Skip to content
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

Span hygiene data should be serialized to crate metadata #68686

Closed
Aaron1011 opened this issue Jan 31, 2020 · 3 comments · Fixed by #72121
Closed

Span hygiene data should be serialized to crate metadata #68686

Aaron1011 opened this issue Jan 31, 2020 · 3 comments · Fixed by #72121
Labels
A-macros Area: All kinds of macros (custom derive, macro_rules!, proc macros, ..) A-metadata Area: Crate metadata C-enhancement Category: An issue proposing an enhancement or a PR with one. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue.

Comments

@Aaron1011
Copy link
Member

When we serialize a Span to crate metadata, we currently throw away the SyntaxContext:

impl rustc_serialize::UseSpecializedEncodable for Span {
fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
let span = self.data();
s.emit_struct("Span", 2, |s| {
s.emit_struct_field("lo", 0, |s| span.lo.encode(s))?;
s.emit_struct_field("hi", 1, |s| span.hi.encode(s))
})
}
}
impl rustc_serialize::UseSpecializedDecodable for Span {
fn default_decode<D: Decoder>(d: &mut D) -> Result<Span, D::Error> {
d.read_struct("Span", 2, |d| {
let lo = d.read_struct_field("lo", 0, Decodable::decode)?;
let hi = d.read_struct_field("hi", 1, Decodable::decode)?;
Ok(Span::with_root_ctxt(lo, hi))
})
}
}

This is because the backing HygieneData is stored in a thread-local in rustc_span, and not serialized into crate metadata.

The result is that spans deserialized from crate metadata may have less information available than spans from the current crate. If the MIR inlining pass decides to inline a function from another crate, we may end up with suboptimal messages when we invoke span.ctxt() (e.g. when emitting debuginfo, and when evaluating the caller_location intrinsic).

It would be useful if we were to serialize HygieneData into crate metadata, and deserialize spans with the proper SyntaxContext. This will also ensure that parallel compilation works properly, since storing HygieneData in a thread local will cause problems if a Span is used on multiple threads.

I'm not really sure how best to go about doing this. ExpnIds are currently unique per-crate, since they are never serialized. We need some way of making ExpnIds globally unique.

@petrochenkov
Copy link
Contributor

This is a big issue and one of the primary blockers for stabilizing Span::def_site (#54724, #54727) and declarative macros 2.0 (#39412).
There's a bunch of FIXMEs in the compiler about this, and I'm kind of surprised that there was no existing GitHub issue.

#49300 (comment) suggests a cross-crate stable representation for ExpnIds based on def-paths, which can be used for serializing them into metadata.

@Aaron1011

This comment has been minimized.

@rustbot rustbot added the A-macros Area: All kinds of macros (custom derive, macro_rules!, proc macros, ..) label Jan 31, 2020
@jonas-schievink jonas-schievink added A-metadata Area: Crate metadata C-enhancement Category: An issue proposing an enhancement or a PR with one. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Jan 31, 2020
@Aaron1011
Copy link
Member Author

Aaron1011 commented Feb 3, 2020

Here's an idea for how to serialize SyntaxContext itself:

We could just serialize the entire HygieneData, and serialize SyntaxContext as just the underlying u32. When we deserialize, we would lookup the corresponding SyntaxContextData in the serialized HygieneData. If we've already interned that SyntaxContextData, we would map the serialized SyntaxContext id to the id in the thread-local interner. If we haven't yet interned the SyntaxContextData, we would store it in HygieneData.syntax_context_data, and return the new id.

However, this approach interacts very badly with cross-crate serialization. Whenever we need to deserialize a Span, we will need to insert its SyntaxContextData into our local interner. This can happen a result of seemingly unrelated queries (e.g tcx.optimized_mir) which require deserializing Spans. This has the potential to use up a large number of non-interned ctxt_or_zero values in Span, causing us to intern spans that we could otherwise store 'inline'.

The situation gets worse if we need to re-serialize cross-crate spans into our own metadata. This will propagate the extra SyntaxContext ids to all downstream crates, effectively requiring them to intern SyntaxContextData instances for transitive dependencies. If we turn on MIR inlining by default, this could become very common, since libcore and libstd have many small #[inline] functions.

