-
Notifications
You must be signed in to change notification settings - Fork 12.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Mitigate binary size impact of implicit caller location #70580
Comments
Stabilize `#[track_caller]`. # Stabilization Report RFC: [2091] Tracking issue: rust-lang#47809 ## Summary From the [rustc-dev-guide chapter][dev-guide]: > Take this example program: ```rust fn main() { let foo: Option<()> = None; foo.unwrap(); // this should produce a useful panic message! } ``` > Prior to Rust 1.42, panics like this `unwrap()` printed a location in libcore: ``` $ rustc +1.41.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value',...core\macros\mod.rs:15:40 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace. ``` > As of 1.42, we get a much more helpful message: ``` $ rustc +1.42.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value', example.rs:3:5 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace ``` > These error messages are achieved through a combination of changes to `panic!` internals to make use of `core::panic::Location::caller` and a number of `#[track_caller]` annotations in the standard library which propagate caller information. The attribute adds an implicit caller location argument to the ABI of annotated functions, but does not affect the type or MIR of the function. We implement the feature entirely in codegen and in the const evaluator. ## Bottom Line This PR stabilizes the use of `#[track_caller]` everywhere, including traits and extern blocks. It also stabilizes `core::panic::Location::caller`, although the use of that function in a const context remains gated by `#![feature(const_caller_location)]`. The implementation for the feature already changed the output of panic messages for a number of std functions, as described in the [1.42 release announcement]. The attribute's use in `Index` and `IndexMut` traits is visible to users since 1.44. ## Tests All of the tests for this feature live under [src/test/ui/rfc-2091-track-caller][tests] in the repo. Noteworthy cases: * [use of attr in std] * validates user-facing benefit of the feature * [trait attribute inheritance] * covers subtle behavior designed during implementation and not RFC'd * [const/codegen equivalence] * this was the result of a suspected edge case and investigation * [diverging function support] * covers an unresolved question from the RFC * [fn pointers and shims] * covers important potential sources of unsoundness ## Documentation The rustc-dev-guide now has a chapter on [Implicit Caller Location][dev-guide]. I have an [open PR to the reference][attr-reference-pr] documenting the attribute. The intrinsic's [wrapper] includes some examples as well. ## Implementation History * 2019-10-02: [`#[track_caller]` feature gate (RFC 2091 1/N) rust-lang#65037](rust-lang#65037) * Picked up the patch that @ayosec had started on the feature gate. * 2019-10-13: [Add `Instance::resolve_for_fn_ptr` (RFC 2091 rust-lang#2/N) rust-lang#65182](rust-lang#65182) * 2019-10-20: ~~[WIP Add MIR argument for #[track_caller] (RFC 2091 3/N) rust-lang#65258](rust-lang#65258 * Abandoned approach to send location as a MIR argument. * 2019-10-28: [`std::panic::Location` is a lang_item, add `core::intrinsics::caller_location` (RFC 2091 3/N) rust-lang#65664](rust-lang#65664) * 2019-12-07: [Implement #[track_caller] attribute. (RFC 2091 4/N) rust-lang#65881](rust-lang#65881) * 2020-01-04: [libstd uses `core::panic::Location` where possible. rust-lang#67137](rust-lang#67137) * 2020-01-08: [`Option::{expect,unwrap}` and `Result::{expect, expect_err, unwrap, unwrap_err}` have `#[track_caller]` rust-lang#67887](rust-lang#67887) * 2020-01-20: [Fix #[track_caller] and function pointers rust-lang#68302](rust-lang#68302) (fixed rust-lang#68178) * 2020-03-23: [#[track_caller] in traits rust-lang#69251](rust-lang#69251) * 2020-03-24: [#[track_caller] on core::ops::{Index, IndexMut}. rust-lang#70234](rust-lang#70234) * 2020-04-08 [Support `#[track_caller]` on functions in `extern "Rust" { ... }` rust-lang#70916](rust-lang#70916) ## Unresolveds ### From the RFC > Currently the RFC simply prohibit applying #[track_caller] to trait methods as a future-proofing > measure. **Resolved.** See the dev-guide documentation and the tests section above. > Diverging functions should be supported. **Resolved.** See the tests section above. > The closure foo::{{closure}} should inherit most attributes applied to the function foo, ... **Resolved.