Prevent opaque types being instantiated twice with different regions within the same function

[oli-obk]

addresses https://github.com/orgs/rust-lang/projects/22/views/1?pane=issue&itemId=41329537

r? @compiler-errors

[oli-obk]

This test was exactly duplicated with fn foo in tests/ui/type-alias-impl-trait/multiple-def-uses-in-one-fn-lifetimes.rs

[compiler-errors]

This should probably work. Can you check that?

type Tait<'x> = impl Sized;
fn define<'a: 'b, 'b: 'a>(x: &'a u8, y: &'b u8) -> (Tait<'a>, Tait<'b>) {
    ((), ())
}

[compiler-errors]

This technically could break RPIT code, so I'd feel comfortable only after a crater run.

// doesnt work
// (bounds there to make params early-bound)
fn rpit<'a: 'a, 'b: 'b>() -> impl Sized {
    let () = rpit::<'b, 'a>();
}

// should work
fn rpit<'a: 'b, 'b: 'a>() -> impl Sized {
    let () = rpit::<'b, 'a>();
}

@bors try

[bors]

⌛ Trying commit 29b10dc with merge 5f19a76...

[bors]

☀️ Try build successful - checks-actions
Build commit: 5f19a76 (5f19a76e461cf77b2134117fdb950227589d6a08)

[compiler-errors]

@craterbot check

[craterbot]

👌 Experiment pr-116935 created and queued.
🤖 Automatically detected try build 5f19a76
🔍 You can check out the queue and this experiment's details.

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🚧 Experiment pr-116935 is now running

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🎉 Experiment pr-116935 is completed!
📊 84 regressed and 2 fixed (380145 total)
📰 Open the full report.

⚠️ If you notice any spurious failure please add them to the blacklist!
ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[oli-obk]

Almost everything in the root regressions list is spurious disk failures (and the rest is unrelated ICEs, possibly due to disk space issues), though