This is essentially the same problem I described in #68718 (comment), but for SyntaxContext instead of ExpnId. However, I don't think there's a straightforward way to map a SyntaxContext to a DefPath, so we'll need a different approach.

If we assume that 'crate-local' spans are accessed much more frequently than 'cross-crate' spans, then I think we can come up with a better solution. We can turn SyntaxContextData into a two-variant enum:

enum SyntaxContextData {
    Local(/* the current SyntaxContextData fields */),
    Remote(CrateNum, SyntaxContext)
}

When we create a fresh SyntaxContext (which is by definition crate-local), we will use SyntaxContextData::Local with the same fields as before.

When we deserialize Span from another crate's metadata, we will create a most one SyntaxContextData::Remote per crate. It will be of the form SyntaxContextData::Remote(cnum, remote_id), where remote_id stores the SyntaxContext used to index into the HygieneData of the remote crate.

To retrieve information about a SyntaxContextData:Remote, we can either create a hygiene_data(cnum) query that deserializes the HygieneData from crate cnum, or keep a HashMap<CrateNum, HygieneData> in memory to speed up lookups.

This will make access to SyntaxContextData for cross-crate spans slightly slower than accesses for crate-local spans. However, we will save almost all of the Span.ctxt_or_zero slots, allowing us to continue to store Spans in the inline format.

With this approach, we can also heuristically choose to deserialize some cross-crate SyntaxContexts as SyntaxContextData::Local rather than SyntaxContextData::Remote (e.g. if we have reason to believe that they will be accessed frequently).

bors added a commit that referenced this issue Feb 24, 2020
[experiment] rustc_metadata: Load metadata for indirect macro-only dependencies

Imagine this dependency chain between crates
```
Executable crate -> Library crate -> Macro crate
```
where "Library crate" uses the macros from "Macro crate" for some code generation, but doesn't reexport them any further.

Currently, when compiling "Executable crate" we don't even load metadata for it, because why would we want to load any metadata from "Macro crate" if it already did all its code generation job when compiling "Library crate".
Right?

Wrong!
Hygiene data and spans (#68686, #68941) from "Macro crate" still may need to be decoded from "Executable crate".
So we'll have to load them properly.

Questions:
- How this will affect compile times for larger crate trees in practice? How to measure it?
Hygiene/span encoding/decoding will necessarily slow down compilation because right now we just don't do some work that we should do, but this introduces a whole new way to slow down things. E.g. loading metadata for `syn` (and its dependencies) when compiling your executable if one of its library dependencies uses it.
- We are currently detecting whether a crate reexports macros from "Macro crate" or not, could we similarly detect whether a crate "reexports spans" and keep it unloaded if it doesn't?
Or at least "reexports important spans" affecting hygiene, we can probably lose spans that only affect diagnostics.
bors added a commit that referenced this issue Mar 2, 2020
rustc_metadata: Load metadata for indirect macro-only dependencies

Imagine this dependency chain between crates
```
Executable crate -> Library crate -> Macro crate
```
where "Library crate" uses the macros from "Macro crate" for some code generation, but doesn't reexport them any further.

Currently, when compiling "Executable crate" we don't even load metadata for it, because why would we want to load any metadata from "Macro crate" if it already did all its code generation job when compiling "Library crate".
Right?

Wrong!
Hygiene data and spans (#68686, #68941) from "Macro crate" still may need to be decoded from "Executable crate".
So we'll have to load them properly.

Questions:
- How this will affect compile times for larger crate trees in practice? How to measure it?
Hygiene/span encoding/decoding will necessarily slow down compilation because right now we just don't do some work that we should do, but this introduces a whole new way to slow down things. E.g. loading metadata for `syn` (and its dependencies) when compiling your executable if one of its library dependencies uses it.
- We are currently detecting whether a crate reexports macros from "Macro crate" or not, could we similarly detect whether a crate "reexports spans" and keep it unloaded if it doesn't?
Or at least "reexports important spans" affecting hygiene, we can probably lose spans that only affect diagnostics.
Aaron1011 added a commit to Aaron1011/rust that referenced this issue Mar 20, 2020
Previously, we threw away the `Span` associated with a definition's
identifier when we encoded crate metadata, causing us to lose location
and hygiene information.