** This unknown was related to specifics of the implementation which were made irrelevant by the final implementation. ### Binary Size I [instrumented track_caller to use custom sections][measure-size] in a local build and discovered relatively minor binary size usage for the feature overall. I'm leaving the issue open to discuss whether we want to upstream custom section support. There's an [open issue to discuss mitigation strategies][mitigate-size]. Some decisions remain about the "right" strategies to reduce size without overly constraining the compiler implementation. I'd be excited to see someone carry that work forward but my opinion is that we shouldn't block stabilization on implementing compiler flags for redaction. ### Specialization There's an [open issue][specialization] on the semantics of the attribute in specialization chains. I'm inclined to move forward with stabilization without an exact resolution here given that specialization is itself unstable, but I also think it should be an easy question to resolve. ### Location only points to the start of a call span rust-lang#69977 was resolved by rust-lang#73182, and the next step should probably be to [extend `Location` with a notion of the end of a call](rust-lang#73554). ### Regression of std's panic messages rust-lang#70963 should be resolved by serializing span hygeine to crate metadata: rust-lang#68686. [2091]: https://github.com/rust-lang/rfcs/blob/master/text/2091-inline-semantic.md [dev-guide]: https://rustc-dev-guide.rust-lang.org/codegen/implicit-caller-location.html [specialization]: rust-lang#70293 [measure-size]: rust-lang#70579 [mitigate-size]: rust-lang#70580 [attr-reference-pr]: rust-lang/reference#742 [wrapper]: https://doc.rust-lang.org/nightly/core/panic/struct.Location.html#method.caller [tests]: https://github.com/rust-lang/rust/tree/master/src/test/ui/rfc-2091-track-caller [const/codegen equivalence]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/caller-location-fnptr-rt-ctfe-equiv.rs [diverging function support]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/diverging-caller-location.rs [use of attr in std]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/std-panic-locations.rs [fn pointers and shims]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-fn-ptr-with-arg.rs [trait attribute inheritance]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-trait-impls.rs [1.42 release announcement]: https://blog.rust-lang.org/2020/03/12/Rust-1.42.html#useful-line-numbers-in-option-and-result-panic-messages
Stabilize `#[track_caller]`. # Stabilization Report RFC: [2091] Tracking issue: rust-lang#47809 ## Summary From the [rustc-dev-guide chapter][dev-guide]: > Take this example program: ```rust fn main() { let foo: Option<()> = None; foo.unwrap(); // this should produce a useful panic message! } ``` > Prior to Rust 1.42, panics like this `unwrap()` printed a location in libcore: ``` $ rustc +1.41.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value',...core\macros\mod.rs:15:40 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace. ``` > As of 1.42, we get a much more helpful message: ``` $ rustc +1.42.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value', example.rs:3:5 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace ``` > These error messages are achieved through a combination of changes to `panic!` internals to make use of `core::panic::Location::caller` and a number of `#[track_caller]` annotations in the standard library which propagate caller information. The attribute adds an implicit caller location argument to the ABI of annotated functions, but does not affect the type or MIR of the function. We implement the feature entirely in codegen and in the const evaluator. ## Bottom Line This PR stabilizes the use of `#[track_caller]` everywhere, including traits and extern blocks. It also stabilizes `core::panic::Location::caller`, although the use of that function in a const context remains gated by `#![feature(const_caller_location)]`. The implementation for the feature already changed the output of panic messages for a number of std functions, as described in the [1.42 release announcement]. The attribute's use in `Index` and `IndexMut` traits is visible to users since 1.44. ## Tests All of the tests for this feature live under [src/test/ui/rfc-2091-track-caller][tests] in the repo. Noteworthy cases: * [use of attr in std] * validates user-facing benefit of the feature * [trait attribute inheritance] * covers subtle behavior designed during implementation and not RFC'd * [const/codegen equivalence] * this was the result of a suspected edge case and investigation * [diverging function support] * covers an unresolved question from the RFC * [fn pointers and shims] * covers important potential sources of unsoundness ## Documentation The rustc-dev-guide now has a chapter on [Implicit Caller Location][dev-guide]. I have an [open PR to the reference][attr-reference-pr] documenting the attribute. The intrinsic's [wrapper] includes some examples as well. ## Implementation History * 2019-10-02: [`#[track_caller]` feature gate (RFC 2091 