In a language like Rust, the compiler will “lock” the pointers for you, and you can’t forget.
In a language like C++ (and presumably Zig), one could, in theory at least, have the iterators and slices that reference the storage of a dynamic array hold some sort of lock that pins the storage.
But this API requires the programmer to remember to lock the pointers and also requires the programmer to keep the lock alive for the correct region of code. And it looks to me like even the example in the blog post has the lock taken completely outside the function that requires stability, so there is nothing whatsoever that gets the lock scoping right. Even the type system can’t help — the offending parse function can’t declare that it wants a pointer-locked ArrayList parameter.
“I use it in a lot of places where I know the max capacity ahead of time -- ensureCapacity() followed by a lot of AssumeCapacity()-styled commands. It's convenient for all of the ... convenience ... methods (append() requires some bookkeeping somewhere, appendSlice() requires more, and so on). In those usages, it's basically syntactic sugar over a slice”*
I suspect “where I know the max capacity ahead of time” covers most if not all use cases (if it you use this without knowing max capacity, you either accept your code may panic, or you do some unlock, grow, lock again dance when you discover your initial estimate is wrong)
If so, wouldn’t adding a growable container where you specify capacity at construction time and removing access to the internal pointers of ArrayList be a better way to handle this?
Do any languages have a notion of "relative pointers"? So in the example if instead of appending "line" as ptr & len, it'd instead be appending an offset & len which could in theory be used to safely compute the actual location even with relocations.
Languages with dependent types can express things like “this offset is in bounds relative to this other array”, which is maybe what you’re thinking of.
in c++, boost interprocess has offset_ptr which is useful since the shared data structure may be mapped at different locations in memory in each process
The FS and GS segment selectors are still used in x86-64, typically for `thread_local` storage, but they can be repurposed.
`thread_local` is an example of a "relative pointer" though. Instructions to access the thread local are prefixed with `fs:` or `gs:`, and point relative to the address in the respective segment register.
A far pointer sounds like the global based pointer described in that article. The far pointer Wikipedia article says they are problematic but doesn't give much reasoning as to why.
Far pointers are for accessing memory in different segments. They're basically obsolete now. They were necessary in older machines with limited sized pointers or address spaces.
GCC still supports `__seg_fs` and `__seg_gs`, which behave similar to `far` in the example on the wiki page, as the FS and GS segment registers are still valid in x86-64 and used for TLS. Clang uses attributes `address_space(257)` and `address_space(256)` for the same thing.
The `__based` pointer in MSVC exploits the addressing modes by pinning the base in eg: `[base+index*scale+displacement]`. It's unrelated to segmentation.
That's Fat pointers, not Far pointers. A fat pointer is a pointer with some other associated data which is stored in the pointer itself - typically by widening the number of bits used to hold a pointer value. The addressable bits usually remain unchanged - the added bits contain the auxiliary data.
Segmentation isn't used. There's no separate registers to hold the bounds information in CHERI - the bounds are held in the pointer value, unlike for example, the now obsolete Intel MPX, which held bounds information in separate registers.
There's some similarity to segmentation because the CHERI pointer restricts which addresses can be accessed, but I wouldn't compare them to far pointers.
Most modern processors have a single linear virtual address space and don't use segmentation, and even where segment registers exist (eg, FS and GS on x86-64), they're only superficial "address spaces" - allocated sections of the process's linear virtual address space which could be accessed without segmentation registers if you knew the base address held in FS or GS.
I think parent was after base+offset+index rather than just base+index.
Examples would be eg, `string_view` or `ArraySegment`. They hold some offset relative to a base allocation, and when we index the string_view or ArraySegment we're indexing relative to that offset.
This is how it is with languages which provide less guarantees than Rust. Sure you can try to hold all the invariants and restrictions in your head, but a sufficiently advanced compiler can do this for you without the possibility of making mistakes. I have no idea why people claim that's too restrictive - if you're not enforcing those rules manually you're just setting yourself up for issues down the road.
I reach for a low-level language only when I want low-level control over what operations happen and when, what memory is used and when etc.. At present, no language offers me this control and safety at the same time. With Rust, when I need such control (which is always, otherwise I would use a higher-level language), I need to give up safety, anyway, at which point I have no safety and the complexity of a language that offers safety.
So right now, when we want control, we need to give up some safety, but weaker things are still helpful.
Also, in low-level code, the problem of "I might forget to do something" sometimes clashes with the problem of "I need to see exactly what operations are done and where". Various kinds of implicitness help with the former at the expense of the latter.
