From 613f53ef19c94e461428dff1df3db7fe9bed5fb9 Mon Sep 17 00:00:00 2001 From: Ralf Jung Date: Sun, 3 Nov 2024 19:58:57 +0100 Subject: [PATCH] add const_eval_select macro to reduce redundancy also move internal const_panic helpers to a better location --- library/core/src/char/methods.rs | 2 +- library/core/src/ffi/c_str.rs | 95 +++++++------- library/core/src/intrinsics.rs | 83 ++++++++++-- library/core/src/macros/mod.rs | 61 --------- library/core/src/num/f128.rs | 2 +- library/core/src/num/f16.rs | 2 +- library/core/src/num/f32.rs | 2 +- library/core/src/num/f64.rs | 2 +- library/core/src/num/mod.rs | 2 +- library/core/src/panic.rs | 56 ++++++++ library/core/src/panicking.rs | 62 +++++---- library/core/src/ptr/const_ptr.rs | 128 +++++++++--------- library/core/src/ptr/mut_ptr.rs | 81 ++++++------ library/core/src/slice/ascii.rs | 158 +++++++++++------------ library/core/src/slice/index.rs | 2 +- library/core/src/slice/memchr.rs | 98 +++++++------- library/core/src/str/validations.rs | 17 +-- library/core/src/ub_checks.rs | 83 ++++++------ tests/ui/consts/const-ptr-is-null.stderr | 4 +- 19 files changed, 481 insertions(+), 459 deletions(-) diff --git a/library/core/src/char/methods.rs b/library/core/src/char/methods.rs index 701e34b135e..3915afb49d6 100644 --- a/library/core/src/char/methods.rs +++ b/library/core/src/char/methods.rs @@ -1,7 +1,7 @@ //! impl char {} use super::*; -use crate::macros::const_panic; +use crate::panic::const_panic; use crate::slice; use crate::str::from_utf8_unchecked_mut; use crate::unicode::printable::is_printable; diff --git a/library/core/src/ffi/c_str.rs b/library/core/src/ffi/c_str.rs index 4ea5cbf8626..85571222b5c 100644 --- a/library/core/src/ffi/c_str.rs +++ b/library/core/src/ffi/c_str.rs @@ -3,11 +3,12 @@ use crate::cmp::Ordering; use crate::error::Error; use crate::ffi::c_char; +use crate::intrinsics::const_eval_select; use crate::iter::FusedIterator; use crate::marker::PhantomData; use crate::ptr::NonNull; use crate::slice::memchr; -use crate::{fmt, intrinsics, ops, slice, str}; +use crate::{fmt, ops, slice, str}; // FIXME: because this is doc(inline)d, we *have* to use intra-doc links because the actual link // depends on where the item is being documented. however, since this is libcore, we can't @@ -411,37 +412,35 @@ pub const fn from_bytes_with_nul(bytes: &[u8]) -> Result<&Self, FromBytesWithNul #[rustc_const_stable(feature = "const_cstr_unchecked", since = "1.59.0")] #[rustc_allow_const_fn_unstable(const_eval_select)] pub const unsafe fn from_bytes_with_nul_unchecked(bytes: &[u8]) -> &CStr { - #[inline] - fn rt_impl(bytes: &[u8]) -> &CStr { - // Chance at catching some UB at runtime with debug builds. - debug_assert!(!bytes.is_empty() && bytes[bytes.len() - 1] == 0); + const_eval_select!( + @capture { bytes: &[u8] } -> &CStr: + if const { + // Saturating so that an empty slice panics in the assert with a good + // message, not here due to underflow. + let mut i = bytes.len().saturating_sub(1); + assert!(!bytes.is_empty() && bytes[i] == 0, "input was not nul-terminated"); - // SAFETY: Casting to CStr is safe because its internal representation - // is a [u8] too (safe only inside std). - // Dereferencing the obtained pointer is safe because it comes from a - // reference. Making a reference is then safe because its lifetime - // is bound by the lifetime of the given `bytes`. - unsafe { &*(bytes as *const [u8] as *const CStr) } - } + // Ending nul byte exists, skip to the rest. + while i != 0 { + i -= 1; + let byte = bytes[i]; + assert!(byte != 0, "input contained interior nul"); + } - const fn const_impl(bytes: &[u8]) -> &CStr { - // Saturating so that an empty slice panics in the assert with a good - // message, not here due to underflow. - let mut i = bytes.len().saturating_sub(1); - assert!(!bytes.is_empty() && bytes[i] == 0, "input was not nul-terminated"); + // SAFETY: See runtime cast comment below. + unsafe { &*(bytes as *const [u8] as *const CStr) } + } else { + // Chance at catching some UB at runtime with debug builds. + debug_assert!(!bytes.is_empty() && bytes[bytes.len() - 1] == 0); - // Ending nul byte exists, skip to the rest. - while i != 0 { - i -= 1; - let byte = bytes[i]; - assert!(byte != 0, "input contained interior nul"); + // SAFETY: Casting to CStr is safe because its internal representation + // is a [u8] too (safe only inside std). + // Dereferencing the obtained pointer is safe because it comes from a + // reference. Making a reference is then safe because its lifetime + // is bound by the lifetime of the given `bytes`. + unsafe { &*(bytes as *const [u8] as *const CStr) } } - - // SAFETY: See `rt_impl` cast. - unsafe { &*(bytes as *const [u8] as *const CStr) } - } - - intrinsics::const_eval_select((bytes,), const_impl, rt_impl) + ) } /// Returns the inner pointer to this C string. @@ -735,29 +734,27 @@ fn as_ref(&self) -> &CStr { #[cfg_attr(bootstrap, rustc_const_stable(feature = "const_cstr_from_ptr", since = "1.81.0"))] #[rustc_allow_const_fn_unstable(const_eval_select)] const unsafe fn strlen(ptr: *const c_char) -> usize { - const fn strlen_ct(s: *const c_char) -> usize { - let mut len = 0; + const_eval_select!