1/N) rust-lang#65037](rust-lang#65037) * Picked up the patch that @ayosec had started on the feature gate. * 2019-10-13: [Add `Instance::resolve_for_fn_ptr` (RFC 2091 rust-lang#2/N) rust-lang#65182](rust-lang#65182) * 2019-10-20: ~~[WIP Add MIR argument for #[track_caller] (RFC 2091 3/N) rust-lang#65258](rust-lang#65258 * Abandoned approach to send location as a MIR argument. * 2019-10-28: [`std::panic::Location` is a lang_item, add `core::intrinsics::caller_location` (RFC 2091 3/N) rust-lang#65664](rust-lang#65664) * 2019-12-07: [Implement #[track_caller] attribute. (RFC 2091 4/N) rust-lang#65881](rust-lang#65881) * 2020-01-04: [libstd uses `core::panic::Location` where possible. rust-lang#67137](rust-lang#67137) * 2020-01-08: [`Option::{expect,unwrap}` and `Result::{expect, expect_err, unwrap, unwrap_err}` have `#[track_caller]` rust-lang#67887](rust-lang#67887) * 2020-01-20: [Fix #[track_caller] and function pointers rust-lang#68302](rust-lang#68302) (fixed rust-lang#68178) * 2020-03-23: [#[track_caller] in traits rust-lang#69251](rust-lang#69251) * 2020-03-24: [#[track_caller] on core::ops::{Index, IndexMut}. rust-lang#70234](rust-lang#70234) * 2020-04-08 [Support `#[track_caller]` on functions in `extern "Rust" { ... }` rust-lang#70916](rust-lang#70916) ## Unresolveds ### From the RFC > Currently the RFC simply prohibit applying #[track_caller] to trait methods as a future-proofing > measure. **Resolved.** See the dev-guide documentation and the tests section above. > Diverging functions should be supported. **Resolved.** See the tests section above. > The closure foo::{{closure}} should inherit most attributes applied to the function foo, ... **Resolved.** This unknown was related to specifics of the implementation which were made irrelevant by the final implementation. ### Binary Size I [instrumented track_caller to use custom sections][measure-size] in a local build and discovered relatively minor binary size usage for the feature overall. I'm leaving the issue open to discuss whether we want to upstream custom section support. There's an [open issue to discuss mitigation strategies][mitigate-size]. Some decisions remain about the "right" strategies to reduce size without overly constraining the compiler implementation. I'd be excited to see someone carry that work forward but my opinion is that we shouldn't block stabilization on implementing compiler flags for redaction. ### Specialization There's an [open issue][specialization] on the semantics of the attribute in specialization chains. I'm inclined to move forward with stabilization without an exact resolution here given that specialization is itself unstable, but I also think it should be an easy question to resolve. ### Location only points to the start of a call span rust-lang#69977 was resolved by rust-lang#73182, and the next step should probably be to [extend `Location` with a notion of the end of a call](rust-lang#73554). ### Regression of std's panic messages rust-lang#70963 should be resolved by serializing span hygeine to crate metadata: rust-lang#68686. [2091]: https://github.com/rust-lang/rfcs/blob/master/text/2091-inline-semantic.md [dev-guide]: https://rustc-dev-guide.rust-lang.org/codegen/implicit-caller-location.html [specialization]: rust-lang#70293 [measure-size]: rust-lang#70579 [mitigate-size]: rust-lang#70580 [attr-reference-pr]: rust-lang/reference#742 [wrapper]: https://doc.rust-lang.org/nightly/core/panic/struct.Location.html#method.caller [tests]: https://github.com/rust-lang/rust/tree/master/src/test/ui/rfc-2091-track-caller [const/codegen equivalence]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/caller-location-fnptr-rt-ctfe-equiv.rs [diverging function support]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/diverging-caller-location.rs [use of attr in std]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/std-panic-locations.rs [fn pointers and shims]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-fn-ptr-with-arg.rs [trait attribute inheritance]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-trait-impls.rs [1.42 release announcement]: https://blog.rust-lang.org/2020/03/12/Rust-1.42.html#useful-line-numbers-in-option-and-result-panic-messages