I'm not saying this is universally better than other approaches, but many people who do serious low-level programming would prefer this.
> With Rust, when I need such control (which is always, otherwise I would use a higher-level language), I need to give up safety, anyway, at which point I have no safety and the complexity of a language that offers safety.
This is a very, very, very common claim. And unfortunately I have no other way to describe it other than a strawman.
In 95% (at least) of the application that need systems programming (not to talk about all applications that don't necessarily need it but will benefit from the performance and it wasn't an option because C++ wasn't an option), you have at most 20% (wildly overestimating) of code that needs to be unsafe. The rest could be completely safe. And amongst code that must be unsafe, you can very commonly encapsulate it in some safe pattern. Many times even extract it to a reusable crate.
And, if you have an unsafe block that is 100% correct, but it relies on safe code being correct, do you need to vet all that safe code? Potentially whole modules needing to be vetted?
Is unsafe Rust code generally harder to get correct than code in other languages, due to:
- Pointer aliasing requirements.
- That unsafe code must have no UB, even if uncontrolled safe code that it might interact with have any sorts of bugs, as per the check in https://doc.rust-lang.org/src/alloc/sync.rs.html#1953 . Since otherwise, if bugs in safe code is allowed to cause UB due to some specific unsafe code not being perfectly and completely 100% resistant to safe-code bugs, the safe-unsafe-split fails.
Not him, but projects that need performance often use unsafe or otherwise allow for UB. Embedded is arguably another example, since no_std allows UB even without unsafe, for instance by causing a stack overflow.
Except the point that Zig should do better than Object Pascal, Modula-2, with solutions already available on Insure++ and friends for use after free, 30 years ago.
But the point of unsafe {} in Rust is not that you should never use it, it's that it creates a clear boundary between code that is safe and the code that needs that lower level control. In other languages, everything is inside an unsafe block. If everything you do requires such low level control over every allocation and access, it sounds like you should be using assembly.
It's a nice feature but I can't help feeling like, if you need a stable pointer to an item in a collection, ArrayList is the wrong data structure to use? Maybe someone can chime in and give me an example of when you'd do this instead of, e.g., just storing an index. Alternatively, you could use an Unrolled Linked List (FKA SegmentedList in Zig before it was removed in 0.16, not sure why).
I count this as a "rookie at system programming" mistake alongside returning a reference to a local variable. Rust is great at this because borrow checker can catch those at compile time and it can _teach_ devs to not do that.
I don't know Zig, but conceptually: a direct pointer is the fastest way to access an object. An arraylist is the fastest dynamic sequence of objects (fattest in access, not in growth). You use these when you need the performance. It's not often but it certainly happens. The most trivial example is a string that you append to but still need to pass to a C API in between that expects it to be contagious, but it's far more useful than just for storing characters.
Indexing into an array is direct pointer access, there's just an addition in front of it but it's hard to imagine that showing up at all in even the tightest of benchmark loops
Sure, in cases where you need elements to be contiguous in memory then certainly an unrolled linked list is not appropriate. There's usually not a meaningful performance difference between a pointer deref and an indexed array access, however.
I use it in a lot of places where I know the max capacity ahead of time -- ensureCapacity() followed by a lot of *AssumeCapacity()-styled commands. It's convenient for all of the ... convenience ... methods (append() requires some bookkeeping somewhere, appendSlice() requires more, and so on). In those usages, it's basically syntactic sugar over a slice. That's not a perfect solution, but it's reasonably good often enough that I keep doing it.
The proposed change doesn't do much for me personally (memory safety is ensured in other ways, and if it weren't I wouldn't be annoyed debugging the allocator-observed errors), but I could see myself using it at some other point in time for the same class of usages, or I could see other people relying on it when they choose that class of coding.
SegmentedList had a weird API, especially the way you control the list growth factor by the size of an inline array. And it hadn't kept up with stdlib norms in recent versions. I do hope it comes back eventually with an improved API.
At least we got Deque in exchange. I use that far more often than I used SegmentedList.
Aha, gotcha. To tell you the truth, I don't think I ever used it. I have a custom implementation I wrote because I didn't realize at the time that SegmentedList was an unrolled linked list :p
ArrayList is a very generic (pun not intended) structure and could be stretched quite freely in any direction with useful property of owning underlying slice. Like readonly preallocated ArrayList is a thing.