 thread 'rustc' panicked at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/compiler/rustc_serialize/src/opaque.rs:233:42:
[INFO] [stderr] range start index 1525578 out of range for slice of length 1269760
[INFO] [stderr] stack backtrace:
[INFO] [stderr]    0:     0x7fa125b6917c - std::backtrace_rs::backtrace::libunwind::trace::h1f94320f2e6a67ac
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/../../backtrace/src/backtrace/libunwind.rs:93:5
[INFO] [stderr]    1:     0x7fa125b6917c - std::backtrace_rs::backtrace::trace_unsynchronized::h20b345c7dcabad28
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/../../backtrace/src/backtrace/mod.rs:66:5
[INFO] [stderr]    2:     0x7fa125b6917c - std::sys_common::backtrace::_print_fmt::h1c6b579a86ac571e
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/sys_common/backtrace.rs:67:5
[INFO] [stderr]    3:     0x7fa125b6917c - <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt::h0676035e212b50ec
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/sys_common/backtrace.rs:44:22
[INFO] [stderr]    4:     0x7fa125bcb400 - core::fmt::rt::Argument::fmt::ha6041a4893633223
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/core/src/fmt/rt.rs:142:9
[INFO] [stderr]    5:     0x7fa125bcb400 - core::fmt::write::h11d7df817e44fac2
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/core/src/fmt/mod.rs:1117:17
[INFO] [stderr]    6:     0x7fa125b5d00f - std::io::Write::write_fmt::h27916ce7b82ceedf
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/io/mod.rs:1762:15
[INFO] [stderr]    7:     0x7fa125b68f64 - std::sys_common::backtrace::_print::h5c780b472386b0af
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/sys_common/backtrace.rs:47:5
[INFO] [stderr]    8:     0x7fa125b68f64 - std::sys_common::backtrace::print::h11bae2f1a0ab82fc
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/sys_common/backtrace.rs:34:9
[INFO] [stderr]    9:     0x7fa125b6bbf7 - std::panicking::default_hook::{{closure}}::h8296a8c48c99a0fb
[INFO] [stderr]   10:     0x7fa125b6b95f - std::panicking::default_hook::h99bc273f02cfb87d
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/panicking.rs:292:9
[INFO] [stderr]     Checking systemstat v0.2.3
[INFO] [stderr]   11:     0x7fa1286a1590 - std[fafc877e3f4be7b4]::panicking::update_hook::<alloc[81b46284dcdbf2c6]::boxed::Box<rustc_driver_impl[ad2e13ae3ebaaf24]::install_ice_hook::{closure#0}>>::{closure#0}
[INFO] [stderr]   12:     0x7fa125b6c338 - <alloc::boxed::Box<F,A> as core::ops::function::Fn<Args>>::call::h17f17d8ab4e43e66
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/alloc/src/boxed.rs:2021:9
[INFO] [stderr]   13:     0x7fa125b6c338 - std::panicking::rust_panic_with_hook::h48200e0ad4451fc0
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/panicking.rs:735:13
[INFO] [stderr]   14:     0x7fa125b6c08e - std::panicking::begin_panic_handler::{{closure}}::ha531ed5b4fb897c6
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/panicking.rs:609:13
[INFO] [stderr]   15:     0x7fa125b69646 - std::sys_common::backtrace::__rust_end_short_backtrace::h934ae3f8127b5c1c
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/sys_common/backtrace.rs:170:18
[INFO] [stderr]   16:     0x7fa125b6bdf2 - rust_begin_unwind
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/panicking.rs:597:5
[INFO] [stderr]   17:     0x7fa125bc7b25 - core::panicking::panic_fmt::hd0329d80ff1c2596
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/core/src/panicking.rs:72:14
[INFO] [stderr]   18:     0x7fa125bcdd72 - core::slice::index::slice_start_index_len_fail_rt::h9029932fc1a5fcc5
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/core/src/slice/index.rs:52:5
[INFO] [stderr]   19:     0x7fa125bcdd72 - core::slice::index::slice_start_index_len_fail::h8e29ebc9dc2644ae
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/core/src/slice/index.rs:40:9
[INFO] [stderr]   20:     0x7fa12a4163bb - <rustc_metadata[e36cc7b7e19843ab]::locator::CrateLocator>::extract_one
[INFO] [stderr]   21:     0x7fa12a41505a - <rustc_metadata[e36cc7b7e19843ab]::locator::CrateLocator>::extract_lib
[INFO] [stderr]   22:     0x7fa12a30c1fa - <rustc_metadata[e36cc7b7e19843ab]::locator::CrateLocator>::maybe_load_library_crate
[INFO] [stderr]   23:     0x7fa12a30b87e - <rustc_metadata[e36cc7b7e19843ab]::creader::CrateLoader>::load
[INFO] [stderr]   24:     0x7fa12a40cf0b - <rustc_metadata[e36cc7b7e19843ab]::creader::CrateLoader>::maybe_resolve_crate
[INFO] [stderr]   25:     0x7fa129f9ccc1 - <rustc_resolve[799b7f14818798b6]::Resolver>::early_resolve_ident_in_lexical_scope
[INFO] [stderr]   26:     0x7fa129fd19d2 - <rustc_resolve[799b7f14818798b6]::Resolver>::resolve_path_with_ribs
[INFO] [stderr]   27:     0x7fa129fa53a2 - <rustc_resolve[799b7f14818798b6]::Resolver as rustc_expand[34fa88f58ed26183]::base::ResolverExpand>::resolve_imports
[INFO] [stderr]   28:     0x7fa12a1c5a1a - <rustc_expand[34fa88f58ed26183]::expand::MacroExpander>::fully_expand_fragment
[INFO] [stderr]   29:     0x7fa12a1c4738 - <rustc_expand[34fa88f58ed26183]::expand::MacroExpander>::expand_crate
[INFO] [stderr]   30:     0x7fa12a51885f - rustc_interface[d6345c2814166138]::passes::resolver_for_lowering
[INFO] [stderr]   31:     0x7fa12a517d2b - rustc_query_impl[be6bc7142f779313]::plumbing::__rust_begin_short_backtrace::<rustc_query_impl[be6bc7142f779313]::query_impl::resolver_for_lowering::dynamic_query::{closure#2}::{closure#0}, rustc_middle[fff2dae698ec3512]::query::erase::Erased<[u8; 8usize]>>
[INFO] [stderr]   32:     0x7fa12a516db3 - rustc_query_system[162c1fefa01d75b4]::query::plumbing::try_execute_query::<rustc_query_impl[be6bc7142f779313]::DynamicConfig<rustc_query_system[162c1fefa01d75b4]::query::caches::SingleCache<rustc_middle[fff2dae698ec3512]::query::erase::Erased<[u8; 8usize]>>, false, false, false>, rustc_query_impl[be6bc7142f779313]::plumbing::QueryCtxt, false>
[INFO] [stderr]   33:     0x7fa12a516449 - rustc_query_impl[be6bc7142f779313]::query_impl::resolver_for_lowering::get_query_non_incr::__rust_end_short_backtrace
[INFO] [stderr]   34:     0x7fa12a52f04b - std[fafc877e3f4be7b4]::sys_common::backtrace::__rust_begin_short_backtrace::<rustc_interface[d6345c2814166138]::util::run_in_thread_with_globals<rustc_interface[d6345c2814166138]::interface::run_compiler<core[a27ae4b4bc8367c2]::result::Result<(), rustc_span[9fa80e483ea6db9a]::ErrorGuaranteed>, rustc_driver_impl[ad2e13ae3ebaaf24]::run_compiler::{closure#1}>::{closure#0}, core[a27ae4b4bc8367c2]::result::Result<(), rustc_span[9fa80e483ea6db9a]::ErrorGuaranteed>>::{closure#0}::{closure#0}, core[a27ae4b4bc8367c2]::result::Result<(), rustc_span[9fa80e483ea6db9a]::ErrorGuaranteed>>
[INFO] [stderr]   35:     0x7fa12a52e533 - <<std[fafc877e3f4be7b4]::thread::Builder>::spawn_unchecked_<rustc_interface[d6345c2814166138]::util::run_in_thread_with_globals<rustc_interface[d6345c2814166138]::interface::run_compiler<core[a27ae4b4bc8367c2]::result::Result<(), rustc_span[9fa80e483ea6db9a]::ErrorGuaranteed>, rustc_driver_impl[ad2e13ae3ebaaf24]::run_compiler::{closure#1}>::{closure#0}, core[a27ae4b4bc8367c2]::result::Result<(), rustc_span[9fa80e483ea6db9a]::ErrorGuaranteed>>::{closure#0}::{closure#0}, core[a27ae4b4bc8367c2]::result::Result<(), rustc_span[9fa80e483ea6db9a]::ErrorGuaranteed>>::{closure#1} as core[a27ae4b4bc8367c2]::ops::function::FnOnce<()>>::call_once::{shim:vtable#0}
[INFO] [stderr]   36:     0x7fa125b77105 - <alloc::boxed::Box<F,A> as core::ops::function::FnOnce<Args>>::call_once::h43cc50cdd0629acf
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/alloc/src/boxed.rs:2007:9
[INFO] [stderr]   37:     0x7fa125b77105 - <alloc::boxed::Box<F,A> as core::ops::function::FnOnce<Args>>::call_once::h5f04d292d31a8e17
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/alloc/src/boxed.rs:2007:9
[INFO] [stderr]   38:     0x7fa125b77105 - std::sys::unix::thread::Thread::new::thread_start::h2e74a97d86ab0436
[INFO] [stderr]                                at /rustc/5f19a76e461cf77b2134117fdb950227589d6a08/library/std/src/sys/unix/thread.rs:108:17
[INFO] [stderr]   39:     0x7fa1258e1ac3 - <unknown>
[INFO] [stderr]   40:     0x7fa125972bf4 - __clone
[INFO] [stderr]   41:                0x0 - <unknown>