Stabilize `#[track_caller]`. # Stabilization Report RFC: [2091] Tracking issue: rust-lang#47809 ## Summary From the [rustc-dev-guide chapter][dev-guide]: > Take this example program: ```rust fn main() { let foo: Option<()> = None; foo.unwrap(); // this should produce a useful panic message! } ``` > Prior to Rust 1.42, panics like this `unwrap()` printed a location in libcore: ``` $ rustc +1.41.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value',...core\macros\mod.rs:15:40 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace. ``` > As of 1.42, we get a much more helpful message: ``` $ rustc +1.42.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value', example.rs:3:5 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace ``` > These error messages are achieved through a combination of changes to `panic!` internals to make use of `core::panic::Location::caller` and a number of `#[track_caller]` annotations in the standard library which propagate caller information. The attribute adds an implicit caller location argument to the ABI of annotated functions, but does not affect the type or MIR of the function. We implement the feature entirely in codegen and in the const evaluator. ## Bottom Line This PR stabilizes the use of `#[track_caller]` everywhere, including traits and extern blocks. It also stabilizes `core::panic::Location::caller`, although the use of that function in a const context remains gated by `#![feature(const_caller_location)]`. The implementation for the feature already changed the output of panic messages for a number of std functions, as described in the [1.42 release announcement]. The attribute's use in `Index` and `IndexMut` traits is visible to users since 1.44. ## Tests All of the tests for this feature live under [src/test/ui/rfc-2091-track-caller][tests] in the repo. Noteworthy cases: * [use of attr in std] * validates user-facing benefit of the feature * [trait attribute inheritance] * covers subtle behavior designed during implementation and not RFC'd * [const/codegen equivalence] * this was the result of a suspected edge case and investigation * [diverging function support] * covers an unresolved question from the RFC * [fn pointers and shims] * covers important potential sources of unsoundness ## Documentation The rustc-dev-guide now has a chapter on [Implicit Caller Location][dev-guide]. I have an [open PR to the reference][attr-reference-pr] documenting the attribute. The intrinsic's [wrapper] includes some examples as well. ## Implementation History * 2019-10-02: [`#[track_caller]` feature gate (RFC 2091 1/N) rust-lang#65037](rust-lang#65037) * Picked up the patch that @ayosec had started on the feature gate. * 2019-10-13: [Add `Instance::resolve_for_fn_ptr` (RFC 2091 rust-lang#2/N) rust-lang#65182](rust-lang#65182) * 2019-10-20: ~~[WIP Add MIR argument for #[track_caller] (RFC 2091 3/N) rust-lang#65258](rust-lang#65258 * Abandoned approach to send location as a MIR argument. * 2019-10-28: [`std::panic::Location` is a lang_item, add `core::intrinsics::caller_location` (RFC 2091 3/N) rust-lang#65664](rust-lang#65664) * 2019-12-07: [Implement #[track_caller] attribute. (RFC 2091 4/N) rust-lang#65881](rust-lang#65881) * 2020-01-04: [libstd uses `core::panic::Location` where possible. rust-lang#67137](rust-lang#67137) * 2020-01-08: [`Option::{expect,unwrap}` and `Result::{expect, expect_err, unwrap, unwrap_err}` have `#[track_caller]` rust-lang#67887](rust-lang#67887) * 2020-01-20: [Fix #[track_caller] and function pointers rust-lang#68302](rust-lang#68302) (fixed rust-lang#68178) * 2020-03-23: [#[track_caller] in traits rust-lang#69251](rust-lang#69251) * 2020-03-24: [#[track_caller] on core::ops::{Index, IndexMut}. rust-lang#70234](rust-lang#70234) * 2020-04-08 [Support `#[track_caller]` on functions in `extern "Rust" { ... }` rust-lang#70916](rust-lang#70916) ## Unresolveds ### From the RFC > Currently the RFC simply prohibit applying #[track_caller] to trait methods as a future-proofing > measure. **Resolved.** See the dev-guide documentation and the tests section above. > Diverging functions should be supported. **Resolved.** See the tests section above. > The closure foo::{{closure}} should inherit most attributes applied to the function foo, ... **Resolved.** This unknown was related to specifics of the implementation which were made irrelevant by the final implementation. ### Binary Size I [instrumented track_caller to use custom sections][measure-size] in a local build and discovered relatively minor binary size usage for the feature overall. I'm leaving the issue open to discuss whether we want to upstream custom section support. There's an [open issue to discuss mitigation strategies][mitigate-size]. Some decisions remain about the "right" strategies to reduce size without overly constraining the compiler implementation. I'd be excited to see someone carry that work forward but my opinion is that we shouldn't block stabilization on implementing compiler flags for redaction. ### Specialization There's an [open issue][specialization] on the semantics of the attribute in specialization chains. I'm inclined to move forward with stabilization without an exact resolution here given that specialization is itself unstable, but I also think it should be an easy question to resolve. ### Location only points to the start of a call span rust-lang#69977 was resolved by