My mental model of Zig is that it is explicitly the language for developers who prefer using pointers in business logic (instead of just in MMIMO, and are looking for something with improvements over C); i.e. exactly this class of abstraction.
Having written a bunch of Zig, I wouldn't say that the language design or culture explicitly encourages the use of pointers over indices in such situations. I would say it's more a language which trusts the programmer to make correct decisions about which constructs are appropriate in any given circumstance.
Hasn't the past 30 years of the Internet age taught us that given such trust, programmers will make the incorrect decision with horrifying predictability? The most trivial level of software security requires that pointer safety needs to be mathematically proven not up to human (or LLM) judgment.
This makes a lot of sense if you consider that it is consistent with the rest of the language. It is one more way to set up tripwires in your code to to catch your own programming errors. Similar to using asserts in your functions to vet input and output.
I use Array list a lot so excited to add this throughout the code to harden them.
I can imagine this is not everyone's cup of tea, but then you probably also wouldn't enjoy any of the other explicitness.
They forgot to add (I believe this was not deliberate, maybe their users already infer that) that this only actually performs the check on Debug and ReleaseSafe modes, not on ReleaseFast mode. Which is reasonable I guess because this is a memory write/read/branch in a super hot code path, but undermines a large part of the guarantee in my opinion (doesn't Zig have a debug allocator that could catch the mistake in the example just as well?).
Debug allocator can't catch it because it's not an allocation bug. Debug allocator finds bugs by marking memory during alloc/free and inspects them upon deinit. Pointer to a memory location change is not something allocator has control over. Possible solutions: smart array list implementation (this article), move semantic analysis (Rust's borrow checker), runtime introspection (https://fil-c.org/).
It is an allocation bug, it's a use-after-free. It's only a UAF if you actually have a reallocation (something that happens in the given example), but debug allocators don't require you to annotate your code.
It took me a minute to understand that this asserts on pointer change within the container, rather than lock/unlock the data structure like a SDL surface.
I recently implemented a custom C++ container for a path whose components could be iterated, backed by a std::string. I just store indices and a reference to the string, such that my iterators are not invalidated if the std::string gets reallocated after being modified. Far less error prone for little added cost.
This is a gripe of mine, and I will admit it is weak.
Changing a segfault to a panic with a stack trace is an improvement in developer experience. It does not make better software. The advantage of automatic strategies to mitigate memory safety mistakes either by using GC to make the program sound or static analysis to prevent the mistake by construction is plainly better.
There is a direction in some systems programming circles away from this by eschewing "complexity" (in other words, fixing the damn problems) for programs that have better error messages when the programmer made a mistake. I don't see that as better software.
In a language like Rust, the compiler will “lock” the pointers for you, and you can’t forget.
In a language like C++ (and presumably Zig), one could, in theory at least, have the iterators and slices that reference the storage of a dynamic array hold some sort of lock that pins the storage.
But this API requires the programmer to remember to lock the pointers and also requires the programmer to keep the lock alive for the correct region of code. And it looks to me like even the example in the blog post has the lock taken completely outside the function that requires stability, so there is nothing whatsoever that gets the lock scoping right. Even the type system can’t help — the offending parse function can’t declare that it wants a pointer-locked ArrayList parameter.
“I use it in a lot of places where I know the max capacity ahead of time -- ensureCapacity() followed by a lot of AssumeCapacity()-styled commands. It's convenient for all of the ... convenience ... methods (append() requires some bookkeeping somewhere, appendSlice() requires more, and so on). In those usages, it's basically syntactic sugar over a slice”*
I suspect “where I know the max capacity ahead of time” covers most if not all use cases (if it you use this without knowing max capacity, you either accept your code may panic, or you do some unlock, grow, lock again dance when you discover your initial estimate is wrong)
If so, wouldn’t adding a growable container where you specify capacity at construction time and removing access to the internal pointers of ArrayList be a better way to handle this?
There is a similar proposal for trait objects in rust.
Similarly, CPU architectures that use descriptors can (have to?) have languages with that notion.
`thread_local` is an example of a "relative pointer" though. Instructions to access the thread local are prefixed with `fs:` or `gs:`, and point relative to the address in the respective segment register.
A far pointer sounds like the global based pointer described in that article. The far pointer Wikipedia article says they are problematic but doesn't give much reasoning as to why.
GCC still supports `__seg_fs` and `__seg_gs`, which behave similar to `far` in the example on the wiki page, as the FS and GS segment registers are still valid in x86-64 and used for TLS. Clang uses attributes `address_space(257)` and `address_space(256)` for the same thing.
