#[ffi_returns_twice]

text/0000-returns-twice.md

@@ -0,0 +1,115 @@
+- Feature Name: ffi-returns-twice
+- Start Date: 2019.02.08
+- RFC PR: (leave this empty)
+- Rust Issue: (leave this empty)
+
+# Summary
+[summary]: #summary
+
+This RFC adds a new function attribute, `#[ffi_returns_twice]`, which indicates
+that a foreign function can return multiple times.
+
+# Motivation
+[motivation]: #motivation
+
+Rust assumes that function calls like:
+
+```rust
+let x = foo();
+```
+
+return only once. That is, when the execution arrives at the function call, the
+function is called, it returns a single value, and that's it. This assumption
+allows Rust to perform many optimizations.
+
+However, some foreign functions like [`setjmp`] and [`vfork`] can return
+multiple times.
+
+The `#[ffi_returns_twice]` attribute specifies that a foreign function might
+returns multiple times, inhibiting optimizations that assume that this is never
+the case.
+
+[`setjmp`]: https://en.cppreference.com/w/cpp/utility/program/setjmp
+[`longjmp`]: https://en.cppreference.com/w/cpp/utility/program/longjmp
+[`vfork`]: http://man7.org/linux/man-pages/man2/vfork.2.html
+
+# Guide-level and reference-level explanation
+[guide-level-explanation]: #guide-level-explanation
+
+The `#[ffi_returns_twice]` function attribute specifies that a foreign function
+might return multiple times, disabling optimizations that are incorrect for such
+functions. Two examples of such functions are [`setjmp`] and [`vfork`].
+
+# Drawbacks
+[drawbacks]: #drawbacks
+
+This complicates the language creating a new kind of functions.
+
+# Prior art
+[prior-art]: #prior-art
+
+This attribute is provided in both [LLVM] and [GCC]. 
+
+[LLVM]: https://llvm.org/docs/LangRef.html#id979
+[GCC]: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html
+
+# Rationale and alternatives
+[rationale-and-alternatives]: #rationale-and-alternatives
+
+In some platforms, efficient implementations of standard library APIs have to
+call low-level platform APIs that return multiple times. For example, the
+widely-used [`musl`] library implements [`posix_spawn`] on Linux by calling
+[`vfork`] (see: [`musl` blog post]). Rust implementations of such libraries (for
+example, see: [`steed`]) should be able to call these platform interfaces
+without invoking undefined behavior.
+
+This RFC introduces a function attribute that disables incorrect optimizations
+only for those functions for which they would be incorrect.
+
+One alternative could be to remove the assumption that Rust functions return at
+most once, eliminating the need for this attribute. This would have negative
+performance implications for most Rust functions, which do not return multiple
+times, making Rust functions a non-zero-cost abstraction.
+
+Another alternative could be to not support interfacing with this type of
+platform APIs, requiring users to use a different programming language (C,
+Assembly, etc.) to do so.
+
+[`posix_spawn`]: http://man7.org/linux/man-pages/man3/posix_spawn.3.html
+[`musl`]: https://www.musl-libc.org/
+[`musl` blog post]: https://ewontfix.com/7/
+[`steed`]: https://github.com/japaric/steed
+
+# Unresolved questions
+[unresolved-questions]: #unresolved-questions
+
+* Is this attribute sufficient to make write programs with functions that return
+multiple times possible?
+
+* Should we give this attribute a different name, e.g.,
+  `#[might_return_multiple_times]` or similar? Currently, this attribute is
+  called `returns_twice` because that's how the C attribute is called. Using the
+  same name as C here makes life easier for them.
+
+* Should we namespace `ffi`-only attributes somehow (e.g. `#[ffi(returns_twice,
+  foo, bar)]` ? See: https://github.com/rust-lang/rfcs/issues/2637
+
+# Future possibilities
+[future-possibilities]: #future-possibilities
+
+This RFC attempts to make writing programs that interface with foreign functions
+that return multiple times possible, but doing so is very challenging because it
+is trivial to introduce undefined behavior in those programs.
+
+Future RFCs could try to make it easier to avoid certain types of undefined
+behavior. For example, we could extend the borrow checker to take
+`#[ffi_returns_twice]` into account and reject programs that would have
+undefined behavior of the type "use-after-move".
+
+In the presence of types that implement `Drop`, usage of APIs that return
+multiple times requires extreme care to avoid deallocating memory without
+invoking destructors - this is critical for using [`Pin`] safely. We could also
+explore diagnosing these cases.
+
+[`Drop`]: https://doc.rust-lang.org/std/ops/trait.Drop.html
+[`Pin`]: https://doc.rust-lang.org/std/pin/struct.Pin.html