There have been some small soundness holes in the typesystem that we have fixed, resulting in breakage, but since that’s in the language, that’s the only way. A library API is different. There haven’t been many of these though. We did have some point releases which immediately fixed some library errors that were introduced by a release, see 1.15.1. But that was in a new API, and so the chance of breakage was extremely low. This API has been around for years. It is also not used much, given it’s a *NIX specific extension you have to explicitly import.
Note that upon using it, you’ll get a warning, so everyone will at least be notified.
Totally the correct way to go about it, IMO. Thus my comment about deprecation.
>> There will not be a Rust 2.0.
>> Note that upon using it, you’ll get a warning, so everyone will at least be notified.
:/ That I've mixed feelings about. Why not? Is it because of the fiascos between Perl 5/6 and Python 2/3? Deprecations and warnings about using deprecated features are indeed the correct thing to do, but IMO, aren't the whole picture. I think to all the times I went to compile something, and got to watch a constant stream of warnings for it (or worse, got hit by bugs I shouldn't have because the programmer took time to finally suppress that annoying warning pointing to the flaw in his code). A (very occasional) major version bump to show "These things we found out were actually wrong and said that you shouldn't do N years ago? We meant that, those are gone now," strikes me as the Right Thing for a language with a major focus on correctness, like Rust.
But, since I'm here armchair quarterbacking here on HN, instead of getting my hands dirty building a major language, I'm willing to concede that my opinion might be different were it informed by experience.
Rust does have a strong commitment to safety, but not an absolute commitment to correctness. These things happen very infrequently. Is it really worth taking an extreme action (which a major language version bump is, especially in the systems space) just to turn a few warnings into errors? Currently, we don’t think so. Maybe in 20 years, when (And if! :) ) there is more than one standard library API that suffers this problem, it would be worth it, but at the current time, it just doesn’t seem to make sense.
I think I'd expect to see warnings about the existing safe method being deprecated for a while, then it's complete removal at the time of another edition of Rust. Users of older versions of Rust can still use the removed function, but newer editions ban it.
This could be implemented as a reserved STD function list in the compiler if it must, since as you've mentioned we only have one STD lib atm.
Editions cannot make this kind of change, as also discussed below.
Never mind the fact that the function is still a symbol in the STD library.
I understand if this is too much complier magic, though it seems like the correct idea to me anyway.
Am I missing something?
P.S. symbol version pinning seems like a great idea, I've even dreamt of having exposed syntax for it, e.g. `Foo::<i32>::[email protected](arg1, arg2)`.
Sorry if I'm not making my thought very clear.
I hope I didn't come off as being to negative about the change. Warnings are still a pretty good solution here.
We can’t get rid of it because we have a commitment to not breaking users’ code. There will not be
a Rust 2.0.
This is interesting. The C Standard people have, for example, removed “gets” from the C11 standard.
Go, too, has exceptions to its Go 1 Compatibility Promise[1], which include security, unspecified
behaviour, and bugs, both in the language and in the standard library. I don't remember for sure,
but I think there were things removed from C++ specifications as well; please correct me if I am
wrong. Did Rust decided to never break stdlib compatibility because the mechanism of editions was
considered from the start, or was this rather ad-hoc?We have to do such things extremely judiciously though. Users can only tolerate so much breakage, even if you say “we did say we reserved the right to do this.” This API is just so rarely used that it was judged not a good time to pull the “technically we are allowed to do this” card.
When you don't have a standard, you're left at the mercy of languages promising not to break stuff. If they do anyway, your options aren't as clean as just sticking to the older standard.
Isn’t Rust 2018 effectively Rust 2.0 (since it introduced new keywords and requires changes to old code if a user opts in to Rust 2018)? So the function could be made unsafe in Rust 2021 (or whatever it’s called).
The standard library cannot change with editions for exactly this reason. It’s compiled with a particular edition, like any other crate, and so can’t work differently in different editions.
What about code examples from StackOverflow, or Github issues? Will these work without adjustments in my local project?
Will a future syntax highlighter always be able to highlight the code in my old Rust articles?
Will future tooling always be able to read my current code?
Can I make editor macros and snippets that will keep working in every version of Rust?
If I can write code in one crate and it will compile, but then write it again in another, newer crate and it doesn't, then it isn't backwards compatible. Plus, as noted above, "code will keep compiling" isn't even a guarantee that the language team gives you.
2. Same thing. You can construct an example that breaks any language in existence.
3. Yes.
4. Yes.
5. Yes.
6. You’re using an idiosyncratic definition, so yes, it’s not the same. Backwards compatibility is about the same thing continuing to compile, not about changing things and expecting it to still compile; that’s forwards compatibility.
I also don't know how current code running in a future compiler is forward-compatibility. And the Rust guarantee only kind-of holds if you define "code" as "crate". Everything outside of or crossing that boundary is not compatible.
To clarify and bring it to an example: Are you guaranteeing that `(a<b, c>(d))` will always be parsed as a tuple with two comparisons? Because if so I believe you're the only one.