had me stumped for a moment until I realized that opaque.rs is not opaque types, but some encoder terminology

[compiler-errors]

I'm gonna block this on T-lang approval of the new Min-TAIT plan. Otherwise implementation looks great to me.

@rustbot blocked

[traviscross]

@rustbot labels +I-lang-nominated

Nominating for T-lang to decide whether to adopt the following rule for Rust:

Within a single item, when multiple defining uses of an opaque type differ only by lifetime arguments, we must be able to prove that those lifetime arguments are equal, otherwise the code will be rejected, and once the new trait solver is stabilized, it will likely not be possible to lift this restriction in a fully backward compatible way.

This is called the "once modulo regions restriction".

This rule was described and proposed in the 2023-11-08 T-lang Mini-TAIT design meeting (minutes): rust-lang/lang-team#233.

For a full discussion of this question, see:

• Once modulo regions restriction
• Appendix C: Once modulo regions restriction details

The effect of this rule would be to reject the following RPIT code which is currently accepted:

pub trait Trait<'x> {}
impl<'x, T> Trait<'x> for T {}

pub fn test<'a, 'b>(a: &'a (), b: &'b ()) -> impl Trait<'a> {
    // ~^ ERROR lifetime may not live long enough
    // ~| NOTE argument requires that `'b` must outlive `'a`
    if false {
        let _ = test(b, b);
    }
    a
}

And to reject similar code under TAIT:

#![feature(type_alias_impl_trait)]

type Tait<'x> = impl Sized;
fn test<'a, 'b>(a: &'a (), b: &'b ()) -> Tait<'a> {
    // ~^ ERROR lifetime may not live long enough
    // ~| NOTE requires that `'a` must outlive `'b`
    // ~| NOTE requires that `'b` must outlive `'a`
    if false {
        let _: Tait<'b> = b;
    }
    a
}

And to also reject, e.g.:

#![feature(type_alias_impl_trait)]

type Tait<'x> = impl Sized;
fn test<'a, 'b>(a: &'a (), b: &'b ()) -> (Tait<'a>, Tait<'b>) {
    // ~^ ERROR lifetime may not live long enough
    // ~| NOTE requires that `'a` must outlive `'b`
    // ~| NOTE requires that `'b` must outlive `'a`
    (a, b)
}

[Jules-Bertholet]

This proposal worries me. It would reject code that:

• Is sound
• Is reasonable, plausibly could have real-world application
• Intuitively "should work"
• Has no obvious work-around
• Works with the old trait solver
• And there's apparently no hope of ever making it work again, ever?