We now store the identifier's `Span` in the crate metadata.
When we decode items from the metadata, we combine
the name and span back into an `Ident`.

This improves the output of several tests, which previously had messages
suppressed due to dummy spans.

This is a prerequisite for rust-lang#68686, since throwing away a `Span` means
that we lose hygiene information.
Dylan-DPC-zz pushed a commit to Dylan-DPC-zz/rust that referenced this issue Mar 22, 2020
…t, r=petrochenkov

Store idents for `DefPathData` into crate metadata

Previously, we threw away the `Span` associated with a definition's
identifier when we encoded crate metadata, causing us to lose location
and hygiene information.

We now store the identifier's `Span` in a side table, which gets encoded
into the crate metadata. When we decode items from the metadata, we
combine the name and span back into an `Ident`.

This improves the output of several tests, which previously had messages
suppressed due to dummy spans.

This is a prerequisite for rust-lang#68686, since throwing away a `Span` means
that we lose hygiene information.
Aaron1011 added a commit to Aaron1011/rust that referenced this issue Mar 23, 2020
Previously, we threw away the `Span` associated with a definition's
identifier when we encoded crate metadata, causing us to lose location
and hygiene information.

We now store the identifier's `Span` in the crate metadata.
When we decode items from the metadata, we combine
the name and span back into an `Ident`.

This improves the output of several tests, which previously had messages
suppressed due to dummy spans.

This is a prerequisite for rust-lang#68686, since throwing away a `Span` means
that we lose hygiene information.
Centril added a commit to Centril/rust that referenced this issue Mar 24, 2020
…t, r=petrochenkov

Store idents for `DefPathData` into crate metadata

Previously, we threw away the `Span` associated with a definition's
identifier when we encoded crate metadata, causing us to lose location
and hygiene information.

We now store the identifier's `Span` in a side table, which gets encoded
into the crate metadata. When we decode items from the metadata, we
combine the name and span back into an `Ident`.

This improves the output of several tests, which previously had messages
suppressed due to dummy spans.

This is a prerequisite for rust-lang#68686, since throwing away a `Span` means
that we lose hygiene information.
Centril added a commit to Centril/rust that referenced this issue Mar 24, 2020
…t, r=petrochenkov

Store idents for `DefPathData` into crate metadata

Previously, we threw away the `Span` associated with a definition's
identifier when we encoded crate metadata, causing us to lose location
and hygiene information.

We now store the identifier's `Span` in a side table, which gets encoded
into the crate metadata. When we decode items from the metadata, we
combine the name and span back into an `Ident`.

This improves the output of several tests, which previously had messages
suppressed due to dummy spans.

This is a prerequisite for rust-lang#68686, since throwing away a `Span` means
that we lose hygiene information.
Centril added a commit to Centril/rust that referenced this issue Mar 24, 2020
…t, r=petrochenkov

Store idents for `DefPathData` into crate metadata

Previously, we threw away the `Span` associated with a definition's
identifier when we encoded crate metadata, causing us to lose location
and hygiene information.

We now store the identifier's `Span` in a side table, which gets encoded
into the crate metadata. When we decode items from the metadata, we
combine the name and span back into an `Ident`.

This improves the output of several tests, which previously had messages
suppressed due to dummy spans.

This is a prerequisite for rust-lang#68686, since throwing away a `Span` means
that we lose hygiene information.
Manishearth added a commit to Manishearth/rust that referenced this issue Jun 30, 2020
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
Manishearth added a commit to Manishearth/rust that referenced this issue Jun 30, 2020
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
Manishearth added a commit to Manishearth/rust that referenced this issue Jul 1, 2020
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
Manishearth added a commit to Manishearth/rust that referenced this issue Jul 1, 2020
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
Manishearth added a commit to Manishearth/rust that referenced this issue Jul 1, 2020
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
@bors bors closed this as completed in fa36f96 Jul 27, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
A-macros Area: All kinds of macros (custom derive, macro_rules!, proc macros, ..) A-metadata Area: Crate metadata C-enhancement Category: An issue proposing an enhancement or a PR with one. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue.
Projects
None yet
Development

Successfully merging a pull request may close this issue.

4 participants