( + @capture { s: *const c_char = ptr } -> usize: + if const { + let mut len = 0; - // SAFETY: Outer caller has provided a pointer to a valid C string. - while unsafe { *s.add(len) } != 0 { - len += 1; + // SAFETY: Outer caller has provided a pointer to a valid C string. + while unsafe { *s.add(len) } != 0 { + len += 1; + } + + len + } else { + extern "C" { + /// Provided by libc or compiler_builtins. + fn strlen(s: *const c_char) -> usize; + } + + // SAFETY: Outer caller has provided a pointer to a valid C string. + unsafe { strlen(s) } } - - len - } - - #[inline] - fn strlen_rt(s: *const c_char) -> usize { - extern "C" { - /// Provided by libc or compiler_builtins. - fn strlen(s: *const c_char) -> usize; - } - - // SAFETY: Outer caller has provided a pointer to a valid C string. - unsafe { strlen(s) } - } - - intrinsics::const_eval_select((ptr,), strlen_ct, strlen_rt) + ) } /// An iterator over the bytes of a [`CStr`], without the nul terminator. diff --git a/library/core/src/intrinsics.rs b/library/core/src/intrinsics.rs index fc09da7bcbc..3e5c0859230 100644 --- a/library/core/src/intrinsics.rs +++ b/library/core/src/intrinsics.rs @@ -2788,6 +2788,68 @@ pub const fn const_eval_select( unreachable!() } +/// A macro to make it easier to invoke const_eval_select. Use as follows: +/// ```rust,ignore (just a macro example) +/// const_eval_select!( +/// @capture { arg1: i32 = some_expr, arg2: T = other_expr } -> U: +/// if const #[attributes_for_const_arm] { +/// // Compile-time code goes here. +/// } else #[attributes_for_runtime_arm] { +/// // Run-time code goes here. +/// } +/// ) +/// ``` +/// The `@capture` block declares which surrounding variables / expressions can be +/// used inside the `if const`. +/// Note that the two arms of this `if` really each become their own function, which is why the +/// macro supports setting attributes for those functions. The runtime function is always +/// markes as `#[inline]`. +/// +/// See [`const_eval_select()`] for the rules and requirements around that intrinsic. +pub(crate) macro const_eval_select { + ( + @capture { $($arg:ident : $ty:ty = $val:expr),* $(,)? } $( -> $ret:ty )? : + if const + $(#[$compiletime_attr:meta])* $compiletime:block + else + $(#[$runtime_attr:meta])* $runtime:block + ) => {{ + #[inline] // avoid the overhead of an extra fn call + $(#[$runtime_attr])* + fn runtime($($arg: $ty),*) $( -> $ret )? { + $runtime + } + + #[inline] // prevent codegen on this function + $(#[$compiletime_attr])* + const fn compiletime($($arg: $ty),*) $( -> $ret )? { + // Don't warn if one of the arguments is unused. + $(let _ = $arg;)* + + $compiletime + } + + const_eval_select(($($val,)*), compiletime, runtime) + }}, + // We support leaving away the `val` expressions for *all* arguments + // (but not for *some* arguments, that's too tricky). + ( + @capture { $($arg:ident : $ty:ty),* $(,)? } $( -> $ret:ty )? : + if const + $(#[$compiletime_attr:meta])* $compiletime:block + else + $(#[$runtime_attr:meta])* $runtime:block + ) => { + $crate::intrinsics::const_eval_select!( + @capture { $($arg : $ty = $arg),* } $(-> $ret)? : + if const + $(#[$compiletime_attr])* $compiletime + else + $(#[$runtime_attr])* $runtime + ) + }, +} + /// Returns whether the argument's value is statically known at /// compile-time. /// @@ -2830,7 +2892,7 @@ pub const fn const_eval_select( /// # Stability concerns /// /// While it is safe to call, this intrinsic may behave differently in -/// a `const` context than otherwise. See the [`const_eval_select`] +/// a `const` context than otherwise. See the [`const_eval_select()`] /// documentation for an explanation of the issues this can cause. Unlike /// `const_eval_select`, this intrinsic isn't guaranteed to behave /// deterministically even in a `const` context. @@ -3734,14 +3796,15 @@ pub(crate) const fn miri_promise_symbolic_alignment(ptr: *const (), align: usize fn miri_promise_symbolic_alignment(ptr: *const (), align: usize); } - fn runtime(ptr: *const (), align: usize) { - // SAFETY: this call is always safe. - unsafe { - miri_promise_symbolic_alignment(ptr, align); + const_eval_select!( + @capture { ptr: *const (), align: usize}: + if const { + // Do nothing. + } else { + // SAFETY: this call is always safe. + unsafe { + miri_promise_symbolic_alignment(ptr, align); + } } - } - - const fn compiletime(_ptr: *const (), _align: usize) {} - - const_eval_select((ptr, align), compiletime, runtime); + ) } diff --git a/library/core/src/macros/mod.rs b/library/core/src/macros/mod.rs index 9a91ff82acd..771c2d31b60 100644 --- a/library/core/src/macros/mod.rs +++ b/library/core/src/macros/mod.rs @@ -12,54 +12,6 @@ macro_rules! panic { }; } -/// Helper macro for panicking in a `const fn`. -/// Invoke as: -/// ```rust,ignore (just an example) -/// core::macros::const_panic!("boring message", "flavored message {a} {b:?}", a: u32 = foo.len(), b: Something = bar); -/// ``` -/// where the first message will be printed in const-eval, -/// and the second message will be printed at runtime. -// All uses of this macro are FIXME(const-hack). -#[unstable(feature = "panic_internals", issue = "none")] -#[doc(hidden)] -pub macro const_panic { - ($const_msg:literal, $runtime_msg:literal, $($arg:ident : $ty:ty = $val:expr),* $(,)?) => {{ - #[inline] - #[track_caller] - fn runtime($($arg: $ty),*) -> ! { - $crate::panic!($runtime_msg); - } - - #[inline] - #[track_caller] - const fn compiletime($(_: $ty),*) -> ! { - $crate::panic!($const_msg); - } - - // Wrap call to `const_eval_select` in a function so that we can - // add the `rustc_allow_const_fn_unstable`. This is okay to do - // because both variants will panic, just with different messages. - #[rustc_allow_const_fn_unstable(const_eval_select)] - #[inline(always)] - #[track_caller] - #[cfg_attr(bootstrap, rustc_const_stable(feature = "const_panic", since = "CURRENT_RUSTC_VERSION"))] - const fn do_panic($($arg: $ty),*) -> ! { - $crate::intrinsics::const_eval_select(($($arg),* ,), compiletime, runtime) - } - - do_panic($($val),*) - }}, - // We support leaving away the `val` expressions for *all* arguments - // (but not for *some* arguments, that's too tricky). - ($const_msg:literal, $runtime_msg:literal, $($arg:ident : $ty:ty),* $(,)?) => { - $crate::macros::const_panic!