rust-lang#73182, and the next step should probably be to [extend `Location` with a notion of the end of a call](rust-lang#73554). ### Regression of std's panic messages rust-lang#70963 should be resolved by serializing span hygeine to crate metadata: rust-lang#68686. [2091]: https://github.com/rust-lang/rfcs/blob/master/text/2091-inline-semantic.md [dev-guide]: https://rustc-dev-guide.rust-lang.org/codegen/implicit-caller-location.html [specialization]: rust-lang#70293 [measure-size]: rust-lang#70579 [mitigate-size]: rust-lang#70580 [attr-reference-pr]: rust-lang/reference#742 [wrapper]: https://doc.rust-lang.org/nightly/core/panic/struct.Location.html#method.caller [tests]: https://github.com/rust-lang/rust/tree/master/src/test/ui/rfc-2091-track-caller [const/codegen equivalence]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/caller-location-fnptr-rt-ctfe-equiv.rs [diverging function support]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/diverging-caller-location.rs [use of attr in std]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/std-panic-locations.rs [fn pointers and shims]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-fn-ptr-with-arg.rs [trait attribute inheritance]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-trait-impls.rs [1.42 release announcement]: https://blog.rust-lang.org/2020/03/12/Rust-1.42.html#useful-line-numbers-in-option-and-result-panic-messages
Stabilize `#[track_caller]`. # Stabilization Report RFC: [2091] Tracking issue: rust-lang#47809 ## Summary From the [rustc-dev-guide chapter][dev-guide]: > Take this example program: ```rust fn main() { let foo: Option<()> = None; foo.unwrap(); // this should produce a useful panic message! } ``` > Prior to Rust 1.42, panics like this `unwrap()` printed a location in libcore: ``` $ rustc +1.41.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value',...core\macros\mod.rs:15:40 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace. ``` > As of 1.42, we get a much more helpful message: ``` $ rustc +1.42.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value', example.rs:3:5 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace ``` > These error messages are achieved through a combination of changes to `panic!` internals to make use of `core::panic::Location::caller` and a number of `#[track_caller]` annotations in the standard library which propagate caller information. The attribute adds an implicit caller location argument to the ABI of annotated functions, but does not affect the type or MIR of the function. We implement the feature entirely in codegen and in the const evaluator. ## Bottom Line This PR stabilizes the use of `#[track_caller]` everywhere, including traits and extern blocks. It also stabilizes `core::panic::Location::caller`, although the use of that function in a const context remains gated by `#![feature(const_caller_location)]`. The implementation for the feature already changed the output of panic messages for a number of std functions, as described in the [1.42 release announcement]. The attribute's use in `Index` and `IndexMut` traits is visible to users since 1.44. ## Tests All of the tests for this feature live under [src/test/ui/rfc-2091-track-caller][tests] in the repo. Noteworthy cases: * [use of attr in std] * validates user-facing benefit of the feature * [trait attribute inheritance] * covers subtle behavior designed during implementation and not RFC'd * [const/codegen equivalence] * this was the result of a suspected edge case and investigation * [diverging function support] * covers an unresolved question from the RFC * [fn pointers and shims] * covers important potential sources of unsoundness ## Documentation The rustc-dev-guide now has a chapter on [Implicit Caller Location][dev-guide]. I have an [open PR to the reference][attr-reference-pr] documenting the attribute. The intrinsic's [wrapper] includes some examples as well. ## Implementation History * 2019-10-02: [`#[track_caller]` feature gate (RFC 2091 1/N) rust-lang#65037](rust-lang#65037) * Picked up the patch that @ayosec had started on the feature gate. * 2019-10-13: [Add `Instance::resolve_for_fn_ptr` (RFC 2091 rust-lang#2/N) rust-lang#65182](rust-lang#65182) * 2019-10-20: ~~[WIP Add MIR argument for #[track_caller] (RFC 2091 3/N) rust-lang#65258](rust-lang#65258 * Abandoned approach to send location as a MIR argument. * 2019-10-28: [`std::panic::Location` is a lang_item, add `core::intrinsics::caller_location` (RFC 2091 3/N) rust-lang#65664](rust-lang#65664) * 2019-12-07: [Implement #[track_caller] attribute. (RFC 2091 4/N) rust-lang#65881](rust-lang#65881) * 2020-01-04: [libstd uses `core::panic::Location` where possible. rust-lang#67137](rust-lang#67137) * 2020-01-08: [`Option::{expect,unwrap}` and `Result::{expect, expect_err, unwrap, unwrap_err}` have `#[track_caller]` rust-lang#67887](rust-lang#67887) * 2020-01-20: [Fix #[track_caller] and function pointers rust-lang#68302](rust-lang#68302) (fixed rust-lang#68178) * 2020-03-23: [#[track_caller] in traits rust-lang#69251](rust-lang#69251) * 2020-03-24: [#[track_caller] on core::ops::{Index, IndexMut}. rust-lang#70234](rust-lang#70234) * 2020-04-08 [Support `#[track_caller]` on functions in `extern "Rust" { ... }` rust-lang#70916](rust-lang#70916) ## Unresolveds ### From the RFC > Currently the RFC simply prohibit applying #[track_caller] to trait methods as a future-proofing > measure. **Resolved.