Perhaps I just need to learn more about the internals of the new solver to understand the motivation. But from an outside perspective, this does not inspire confidence.

[compiler-errors]

Let me break down this list:

• Is sound

There's lots of sound code that doesn't work in the old trait solver or the new trait solver.

Especially when it comes to things that are equal modulo regions, we already have a lot of limitations when it comes to the trait solver distinguishing choices (e.g. where clauses), because the trait solver fundamentally is not built to distinguish lifetimes.

• Is reasonable, plausibly could have real-world application

I haven't seen any examples of code that rely on this property. I agree that it's plausible, but it's clear that our guestimates of the probability of this being an issue that is hit in practice differ.

My understanding from discussing this briefly with @oli-obk is that he only wrote the tests that exercise this behavior because they could work, not because they were motivated by any real-life examples (though I may be misremembering, so any paraphrasing is my own).

• Intuitively "should work"

Again, lots of things that intuitively should work don't due to real, concrete limitations to the compiler, especially ones that enable other code to work, and for the trait solver to be sound. I go into more detail below, but this limitation exists to make space for lazy normalization and region uniquification.

• Has no obvious work-around

You can use type or const parameters to work around this, but again, a fix would likely be motivated by first understanding what is trying to be achieved at a high level.

• Works with the old trait solver

The purpose of this limitation being put in place is precisely to level out the differences between the new trait solver and the old trait solver. See below.

• And there's apparently no hope of ever making it work again, ever?

We are not technically committing to this behavior until the new trait solver lands, but yes, I don't think there's a way to make this work soundly in a language that doesn't do borrow-checking in tandem with type-checking.

Hopefully at the point of stabilization of the new solver, we will be able to more precisely motivate the reasons why the trait solver requires this to work the way it does. but TL;DR is "lazy normalization" and "region uniquification". The former is a property that we need to have to make the trait solver sound, and the latter is one that gives us better caching and is a requirement to make the trait solver work properly between typeck and borrowck.

The side-effect of this limitation is that we can treat defining instantiations of TAITs as being defined anywhere in the body. For example (https://godbolt.org/z/v9GjnP769) (though don't go trying it with region args yet, that still fails due to bugs with aliases in the new trait solver).

[aliemjay]

While the concern of higher-ranked-lifetimes is still applicable and may block the new capture rules (#117587), I don't see a reason for it to block ATPIT stabilization since we have a hard error now when opaque types mention higher ranked regions (#121386) and I believe that's forward compatible with any future decision. For this reason I am going to resolve my concern here and mention it as a blocker in #117587. Thanks to @traviscross for bringing this to my attention.

@rfcbot resolve higher-ranked-lifetimes
@rfcbot reviewed

[rfcbot]

🔔 This is now entering its final comment period, as per the review above. 🔔

[traviscross]

@rustbot labels -I-lang-nominated

Huge thanks to @aliemjay for raising this important issue and for discussing, analyzing, and working through it with me.

Since this is now in FCP, we can remove the nomination.

[rfcbot]

The final comment period, with a disposition to merge, as per the review above, is now complete.

As the automated representative of the governance process, I would like to thank the author for their work and everyone else who contributed.

This will be merged soon.

[compiler-errors]

@bors r+ rollup=never (bisectability)

[bors]

📌 Commit be9317d has been approved by compiler-errors

It is now in the queue for this repository.

[bors]

⌛ Testing commit be9317d with merge a7e4de1...

[bors]

☀️ Test successful - checks-actions
Approved by: compiler-errors
Pushing a7e4de1 to master...

[rust-timer]

Finished benchmarking commit (a7e4de1): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	2.1%	[2.0%, 2.2%]	3
Regressions ❌ (secondary)	3.5%	[2.1%, 5.6%]	29
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	2.1%	[2.0%, 2.2%]	3

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 670.459s -> 668.419s (-0.30%)
Artifact size: 312.77 MiB -> 312.82 MiB (0.02%)