( - $const_msg, - $runtime_msg, - $($arg: $ty = $arg),* - ) - }, -} - /// Asserts that two expressions are equal to each other (using [`PartialEq`]). /// /// Assertions are always checked in both debug and release builds, and cannot @@ -244,19 +196,6 @@ macro_rules! assert_ne { }, } -/// A version of `assert` that prints a non-formatting message in const contexts. -/// -/// See [`const_panic!`]. -#[unstable(feature = "panic_internals", issue = "none")] -#[doc(hidden)] -pub macro const_assert { - ($condition: expr, $const_msg:literal, $runtime_msg:literal, $($arg:tt)*) => {{ - if !$crate::intrinsics::likely($condition) { - $crate::macros::const_panic!($const_msg, $runtime_msg, $($arg)*) - } - }} -} - /// A macro for defining `#[cfg]` match-like statements. /// /// It is similar to the `if/elif` C preprocessor macro by allowing definition of a cascade of diff --git a/library/core/src/num/f128.rs b/library/core/src/num/f128.rs index 7709e7de01b..0484611958d 100644 --- a/library/core/src/num/f128.rs +++ b/library/core/src/num/f128.rs @@ -14,9 +14,9 @@ use crate::convert::FloatToInt; #[cfg(not(test))] use crate::intrinsics; -use crate::macros::const_assert; use crate::mem; use crate::num::FpCategory; +use crate::panic::const_assert; /// Basic mathematical constants. #[unstable(feature = "f128", issue = "116909")] diff --git a/library/core/src/num/f16.rs b/library/core/src/num/f16.rs index eb0225c58b8..898caf835bf 100644 --- a/library/core/src/num/f16.rs +++ b/library/core/src/num/f16.rs @@ -14,9 +14,9 @@ use crate::convert::FloatToInt; #[cfg(not(test))] use crate::intrinsics; -use crate::macros::const_assert; use crate::mem; use crate::num::FpCategory; +use crate::panic::const_assert; /// Basic mathematical constants. #[unstable(feature = "f16", issue = "116909")] diff --git a/library/core/src/num/f32.rs b/library/core/src/num/f32.rs index 686a6c50927..20ece883da6 100644 --- a/library/core/src/num/f32.rs +++ b/library/core/src/num/f32.rs @@ -14,9 +14,9 @@ use crate::convert::FloatToInt; #[cfg(not(test))] use crate::intrinsics; -use crate::macros::const_assert; use crate::mem; use crate::num::FpCategory; +use crate::panic::const_assert; /// The radix or base of the internal representation of `f32`. /// Use [`f32::RADIX`] instead. diff --git a/library/core/src/num/f64.rs b/library/core/src/num/f64.rs index 798cb4b1b5c..5640e71788b 100644 --- a/library/core/src/num/f64.rs +++ b/library/core/src/num/f64.rs @@ -14,9 +14,9 @@ use crate::convert::FloatToInt; #[cfg(not(test))] use crate::intrinsics; -use crate::macros::const_assert; use crate::mem; use crate::num::FpCategory; +use crate::panic::const_assert; /// The radix or base of the internal representation of `f64`. /// Use [`f64::RADIX`] instead. diff --git a/library/core/src/num/mod.rs b/library/core/src/num/mod.rs index f4930ca5c7d..5a69dc0c724 100644 --- a/library/core/src/num/mod.rs +++ b/library/core/src/num/mod.rs @@ -2,7 +2,7 @@ #![stable(feature = "rust1", since = "1.0.0")] -use crate::macros::const_panic; +use crate::panic::const_panic; use crate::str::FromStr; use crate::ub_checks::assert_unsafe_precondition; use crate::{ascii, intrinsics, mem}; diff --git a/library/core/src/panic.rs b/library/core/src/panic.rs index c95a000561c..f8f3962ce55 100644 --- a/library/core/src/panic.rs +++ b/library/core/src/panic.rs @@ -189,3 +189,59 @@ fn as_str(&mut self) -> Option<&str> { None } } + +/// Helper macro for panicking in a `const fn`. +/// Invoke as: +/// ```rust,ignore (just an example) +/// core::macros::const_panic!("boring message", "flavored message {a} {b:?}", a: u32 = foo.len(), b: Something = bar); +/// ``` +/// where the first message will be printed in const-eval, +/// and the second message will be printed at runtime. +// All uses of this macro are FIXME(const-hack). +#[unstable(feature = "panic_internals", issue = "none")] +#[doc(hidden)] +pub macro const_panic { + ($const_msg:literal, $runtime_msg:literal, $($arg:ident : $ty:ty = $val:expr),* $(,)?) => {{ + // Wrap call to `const_eval_select` in a function so that we can + // add the `rustc_allow_const_fn_unstable`. This is okay to do + // because both variants will panic, just with different messages. + #[rustc_allow_const_fn_unstable(const_eval_select)] + #[inline(always)] + #[track_caller] + #[cfg_attr(bootstrap, rustc_const_stable(feature = "const_panic", since = "CURRENT_RUSTC_VERSION"))] + const fn do_panic($($arg: $ty),*) -> ! { + $crate::intrinsics::const_eval_select!( + @capture { $($arg: $ty),* } -> !: + if const #[track_caller] { + $crate::panic!($const_msg) + } else #[track_caller] { + $crate::panic!($runtime_msg) + } + ) + } + + do_panic($($val),*) + }}, + // We support leaving away the `val` expressions for *all* arguments + // (but not for *some* arguments, that's too tricky). + ($const_msg:literal, $runtime_msg:literal, $($arg:ident : $ty:ty),* $(,)?) => { + $crate::panic::const_panic!( + $const_msg, + $runtime_msg, + $($arg: $ty = $arg),* + ) + }, +} + +/// A version of `assert` that prints a non-formatting message in const contexts. +/// +/// See [`const_panic!`]. +#[unstable(feature = "panic_internals", issue = "none")] +#[doc(hidden)] +pub macro const_assert { + ($condition: expr, $const_msg:literal, $runtime_msg:literal, $($arg:tt)*) => {{ + if !$crate::intrinsics::likely($condition) { + $crate::panic::const_panic!