** See the dev-guide documentation and the tests section above. > Diverging functions should be supported. **Resolved.** See the tests section above. > The closure foo::{{closure}} should inherit most attributes applied to the function foo, ... **Resolved.** This unknown was related to specifics of the implementation which were made irrelevant by the final implementation. ### Binary Size I [instrumented track_caller to use custom sections][measure-size] in a local build and discovered relatively minor binary size usage for the feature overall. I'm leaving the issue open to discuss whether we want to upstream custom section support. There's an [open issue to discuss mitigation strategies][mitigate-size]. Some decisions remain about the "right" strategies to reduce size without overly constraining the compiler implementation. I'd be excited to see someone carry that work forward but my opinion is that we shouldn't block stabilization on implementing compiler flags for redaction. ### Specialization There's an [open issue][specialization] on the semantics of the attribute in specialization chains. I'm inclined to move forward with stabilization without an exact resolution here given that specialization is itself unstable, but I also think it should be an easy question to resolve. ### Location only points to the start of a call span rust-lang#69977 was resolved by rust-lang#73182, and the next step should probably be to [extend `Location` with a notion of the end of a call](rust-lang#73554). ### Regression of std's panic messages rust-lang#70963 should be resolved by serializing span hygeine to crate metadata: rust-lang#68686. [2091]: https://github.com/rust-lang/rfcs/blob/master/text/2091-inline-semantic.md [dev-guide]: https://rustc-dev-guide.rust-lang.org/codegen/implicit-caller-location.html [specialization]: rust-lang#70293 [measure-size]: rust-lang#70579 [mitigate-size]: rust-lang#70580 [attr-reference-pr]: rust-lang/reference#742 [wrapper]: https://doc.rust-lang.org/nightly/core/panic/struct.Location.html#method.caller [tests]: https://github.com/rust-lang/rust/tree/master/src/test/ui/rfc-2091-track-caller [const/codegen equivalence]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/caller-location-fnptr-rt-ctfe-equiv.rs [diverging function support]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/diverging-caller-location.rs [use of attr in std]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/std-panic-locations.rs [fn pointers and shims]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-fn-ptr-with-arg.rs [trait attribute inheritance]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-trait-impls.rs [1.42 release announcement]: https://blog.rust-lang.org/2020/03/12/Rust-1.42.html#useful-line-numbers-in-option-and-result-panic-messages
Stabilize `#[track_caller]`. # Stabilization Report RFC: [2091] Tracking issue: rust-lang#47809 ## Summary From the [rustc-dev-guide chapter][dev-guide]: > Take this example program: ```rust fn main() { let foo: Option<()> = None; foo.unwrap(); // this should produce a useful panic message! } ``` > Prior to Rust 1.42, panics like this `unwrap()` printed a location in libcore: ``` $ rustc +1.41.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value',...core\macros\mod.rs:15:40 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace. ``` > As of 1.42, we get a much more helpful message: ``` $ rustc +1.42.0 example.rs; example.exe thread 'main' panicked at 'called `Option::unwrap()` on a `None` value', example.rs:3:5 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace ``` > These error messages are achieved through a combination of changes to `panic!