The `__based` pointer in MSVC exploits the addressing modes by pinning the base in eg: `[base+index*scale+displacement]`. It's unrelated to segmentation.
Project CHERI would like to disagree.
Segmentation isn't used. There's no separate registers to hold the bounds information in CHERI - the bounds are held in the pointer value, unlike for example, the now obsolete Intel MPX, which held bounds information in separate registers.
There's some similarity to segmentation because the CHERI pointer restricts which addresses can be accessed, but I wouldn't compare them to far pointers.
Most modern processors have a single linear virtual address space and don't use segmentation, and even where segment registers exist (eg, FS and GS on x86-64), they're only superficial "address spaces" - allocated sections of the process's linear virtual address space which could be accessed without segmentation registers if you knew the base address held in FS or GS.
Examples would be eg, `string_view` or `ArraySegment`. They hold some offset relative to a base allocation, and when we index the string_view or ArraySegment we're indexing relative to that offset.
So right now, when we want control, we need to give up some safety, but weaker things are still helpful.
Also, in low-level code, the problem of "I might forget to do something" sometimes clashes with the problem of "I need to see exactly what operations are done and where". Various kinds of implicitness help with the former at the expense of the latter.
I'm not saying this is universally better than other approaches, but many people who do serious low-level programming would prefer this.
This is a very, very, very common claim. And unfortunately I have no other way to describe it other than a strawman.
In 95% (at least) of the application that need systems programming (not to talk about all applications that don't necessarily need it but will benefit from the performance and it wasn't an option because C++ wasn't an option), you have at most 20% (wildly overestimating) of code that needs to be unsafe. The rest could be completely safe. And amongst code that must be unsafe, you can very commonly encapsulate it in some safe pattern. Many times even extract it to a reusable crate.
That is the point of Rust. Not avoiding unsafety, but limiting and encapsulating it. And evidence proves that to work (for example https://blog.google/security/rust-in-android-move-fast-fix-t...).
And, if you have an unsafe block that is 100% correct, but it relies on safe code being correct, do you need to vet all that safe code? Potentially whole modules needing to be vetted?
Is unsafe Rust code generally harder to get correct than code in other languages, due to:
- Pointer aliasing requirements.
- That unsafe code must have no UB, even if uncontrolled safe code that it might interact with have any sorts of bugs, as per the check in https://doc.rust-lang.org/src/alloc/sync.rs.html#1953 . Since otherwise, if bugs in safe code is allowed to cause UB due to some specific unsafe code not being perfectly and completely 100% resistant to safe-code bugs, the safe-unsafe-split fails.
- Pinning.
- Etc.
?
Do Rust libraries, including std, historically have had UB bugs? https://materialize.com/blog/rust-concurrency-bug-unbounded-...
Can Miri catch everything? https://github.com/rust-lang/rust/pull/139553#issuecomment-2...
Are all the rules of unsafe, pinning, etc. fully specified and easy to learn and reason about?
The proposed change doesn't do much for me personally (memory safety is ensured in other ways, and if it weren't I wouldn't be annoyed debugging the allocator-observed errors), but I could see myself using it at some other point in time for the same class of usages, or I could see other people relying on it when they choose that class of coding.
At least we got Deque in exchange. I use that far more often than I used SegmentedList.
Try running both of these examples. They only differ in a pair of curly braces.
https://play.rust-lang.org/?version=stable&mode=debug&editio...
https://play.rust-lang.org/?version=stable&mode=debug&editio...
https://fasterthanli.me/articles/a-rust-match-made-in-hell
Mojo handles this significantly better than Rust.
I use Array list a lot so excited to add this throughout the code to harden them.
I can imagine this is not everyone's cup of tea, but then you probably also wouldn't enjoy any of the other explicitness.
I recently implemented a custom C++ container for a path whose components could be iterated, backed by a std::string. I just store indices and a reference to the string, such that my iterators are not invalidated if the std::string gets reallocated after being modified. Far less error prone for little added cost.
Changing a segfault to a panic with a stack trace is an improvement in developer experience. It does not make better software. The advantage of automatic strategies to mitigate memory safety mistakes either by using GC to make the program sound or static analysis to prevent the mistake by construction is plainly better.
There is a direction in some systems programming circles away from this by eschewing "complexity" (in other words, fixing the damn problems) for programs that have better error messages when the programmer made a mistake. I don't see that as better software.