($const_msg, $runtime_msg, $($arg)*) + } + }} +} diff --git a/library/core/src/panicking.rs b/library/core/src/panicking.rs index 9071d6719a3..f603eb2971f 100644 --- a/library/core/src/panicking.rs +++ b/library/core/src/panicking.rs @@ -29,6 +29,7 @@ )] use crate::fmt; +use crate::intrinsics::const_eval_select; use crate::panic::{Location, PanicInfo}; #[cfg(feature = "panic_immediate_abort")] @@ -89,40 +90,35 @@ pub const fn panic_fmt(fmt: fmt::Arguments<'_>) -> ! { #[cfg_attr(not(bootstrap), rustc_const_stable_indirect)] // must follow stable const rules since it is exposed to stable #[rustc_allow_const_fn_unstable(const_eval_select)] pub const fn panic_nounwind_fmt(fmt: fmt::Arguments<'_>, force_no_backtrace: bool) -> ! { - #[inline] // this should always be inlined into `panic_nounwind_fmt` - #[track_caller] - fn runtime(fmt: fmt::Arguments<'_>, force_no_backtrace: bool) -> ! { - if cfg!(feature = "panic_immediate_abort") { - super::intrinsics::abort() + const_eval_select!( + @capture { fmt: fmt::Arguments<'_>, force_no_backtrace: bool } -> !: + if const #[track_caller] { + // We don't unwind anyway at compile-time so we can call the regular `panic_fmt`. + panic_fmt(fmt) + } else #[track_caller] { + if cfg!(feature = "panic_immediate_abort") { + super::intrinsics::abort() + } + + // NOTE This function never crosses the FFI boundary; it's a Rust-to-Rust call + // that gets resolved to the `#[panic_handler]` function. + extern "Rust" { + #[lang = "panic_impl"] + fn panic_impl(pi: &PanicInfo<'_>) -> !; + } + + // PanicInfo with the `can_unwind` flag set to false forces an abort. + let pi = PanicInfo::new( + &fmt, + Location::caller(), + /* can_unwind */ false, + force_no_backtrace, + ); + + // SAFETY: `panic_impl` is defined in safe Rust code and thus is safe to call. + unsafe { panic_impl(&pi) } } - - // NOTE This function never crosses the FFI boundary; it's a Rust-to-Rust call - // that gets resolved to the `#[panic_handler]` function. - extern "Rust" { - #[lang = "panic_impl"] - fn panic_impl(pi: &PanicInfo<'_>) -> !; - } - - // PanicInfo with the `can_unwind` flag set to false forces an abort. - let pi = PanicInfo::new( - &fmt, - Location::caller(), - /* can_unwind */ false, - force_no_backtrace, - ); - - // SAFETY: `panic_impl` is defined in safe Rust code and thus is safe to call. - unsafe { panic_impl(&pi) } - } - - #[inline] - #[track_caller] - const fn comptime(fmt: fmt::Arguments<'_>, _force_no_backtrace: bool) -> ! { - // We don't unwind anyway at compile-time so we can call the regular `panic_fmt`. - panic_fmt(fmt); - } - - super::intrinsics::const_eval_select((fmt, force_no_backtrace), comptime, runtime); + ) } // Next we define a bunch of higher-level wrappers that all bottom out in the two core functions diff --git a/library/core/src/ptr/const_ptr.rs b/library/core/src/ptr/const_ptr.rs index a4e8e373e04..2d7507e2d53 100644 --- a/library/core/src/ptr/const_ptr.rs +++ b/library/core/src/ptr/const_ptr.rs @@ -33,26 +33,23 @@ impl *const T { #[rustc_diagnostic_item = "ptr_const_is_null"] #[inline] pub const fn is_null(self) -> bool { - #[inline] - fn runtime_impl(ptr: *const u8) -> bool { - ptr.addr() == 0 - } - - #[inline] - #[rustc_const_unstable(feature = "const_ptr_is_null", issue = "74939")] - const fn const_impl(ptr: *const u8) -> bool { - match (ptr).guaranteed_eq(null_mut()) { - Some(res) => res, - // To remain maximally convervative, we stop execution when we don't - // know whether the pointer is null or not. - // We can *not* return `false` here, that would be unsound in `NonNull::new`! - None => panic!("null-ness of this pointer cannot be determined in const context"), - } - } - // Compare via a cast to a thin pointer, so fat pointers are only // considering their "data" part for null-ness. - const_eval_select((self as *const u8,), const_impl, runtime_impl) + let ptr = self as *const u8; + const_eval_select!( + @capture { ptr: *const u8 } -> bool: + if const #[rustc_const_unstable(feature = "const_ptr_is_null", issue = "74939")] { + match (ptr).guaranteed_eq(null_mut()) { + Some(res) => res, + // To remain maximally convervative, we stop execution when we don't + // know whether the pointer is null or not. + // We can *not* return `false` here, that would be unsound in `NonNull::new`! + None => panic!("null-ness of this pointer cannot be determined in const context"), + } + } else { + ptr.addr() == 0 + } + ) } /// Casts to a pointer of another type. @@ -410,22 +407,21 @@ pub const fn to_raw_parts(self) -> (*const (), ::Metadata) #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_offset_nowrap(this: *const (), count: isize, size: usize) -> bool { - #[inline] - fn runtime(this: *const (), count: isize, size: usize) -> bool { - // We know `size <= isize::MAX` so the `as` cast here is not lossy. - let Some(byte_offset) = count.checked_mul(size as isize) else { - return false; - }; - let (_, overflow) = this.addr().overflowing_add_signed(byte_offset); - !overflow - } - - const fn comptime(_: *const (), _: isize, _: usize) -> bool { - true - } - // We can use const_eval_select here because this is only for UB checks. - intrinsics::const_eval_select((this, count, size), comptime, runtime) + const_eval_select!( + @capture { this: *const (), count: isize, size: usize } -> bool: + if const { + true + } else { + // `size` is the size of a Rust type, so we know that + // `size <= isize::MAX` and thus `as` cast here is not lossy. + let Some(byte_offset) = count.checked_mul(size as isize) else { + return false; + }; + let (_, overflow) = this.addr().overflowing_add_signed(byte_offset); + !overflow + } + ) } ub_checks::assert_unsafe_precondition!( @@ -763,14 +759,14 @@ pub fn mask(self, mask: usize) -> *const T { { #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_ptr_ge(this: *const (), origin: *const ()) -> bool { - fn runtime(this: *const (), origin: *const ()) -> bool { - this >= origin - } - const fn comptime(_: *const (), _: *const ()) -> bool { - true - } - - intrinsics::const_eval_select((this, origin), comptime, runtime) + const_eval_select!