` internals to make use of `core::panic::Location::caller` and a number of `#[track_caller]` annotations in the standard library which propagate caller information. The attribute adds an implicit caller location argument to the ABI of annotated functions, but does not affect the type or MIR of the function. We implement the feature entirely in codegen and in the const evaluator. ## Bottom Line This PR stabilizes the use of `#[track_caller]` everywhere, including traits and extern blocks. It also stabilizes `core::panic::Location::caller`, although the use of that function in a const context remains gated by `#![feature(const_caller_location)]`. The implementation for the feature already changed the output of panic messages for a number of std functions, as described in the [1.42 release announcement]. The attribute's use in `Index` and `IndexMut` traits is visible to users since 1.44. ## Tests All of the tests for this feature live under [src/test/ui/rfc-2091-track-caller][tests] in the repo. Noteworthy cases: * [use of attr in std] * validates user-facing benefit of the feature * [trait attribute inheritance] * covers subtle behavior designed during implementation and not RFC'd * [const/codegen equivalence] * this was the result of a suspected edge case and investigation * [diverging function support] * covers an unresolved question from the RFC * [fn pointers and shims] * covers important potential sources of unsoundness ## Documentation The rustc-dev-guide now has a chapter on [Implicit Caller Location][dev-guide]. I have an [open PR to the reference][attr-reference-pr] documenting the attribute. The intrinsic's [wrapper] includes some examples as well. ## Implementation History * 2019-10-02: [`#[track_caller]` feature gate (RFC 2091 1/N) rust-lang#65037](rust-lang#65037) * Picked up the patch that @ayosec had started on the feature gate. * 2019-10-13: [Add `Instance::resolve_for_fn_ptr` (RFC 2091 rust-lang#2/N) rust-lang#65182](rust-lang#65182) * 2019-10-20: ~~[WIP Add MIR argument for #[track_caller] (RFC 2091 3/N) rust-lang#65258](rust-lang#65258 * Abandoned approach to send location as a MIR argument. * 2019-10-28: [`std::panic::Location` is a lang_item, add `core::intrinsics::caller_location` (RFC 2091 3/N) rust-lang#65664](rust-lang#65664) * 2019-12-07: [Implement #[track_caller] attribute. (RFC 2091 4/N) rust-lang#65881](rust-lang#65881) * 2020-01-04: [libstd uses `core::panic::Location` where possible. rust-lang#67137](rust-lang#67137) * 2020-01-08: [`Option::{expect,unwrap}` and `Result::{expect, expect_err, unwrap, unwrap_err}` have `#[track_caller]` rust-lang#67887](rust-lang#67887) * 2020-01-20: [Fix #[track_caller] and function pointers rust-lang#68302](rust-lang#68302) (fixed rust-lang#68178) * 2020-03-23: [#[track_caller] in traits rust-lang#69251](rust-lang#69251) * 2020-03-24: [#[track_caller] on core::ops::{Index, IndexMut}. rust-lang#70234](rust-lang#70234) * 2020-04-08 [Support `#[track_caller]` on functions in `extern "Rust" { ... }` rust-lang#70916](rust-lang#70916) ## Unresolveds ### From the RFC > Currently the RFC simply prohibit applying #[track_caller] to trait methods as a future-proofing > measure. **Resolved.** See the dev-guide documentation and the tests section above. > Diverging functions should be supported. **Resolved.** See the tests section above. > The closure foo::{{closure}} should inherit most attributes applied to the function foo, ... **Resolved.** This unknown was related to specifics of the implementation which were made irrelevant by the final implementation. ### Binary Size I [instrumented track_caller to use custom sections][measure-size] in a local build and discovered relatively minor binary size usage for the feature overall. I'm leaving the issue open to discuss whether we want to upstream custom section support. There's an [open issue to discuss mitigation strategies][mitigate-size]. Some decisions remain about the "right" strategies to reduce size without overly constraining the compiler implementation. I'd be excited to see someone carry that work forward but my opinion is that we shouldn't block stabilization on implementing compiler flags for redaction. ### Specialization There's an [open issue][specialization] on the semantics of the attribute in specialization chains. I'm inclined to move forward with stabilization without an exact resolution here given that specialization is itself unstable, but I also think it should be an easy question to resolve. ### Location only points to the start of a call span rust-lang#69977 was resolved by rust-lang#73182, and the next step should probably be to [extend `Location` with a notion of the end of a call](rust-lang#73554). ### Regression of std's panic messages rust-lang#70963 should be resolved by serializing span hygeine to crate metadata: rust-lang#68686. [2091]: https://github.com/rust-lang/rfcs/blob/master/text/2091-inline-semantic.md [dev-guide]: https://rustc-dev-guide.rust-lang.org/codegen/implicit-caller-location.html [specialization]: rust-lang#70293 [measure-size]: rust-lang#70579 [mitigate-size]: rust-lang#70580 [attr-reference-pr]: rust-lang/reference#742 [wrapper]: https://doc.rust-lang.org/nightly/core/panic/struct.Location.html#method.caller [tests]: https://github.com/rust-lang/rust/tree/master/src/test/ui/rfc-2091-track-caller [const/codegen equivalence]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/caller-location-fnptr-rt-ctfe-equiv.rs [diverging function support]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/diverging-caller-location.rs [use of attr in std]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/std-panic-locations.rs [fn pointers and shims]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-fn-ptr-with-arg.rs [trait attribute inheritance]: https://github.com/rust-lang/rust/blob/master/src/test/ui/rfc-2091-track-caller/tracked-trait-impls.rs [1.42 release announcement]: https://blog.rust-lang.org/2020/03/12/Rust-1.42.html#useful-line-numbers-in-option-and-result-panic-messages