Given the tight relationship of all of these parts (despite each one being capable of being individually versioned), it would seem feasible to add a cfg attribute for the Rust edition, and then make the function unsafe if the user had opted in to Rust 2021 (or whatever).
Think of it this way: crate A uses edition 2015. Crate B uses edition 2018. There’s only one copy of the standard library. It can’t be compiled both ways.
If, in theory, we let you have multiple copies of the standard library, maybe that could work, but that’s not possible nor desirable for a host of reasons.
glibc has a similar commitment to never breaking user code, and yet symbols have been removed from glibc! The trick is to use "symbol versioning"; code compiled against old versions of glibc is transparently rewritten to use the old versions of symbols, while code compiled against new versions is rewritten to use the new versions of the symbols. [0][1]
You can imagine something similar for the Rust standard library, whereby both the unsafe and safe versions are provided, and packages are transparently rewritten to use the correct version based on their specified edition. You don't need two copies of the stdlib; just of the symbols that have diverged. In this case, you wouldn't even need two versions of the symbol in the resulting binary, since the addition of `unsafe` doesn't change the codegen of the function, but merely restricts what code is accepted by the compiler.
It's true that this would require some serious shenanigans in the stdlib, but it's a tractable problem. It's also, in my mind, quite acceptable for the standard library to play games like this, since the compiler, the language, and the stdlib are always going to be tightly coupled, and the complexity can be shielded from users.
All that said, it sounds like the list of deprecated stdlib features is nowhere near long enough to justify adding this kind of complexity.
[0]: https://gcc.gnu.org/wiki/SymbolVersioning
[1]: Note that symbol versioning is kind of a quagmire, but that's because it's seriously underdocumented and practically no developers are aware that it exists. The idea itself is sound.
I think what you’re saying is “it is possible to do this with the correct design” and what I’m saying is “that design has not been proposed nor accepted and so the answer today is “that’s not possible”.” Does that sound about right?
And great! All I wanted to point out was that, should the need arise, it is technically possible to tie stdlib function signatures to a particular edition.
Why would it not be possible to have some additional information that says "from rust edition 2021 onwards this is actually obsolete". Whether crate B merely gives a warning on compilation or an error then depends on crate B's edition.
Sure that means you can still use the method if you really want to (by using older editions), and it also means that if some of your dependencies are still on an older edition you can't prevent it from being used by them, but it would be a higher hurdle to accidental usage.
There may be a way in which it is possible, but it is not currently possible. We’d have to have a different design.
This is not really true though. The commitment is about publicly visible and crater-testable code. If you have your own code in-house, there isn't much guarantee unless all syntax/code you use is widely used in public as well. A known change in language grammar will be evaluated on its publicly visible impact, not on its general breakiness.
It's really disappointing to keep reading about guaranteed backwards-compatibility and then being told in RFC discussions that those don't exist.
If backwards compatibility is guaranteed, turbofish cannot be "fixed". I can't find the link currently, but from my reddit post on this discussion the crucial quote from a language team member was:
> Our bar for doing backwards compatibility breaks has never been soundness fixes. We have in the past done changes given future-compatibility warning with lints and then made such changes without an edition.
The turbofish discussion also contains this quote:
> This RFC technically amounts to a backwards incompatible change without using the edition mechanism. However, as the RFC notes, a crater run was made and the syntax was not encountered at all.
You can see how controversial that RFC was, and for that exact reason.
Can you point me towards an authoritative post in the turbofish discussion that says that it can't happen because of backwards-compatibility changes? Because before it was locked it seemed that the language team wanted to go ahead.
I've been repeatedly told by language team members that the policy you're promising people doesn't exist. I would really appreciate it if there was further clarification.
My understanding is that, again, some lang team members want a different policy. That doesn’t mean that it’s actually different.
So, can you point me to a comment that tells me that the policy still holds and is as you are presenting it?
The resolution of this not only had a notable effect on how Rust thought about memory safety (in particular, that leaking memory could not be considered unsafe as it's always possible to construct a leak using safe Rust if you have access to Rc, and as a result `std::mem::forget()` was changed from unsafe to safe), but also informed the way we used RAII going forward, as the fundamental unsafety was due to the use of an RAII value as a token representing external computation and was not itself used in order to perform that computation, which meant leaking the token didn't prevent computation from happening. In all other known cases of RAII, leaking the value is safe because without the value, you can't access the protected resources, but in this case the token represented a thread and the thread of course would continue to run after you leak the token. As a takeaway from this lesson, you can't use tokens to represent external processes, and probably shouldn't use them to represent e.g. hardware that needs to be put back into a known state when the token is dropped unless you have a mutex in there as well (which would cause deadlock if you drop the token and then reuse the resource, rather than sending bad commands that break the hardware). Instead, any time leaking the token would cause unsafety, you need to structure your code to use explicit scopes instead, such that the library can clean up at the end of the scope.
This is the first such instance of "oops, turns out that wasn't safe after all, that should have been unsafe" that I've heard of in Rust. Is that because this is the first such mistake since the 1.0 milestone (the rest of `unsafe` having been nailed down before 1.0), or have there been other such mistakes that I didn't hear about because I haven't read the notes for all of the prior releases of Rust?