( + @capture { this: *const (), origin: *const () } -> bool: + if const { + true + } else { + this >= origin + } + ) } ub_checks::assert_unsafe_precondition!( @@ -924,20 +920,18 @@ pub const fn guaranteed_ne(self, other: *const T) -> Option #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_add_nowrap(this: *const (), count: usize, size: usize) -> bool { - #[inline] - fn runtime(this: *const (), count: usize, size: usize) -> bool { - let Some(byte_offset) = count.checked_mul(size) else { - return false; - }; - let (_, overflow) = this.addr().overflowing_add(byte_offset); - byte_offset <= (isize::MAX as usize) && !overflow - } - - const fn comptime(_: *const (), _: usize, _: usize) -> bool { - true - } - - intrinsics::const_eval_select((this, count, size), comptime, runtime) + const_eval_select!( + @capture { this: *const (), count: usize, size: usize } -> bool: + if const { + true + } else { + let Some(byte_offset) = count.checked_mul(size) else { + return false; + }; + let (_, overflow) = this.addr().overflowing_add(byte_offset); + byte_offset <= (isize::MAX as usize) && !overflow + } + ) } #[cfg(debug_assertions)] // Expensive, and doesn't catch much in the wild. @@ -1033,19 +1027,17 @@ const fn comptime(_: *const (), _: usize, _: usize) -> bool { #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_sub_nowrap(this: *const (), count: usize, size: usize) -> bool { - #[inline] - fn runtime(this: *const (), count: usize, size: usize) -> bool { - let Some(byte_offset) = count.checked_mul(size) else { - return false; - }; - byte_offset <= (isize::MAX as usize) && this.addr() >= byte_offset - } - - const fn comptime(_: *const (), _: usize, _: usize) -> bool { - true - } - - intrinsics::const_eval_select((this, count, size), comptime, runtime) + const_eval_select!( + @capture { this: *const (), count: usize, size: usize } -> bool: + if const { + true + } else { + let Some(byte_offset) = count.checked_mul(size) else { + return false; + }; + byte_offset <= (isize::MAX as usize) && this.addr() >= byte_offset + } + ) } #[cfg(debug_assertions)] // Expensive, and doesn't catch much in the wild. diff --git a/library/core/src/ptr/mut_ptr.rs b/library/core/src/ptr/mut_ptr.rs index 0d94a7f491c..344ba46a50e 100644 --- a/library/core/src/ptr/mut_ptr.rs +++ b/library/core/src/ptr/mut_ptr.rs @@ -1,5 +1,6 @@ use super::*; use crate::cmp::Ordering::{Equal, Greater, Less}; +use crate::intrinsics::const_eval_select; use crate::mem::SizedTypeProperties; use crate::slice::{self, SliceIndex}; @@ -404,23 +405,21 @@ pub const fn to_raw_parts(self) -> (*mut (), ::Metadata) { #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_offset_nowrap(this: *const (), count: isize, size: usize) -> bool { - #[inline] - fn runtime(this: *const (), count: isize, size: usize) -> bool { - // `size` is the size of a Rust type, so we know that - // `size <= isize::MAX` and thus `as` cast here is not lossy. - let Some(byte_offset) = count.checked_mul(size as isize) else { - return false; - }; - let (_, overflow) = this.addr().overflowing_add_signed(byte_offset); - !overflow - } - - const fn comptime(_: *const (), _: isize, _: usize) -> bool { - true - } - // We can use const_eval_select here because this is only for UB checks. - intrinsics::const_eval_select((this, count, size), comptime, runtime) + const_eval_select!( + @capture { this: *const (), count: isize, size: usize } -> bool: + if const { + true + } else { + // `size` is the size of a Rust type, so we know that + // `size <= isize::MAX` and thus `as` cast here is not lossy. + let Some(byte_offset) = count.checked_mul(size as isize) else { + return false; + }; + let (_, overflow) = this.addr().overflowing_add_signed(byte_offset); + !overflow + } + ) } ub_checks::assert_unsafe_precondition!( @@ -1002,20 +1001,18 @@ pub const fn guaranteed_ne(self, other: *mut T) -> Option #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_add_nowrap(this: *const (), count: usize, size: usize) -> bool { - #[inline] - fn runtime(this: *const (), count: usize, size: usize) -> bool { - let Some(byte_offset) = count.checked_mul(size) else { - return false; - }; - let (_, overflow) = this.addr().overflowing_add(byte_offset); - byte_offset <= (isize::MAX as usize) && !overflow - } - - const fn comptime(_: *const (), _: usize, _: usize) -> bool { - true - } - - intrinsics::const_eval_select((this, count, size), comptime, runtime) + const_eval_select!( + @capture { this: *const (), count: usize, size: usize } -> bool: + if const { + true + } else { + let Some(byte_offset) = count.checked_mul(size) else { + return false; + }; + let (_, overflow) = this.addr().overflowing_add(byte_offset); + byte_offset <= (isize::MAX as usize) && !overflow + } + ) } #[cfg(debug_assertions)] // Expensive, and doesn't catch much in the wild. @@ -1111,19 +1108,17 @@ const fn comptime(_: *const (), _: usize, _: usize) -> bool { #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] const fn runtime_sub_nowrap(this: *const (), count: usize, size: usize) -> bool { - #[inline] - fn runtime(this: *const (), count: usize, size: usize) -> bool { - let Some(byte_offset) = count.checked_mul(size) else { - return false; - }; - byte_offset <= (isize::MAX as usize) && this.addr() >= byte_offset - } - - const fn comptime(_: *const (), _: usize, _: usize) -> bool { - true - } - - intrinsics::const_eval_select((this, count, size), comptime, runtime) + const_eval_select!( + @capture { this: *const (), count: usize, size: usize } -> bool: + if const { + true + } else { + let Some(byte_offset) = count.checked_mul(size) else { + return false; + }; + byte_offset <= (isize::MAX as usize) && this.addr() >= byte_offset + } + ) } #[cfg(debug_assertions)] // Expensive, and doesn't catch much in the wild. diff --git a/library/core/src/slice/ascii.rs b/library/core/src/slice/ascii.rs index 8dcd34929e1..58ba3a1573a 100644 --- a/library/core/src/slice/ascii.rs +++ b/library/core/src/slice/ascii.rs @@ -351,89 +351,87 @@ pub const fn is_ascii_simple(mut bytes: &[u8]) -> bool { const fn is_ascii(s: &[u8]) -> bool { // The runtime version behaves the same as the compiletime version, it's // just more optimized. - return const_eval_select((s,), compiletime, runtime); + const_eval_select!( + @capture { s: &[u8] } -> bool: + if const { + is_ascii_simple(s) + } else { + const USIZE_SIZE: usize = mem::size_of::(); - const fn compiletime(s: &[u8]) -> bool { - is_ascii_simple(s) - } + let len = s.len(); + let align_offset = s.as_ptr().align_offset(USIZE_SIZE); - #[inline] - fn runtime(s: &[u8]) -> bool { - const USIZE_SIZE: usize = mem::size_of::(); - - let len = s.len(); - let align_offset = s.as_ptr().align_offset(USIZE_SIZE); - - // If we wouldn't gain anything from the word-at-a-time implementation, fall - // back to a scalar loop. - // - // We also do this for architectures where `size_of::()` isn't - // sufficient alignment for `usize`, because it's a weird edge case. - if len < USIZE_SIZE || len < align_offset || USIZE_SIZE < mem::align_of::() { - return is_ascii_simple(s); - } - - // We always read the first word unaligned, which means `align_offset` is - // 0, we'd read the same value again for the aligned read. - let offset_to_aligned = if align_offset == 0 { USIZE_SIZE } else { align_offset }; - - let start = s.as_ptr(); - // SAFETY: We verify `len < USIZE_SIZE` above. - let first_word = unsafe { (start as *const usize).read_unaligned() }; - - if contains_nonascii(first_word) { - return false; - } - // We checked this above, somewhat implicitly. Note that `offset_to_aligned` - // is either `align_offset` or `USIZE_SIZE`, both of are explicitly checked - // above. - debug_assert!(offset_to_aligned <= len); - - // SAFETY: word_ptr is the (properly aligned) usize ptr we use to read the - // middle chunk of the slice. - let mut word_ptr = unsafe { start.add(offset_to_aligned) as *const usize }; - - // `byte_pos` is the byte index of `word_ptr`, used for loop end checks. - let mut byte_pos = offset_to_aligned; - - // Paranoia check about alignment, since we're about to do a bunch of - // unaligned loads. In practice this should be impossible barring a bug in - // `align_offset` though. - // While this method is allowed to spuriously fail in CTFE, if it doesn't - // have alignment information it should have given a `usize::MAX` for - // `align_offset` earlier, sending things through the scalar path instead of - // this one, so this check should pass if it's reachable. - debug_assert!(word_ptr.is_aligned_to(mem::align_of::())); - - // Read subsequent words until the last aligned word, excluding the last - // aligned word by itself to be done in tail check later, to ensure that - // tail is always one `usize` at most to extra branch `byte_pos == len`. - while byte_pos < len - USIZE_SIZE { - // Sanity check that the read is in bounds - debug_assert!(byte_pos + USIZE_SIZE <= len); - // And that our assumptions about `byte_pos` hold. - debug_assert!(word_ptr.cast::() == start.wrapping_add(byte_pos)); - - // SAFETY: We know `word_ptr` is properly aligned (because of - // `align_offset`), and we know that we have enough bytes between `word_ptr` and the end - let word = unsafe { word_ptr.read() }; - if contains_nonascii(word) { - return false; + // If we wouldn't gain anything from the word-at-a-time implementation, fall + // back to a scalar loop. + // + // We also do this for architectures where `size_of::()` isn't + // sufficient alignment for `usize`, because it's a weird edge case. + if len < USIZE_SIZE || len < align_offset || USIZE_SIZE < mem::align_of::() { + return is_ascii_simple(s); } - byte_pos += USIZE_SIZE; - // SAFETY: We know that `byte_pos <= len - USIZE_SIZE`, which means that - // after this `add`, `word_ptr` will be at most one-past-the-end. - word_ptr = unsafe { word_ptr.add(1) }; + // We always read the first word unaligned, which means `align_offset` is + // 0, we'd read the same value again for the aligned read. + let offset_to_aligned = if align_offset == 0 { USIZE_SIZE } else { align_offset }; + + let start = s.as_ptr(); + // SAFETY: We verify `len < USIZE_SIZE` above. + let first_word = unsafe { (start as *const usize).read_unaligned() }; + + if contains_nonascii(first_word) { + return false; + } + // We checked this above, somewhat implicitly. Note that `offset_to_aligned` + // is either `align_offset` or `USIZE_SIZE`, both of are explicitly checked + // above. + debug_assert!(offset_to_aligned <= len); + + // SAFETY: word_ptr is the (properly aligned) usize ptr we use to read the + // middle chunk of the slice. + let mut word_ptr = unsafe { start.add(offset_to_aligned) as *const usize }; + + // `byte_pos` is the byte index of `word_ptr`, used for loop end checks. + let mut byte_pos = offset_to_aligned; + + // Paranoia check about alignment, since we're about to do a bunch of + // unaligned loads. In practice this should be impossible barring a bug in + // `align_offset` though. + // While this method is allowed to spuriously fail in CTFE, if it doesn't + // have alignment information it should have given a `usize::MAX` for + // `align_offset` earlier, sending things through the scalar path instead of + // this one, so this check should pass if it's reachable. + debug_assert!(word_ptr.is_aligned_to(mem::align_of::())); + + // Read subsequent words until the last aligned word, excluding the last + // aligned word by itself to be done in tail check later, to ensure that + // tail is always one `usize` at most to extra branch `byte_pos == len`. + while byte_pos < len - USIZE_SIZE { + // Sanity check that the read is in bounds + debug_assert!