For reference, I think the place to branch would probably be in rust/src/librustc_mir/interpret/intrinsics/caller_location.rs Lines 54 to 81 in 1fb612b
|
Also worth noting that if we leave the ABI unaffected then a libstd without track_caller would only be needed for users who really need to squeeze the size, rather than a requirement for anyone compiling without the locations. |
I am interested in taking on this issue. I took some basic measurements for an embedded binary here, and based on my findings I am interested in implementing |
…l, r=davidtwco Implement -Z location-detail flag This PR implements the `-Z location-detail` flag as described in rust-lang/rfcs#2091 . `-Z location-detail=val` controls what location details are tracked when using `caller_location`. This allows users to control what location details are printed as part of panic messages, by allowing them to exclude any combination of filenames, line numbers, and column numbers. This option is intended to provide users with a way to mitigate the size impact of `#[track_caller]`. Some measurements of the savings of this approach on an embedded binary can be found here: rust-lang#70579 (comment) . Closes rust-lang#70580 (unless people want to leave that open as a place for discussion of further improvements). This is my first real PR to rust, so any help correcting mistakes / understanding side effects / improving my tests is appreciated :) I have one question: RFC 2091 specified this as a debugging option (I think that is what -Z implies?). Does that mean this can never be stabilized without a separate MCP? If so, do I need to submit an MCP now, or is the initial RFC specifying this option sufficient for this to be merged as is, and then an MCP would be needed for eventual stabilization?
…l, r=davidtwco Implement -Z location-detail flag This PR implements the `-Z location-detail` flag as described in rust-lang/rfcs#2091 . `-Z location-detail=val` controls what location details are tracked when using `caller_location`. This allows users to control what location details are printed as part of panic messages, by allowing them to exclude any combination of filenames, line numbers, and column numbers. This option is intended to provide users with a way to mitigate the size impact of `#[track_caller]`. Some measurements of the savings of this approach on an embedded binary can be found here: rust-lang#70579 (comment) . Closes rust-lang#70580 (unless people want to leave that open as a place for discussion of further improvements). This is my first real PR to rust, so any help correcting mistakes / understanding side effects / improving my tests is appreciated :) I have one question: RFC 2091 specified this as a debugging option (I think that is what -Z implies?). Does that mean this can never be stabilized without a separate MCP? If so, do I need to submit an MCP now, or is the initial RFC specifying this option sufficient for this to be merged as is, and then an MCP would be needed for eventual stabilization?
Once we've measured the size impact of implicit caller location, we need to decide what (if any) knobs to offer users to mitigate that impact.
The RFC originally proposed a
-Zlocation-detail-control
flag for rustc which would allow users to control which fields ofLocation
were preserved.On the tracking issue I also proposed an all-or-nothing flag like
-Zredact-caller-location
which would either preserveLocation
s or rewrite them all to be the same.Note that both of these options are limited to controlling the locations reported by user code -- unless a user recompiles
std
with whatever flag we offer, there will always be "unredacted" locations in their binary.I'm fond of ideas that would allow us to replace
Location
s with something smaller that can be "symbolized" similarly to the way a debugger for an embedded device can identify the source location for a program counter value in a stacktrace. I think most options in this direction require help from the linker that isn't available, but I'd love to be wrong.The text was updated successfully, but these errors were encountered: