristics, not some front-and-center default.

  • netbsdusers a day ago

    They're not obscure or niche, they are everywhere in operating systems. The linked list is probably the single most essential and the most used data structure in any kernel.

    • simonask a day ago

      Implementing an operating system is incredibly niche and obscure.

      The practices that apply to kernel development do not generally apply to very much else.

    • fcantournet a day ago

      Ah yes, operating system kernels, famously super mainstream programs that almost everyone works on.

  • grayhatter a day ago

    This is true for the kind of programmin most people do. Linked Lists make no sense in JS, Python, Rust, etc. But they are without a doubt still an incredibly useful pattern with unmatched performance characteristics. Especially if you're trying to be memory efficient. It's a systems programming pattern, rarely useful if you're just a web developer.

    But then again... I never thought I'd need to know music waveform theory while I was watching the magic school bus as a kid... Turns out that was incredibly useful when I needed to implement a music note generator for a VoIP client... You never know when something might actually be really useful.

    • simonask a day ago

      This comment makes no sense.

      1. You have linked lists in Rust. They're in the standard library.

      2. Linked lists do not have "unmatched performance" - they are slower than almost anything, outside of highly specialized use cases. Even moreso for intrusive linked lists. Worst of all, they are horrible for cache locality.

      There are valid uses of linked lists, that's why they exist, but most people don't have those use cases.

      • grayhatter a day ago

        Speed isn't the only performance metric that matters. I probably should have said explicitly that I'm referring to uses where it makes sense to use a linked list. But unfortunately I didn't specify because that wasn't the point I really wanted to make.

        There's a lot of things to learn that appear to to not makes sense... until suddenly they do.

        > There are valid uses of linked lists, that's why they exist, but most people don't have those use cases.

        Isn't that pretty much what I said?

        I'm pretty sure we agree that they're useful, but their usefulness isn't as universal nor as common as the pattern is.

        • simonask a day ago

          Linked lists are worse than flat arrays in every single performance metric other than ordered removal. They use more memory (the links), have inferior cache locality, cause more allocations (unless you're allocating anyway, in the case of intrusive lists).

          Their one benefit - O(1) ordered removal - is easily achieved with a flat array, in at least two common ways: swap in the last element if you don't care about the order, or mark the index as vacant if you do care about the order, or if you care about the stability of references.

          I mean, I agree that they're kind of neat from a certain perspective, but they're almost never the right choice.

  • grandempire a day ago

    I’m all about data-oriented design, but I don’t think this is true - you need their unique characteristics in almost every project.

    • simonask a day ago

      In all of my years, I have seen maybe 2 projects that had one valid use case each. They exist, sure. It's not that common.

      • boomlinde a day ago

        Out of curiosity I looked up some of the software I've meaningfully interacted with today. Of all I looked up—the operating system kernel, the init system, the shell, the terminal emulator, the browser, the compilers, the text editor, the windowing system, the window manager, the notification daemon, the audio server, the audio session manager, the GPG agent, the NTP daemon, the SSH daemon, the DHCP client, the wireless network manager, the power manager, the policy manager, D-Bus, the video player, the video editor—each uses linked lists. There's some more system software showing up in ps (which by the way uses linked lists) that I haven't considered but I am rather confident that most of it uses linked lists.

        Maybe you only see these projects as a user, but linked lists are not uncommon. Your experience reflects your, well, experience. We all sit in different niches.

        • simonask a day ago

          I'm wondering what makes you feel confident about the use of linked lists in all of those components.

          Mind you, most of those will be written in C on a typical Linux installation, and linked lists happen to be one of the two collection types that are relatively easy to use in C (the other being a flat array), so I will concede that some software is using linked lists out of desperation, rather than it being the correct choice. :-)

          • boomlinde 10 hours ago

            > I'm wondering what makes you feel confident about the use of linked lists in all of those components.

            Of all of those mentioned I literally looked up their source repositories up and searched for obvious indicators like "linked", "list", ".next" or "->next" and then verified that I was indeed looking at one or more linked lists. Where does your confidence come from? Oh right, you already mentioned it: it's based on your experience of the projects you've worked on.

            The rest of your reply is just moving goalposts, mind reading and a glaring category error. Get back if and when you have something useful to add.

            • simonask 22 minutes ago

              That’s an incredible amount of effort to throw at an argument on Hacker News.

bmandale a day ago

A few notes I would make as an outsider:

- conventionally we put the `node' struct at the start of the `user' struct. The advantage is that we can convert from node* to user* with a simple cast instead of messing around with offsets. Knowing the offset stuff is still potentially useful in case you want multiple `node's in a single struct, but that is seldom the case so you can almost always get by with just a cast.

- this strategy is, especially with the node at the start of the user, equivalent to inheritance. I take it from the fine article that zig doesn't have inheritance? If they are planning on adding more interfaces like this linked list one, then inheritance would be a good idea. We often treat inheritance using analogies to the animal kingdom or to geometric shapes, but data structures like linked lists are (to my understanding) the problem that inheritance is actually designed to solve, and it is really the best approach to solving problems like in the OP.

- just lol on zig speed running being a crappier C++.