(byte_pos + USIZE_SIZE <= len); + // And that our assumptions about `byte_pos` hold. + debug_assert!(word_ptr.cast::() == start.wrapping_add(byte_pos)); + + // SAFETY: We know `word_ptr` is properly aligned (because of + // `align_offset`), and we know that we have enough bytes between `word_ptr` and the end + let word = unsafe { word_ptr.read() }; + if contains_nonascii(word) { + return false; + } + + byte_pos += USIZE_SIZE; + // SAFETY: We know that `byte_pos <= len - USIZE_SIZE`, which means that + // after this `add`, `word_ptr` will be at most one-past-the-end. + word_ptr = unsafe { word_ptr.add(1) }; + } + + // Sanity check to ensure there really is only one `usize` left. This should + // be guaranteed by our loop condition. + debug_assert!(byte_pos <= len && len - byte_pos <= USIZE_SIZE); + + // SAFETY: This relies on `len >= USIZE_SIZE`, which we check at the start. + let last_word = unsafe { (start.add(len - USIZE_SIZE) as *const usize).read_unaligned() }; + + !contains_nonascii(last_word) } - - // Sanity check to ensure there really is only one `usize` left. This should - // be guaranteed by our loop condition. - debug_assert!(byte_pos <= len && len - byte_pos <= USIZE_SIZE); - - // SAFETY: This relies on `len >= USIZE_SIZE`, which we check at the start. - let last_word = unsafe { (start.add(len - USIZE_SIZE) as *const usize).read_unaligned() }; - - !contains_nonascii(last_word) - } + ) } diff --git a/library/core/src/slice/index.rs b/library/core/src/slice/index.rs index ebb4bdb1449..aafa19c0dd3 100644 --- a/library/core/src/slice/index.rs +++ b/library/core/src/slice/index.rs @@ -1,6 +1,6 @@ //! Indexing implementations for `[T]`. -use crate::macros::const_panic; +use crate::panic::const_panic; use crate::ub_checks::assert_unsafe_precondition; use crate::{ops, range}; diff --git a/library/core/src/slice/memchr.rs b/library/core/src/slice/memchr.rs index b7c4a1f6f08..339adad1b17 100644 --- a/library/core/src/slice/memchr.rs +++ b/library/core/src/slice/memchr.rs @@ -56,61 +56,59 @@ const fn memchr_naive(x: u8, text: &[u8]) -> Option { const fn memchr_aligned(x: u8, text: &[u8]) -> Option { // The runtime version behaves the same as the compiletime version, it's // just more optimized. - return const_eval_select((x, text), compiletime, runtime); + const_eval_select!( + @capture { x: u8, text: &[u8] } -> Option: + if const { + memchr_naive(x, text) + } else { + // Scan for a single byte value by reading two `usize` words at a time. + // + // Split `text` in three parts + // - unaligned initial part, before the first word aligned address in text + // - body, scan by 2 words at a time + // - the last remaining part, < 2 word size - const fn compiletime(x: u8, text: &[u8]) -> Option { - memchr_naive(x, text) - } + // search up to an aligned boundary + let len = text.len(); + let ptr = text.as_ptr(); + let mut offset = ptr.align_offset(USIZE_BYTES); - #[inline] - fn runtime(x: u8, text: &[u8]) -> Option { - // Scan for a single byte value by reading two `usize` words at a time. - // - // Split `text` in three parts - // - unaligned initial part, before the first word aligned address in text - // - body, scan by 2 words at a time - // - the last remaining part, < 2 word size - - // search up to an aligned boundary - let len = text.len(); - let ptr = text.as_ptr(); - let mut offset = ptr.align_offset(USIZE_BYTES); - - if offset > 0 { - offset = offset.min(len); - let slice = &text[..offset]; - if let Some(index) = memchr_naive(x, slice) { - return Some(index); - } - } - - // search the body of the text - let repeated_x = usize::repeat_u8(x); - while offset <= len - 2 * USIZE_BYTES { - // SAFETY: the while's predicate guarantees a distance of at least 2 * usize_bytes - // between the offset and the end of the slice. - unsafe { - let u = *(ptr.add(offset) as *const usize); - let v = *(ptr.add(offset + USIZE_BYTES) as *const usize); - - // break if there is a matching byte - let zu = contains_zero_byte(u ^ repeated_x); - let zv = contains_zero_byte(v ^ repeated_x); - if zu || zv { - break; + if offset > 0 { + offset = offset.min(len); + let slice = &text[..offset]; + if let Some(index) = memchr_naive(x, slice) { + return Some(index); } } - offset += USIZE_BYTES * 2; - } - // Find the byte after the point the body loop stopped. - // FIXME(const-hack): Use `?` instead. - // FIXME(const-hack, fee1-dead): use range slicing - let slice = - // SAFETY: offset is within bounds - unsafe { super::from_raw_parts(text.as_ptr().add(offset), text.len() - offset) }; - if let Some(i) = memchr_naive(x, slice) { Some(offset + i) } else { None } - } + // search the body of the text + let repeated_x = usize::repeat_u8(x); + while offset <= len - 2 * USIZE_BYTES { + // SAFETY: the while's predicate guarantees a distance of at least 2 * usize_bytes + // between the offset and the end of the slice. + unsafe { + let u = *(ptr.add(offset) as *const usize); + let v = *(ptr.add(offset + USIZE_BYTES) as *const usize); + + // break if there is a matching byte + let zu = contains_zero_byte(u ^ repeated_x); + let zv = contains_zero_byte(v ^ repeated_x); + if zu || zv { + break; + } + } + offset += USIZE_BYTES * 2; + } + + // Find the byte after the point the body loop stopped. + // FIXME(const-hack): Use `?` instead. + // FIXME(const-hack, fee1-dead): use range slicing + let slice = + // SAFETY: offset is within bounds + unsafe { super::from_raw_parts(text.as_ptr().add(offset), text.len() - offset) }; + if let Some(i) = memchr_naive(x, slice) { Some(offset + i) } else { None } + } + ) } /// Returns the last index matching the byte `x` in `text`. diff --git a/library/core/src/str/validations.rs b/library/core/src/str/validations.rs index 6095b589e18..0f724dd9613 100644 --- a/library/core/src/str/validations.rs +++ b/library/core/src/str/validations.rs @@ -132,19 +132,16 @@ pub(super) const fn run_utf8_validation(v: &[u8]) -> Result<(), Utf8Error> { let ascii_block_size = 2 * USIZE_BYTES; let blocks_end = if len >= ascii_block_size { len - ascii_block_size + 1 } else { 0 }; - let align = { - const fn compiletime(_v: &[u8]) -> usize { + // Below, we safely fall back to a slower codepath if the offset is `usize::MAX`, + // so the end-to-end behavior is the same at compiletime and runtime. + let align = const_eval_select!( + @capture { v: &[u8] } -> usize: + if const { usize::MAX - } - - fn runtime(v: &[u8]) -> usize { + } else { v.as_ptr().align_offset(USIZE_BYTES) } - - // Below, we safely fall back to a slower codepath if the offset is `usize::MAX`, - // so the end-to-end behavior is the same at compiletime and runtime. - const_eval_select((v,), compiletime, runtime) - }; + ); while index < len { let old_offset = index; diff --git a/library/core/src/ub_checks.rs b/library/core/src/ub_checks.rs index dd1454f401e..8fcbda141da 100644 --- a/library/core/src/ub_checks.rs +++ b/library/core/src/ub_checks.rs @@ -95,20 +95,18 @@ const fn precondition_check($($name:$ty),*) { #[inline] #[rustc_allow_const_fn_unstable(const_eval_select)] pub(crate) const fn check_language_ub() -> bool { - #[inline] - fn runtime() -> bool { - // Disable UB checks in Miri. - !cfg!(miri) - } - - #[inline] - const fn comptime() -> bool { - // Always disable UB checks. - false - } - // Only used for UB checks so we may const_eval_select. - intrinsics::ub_checks() && const_eval_select((), comptime, runtime) + intrinsics::ub_checks() + && const_eval_select!( + @capture { } -> bool: + if const { + // Always disable UB checks. + false + } else { + // Disable UB checks in Miri. + !cfg!(miri) + } + ) } /// Checks whether `ptr` is properly aligned with respect to the given alignment, and @@ -120,19 +118,15 @@ const fn comptime() -> bool { #[inline] #[rustc_const_unstable(feature = "const_ub_checks", issue = "none")] pub(crate) const fn is_aligned_and_not_null(ptr: *const (), align: usize, is_zst: bool) -> bool { - #[inline] - fn runtime(ptr: *const (), align: usize, is_zst: bool) -> bool { - ptr.is_aligned_to(align) && (is_zst || !ptr.is_null()) - } - - #[inline] - #[rustc_const_unstable(feature = "const_ub_checks", issue = "none")] - const fn comptime(ptr: *const (), _align: usize, is_zst: bool) -> bool { - is_zst || !ptr.is_null() - } - // This is just for safety checks so we can const_eval_select. - const_eval_select((ptr, align, is_zst), comptime, runtime) + const_eval_select!( + @capture { ptr: *const (), align: usize, is_zst: bool } -> bool: + if const #[rustc_const_unstable(feature = "const_ub_checks", issue = "none")] { + is_zst || !ptr.is_null() + } else { + ptr.is_aligned_to(align) && (is_zst || !ptr.is_null()) + } + ) } #[inline] @@ -154,26 +148,23 @@ pub(crate) const fn is_nonoverlapping( size: usize, count: usize, ) -> bool { - #[inline] - fn runtime(src: *const (), dst: *const (), size: usize, count: usize) -> bool { - let src_usize = src.addr(); - let dst_usize = dst.addr(); - let Some(size) = size.checked_mul(count) else { - crate::panicking::panic_nounwind( - "is_nonoverlapping: `size_of::() * count` overflows a usize", - ) - }; - let diff = src_usize.abs_diff(dst_usize); - // If the absolute distance between the ptrs is at least as big as the size of the buffer, - // they do not overlap. - diff >= size - } - - #[inline] - const fn comptime(_: *const (), _: *const (), _: usize, _: usize) -> bool { - true - } - // This is just for safety checks so we can const_eval_select. - const_eval_select((src, dst, size, count), comptime, runtime) + const_eval_select!( + @capture { src: *const (), dst: *const (), size: usize, count: usize } -> bool: + if const { + true + } else { + let src_usize = src.addr(); + let dst_usize = dst.addr(); + let Some(size) = size.checked_mul(count) else { + crate::panicking::panic_nounwind( + "is_nonoverlapping: `size_of::() * count` overflows a usize", + ) + }; + let diff = src_usize.abs_diff(dst_usize); + // If the absolute distance between the ptrs is at least as big as the size of the buffer, + // they do not overlap. + diff >= size + } + ) } diff --git a/tests/ui/consts/const-ptr-is-null.stderr b/tests/ui/consts/const-ptr-is-null.stderr index 20e44a1401f..5fd35142818 100644 --- a/tests/ui/consts/const-ptr-is-null.stderr +++ b/tests/ui/consts/const-ptr-is-null.stderr @@ -3,7 +3,7 @@ error[E0080]: evaluation of constant value failed | = note: the evaluated program panicked at 'null-ness of this pointer cannot be determined in const context', $SRC_DIR/core/src/ptr/const_ptr.rs:LL:COL | -note: inside `std::ptr::const_ptr::::is_null::const_impl` +note: inside `std::ptr::const_ptr::::is_null::compiletime` --> $SRC_DIR/core/src/ptr/const_ptr.rs:LL:COL note: inside `std::ptr::const_ptr::::is_null` --> $SRC_DIR/core/src/ptr/const_ptr.rs:LL:COL @@ -12,7 +12,7 @@ note: inside `MAYBE_NULL` | LL | assert!(!ptr.wrapping_sub(512).is_null()); | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - = note: this error originates in the macro `$crate::panic::panic_2021` which comes from the expansion of the macro `panic` (in Nightly builds, run with -Z macro-backtrace for more info) + = note: this error originates in the macro `$crate::panic::panic_2021` which comes from the expansion of the macro `const_eval_select` (in Nightly builds, run with -Z macro-backtrace for more info) error: aborting due to 1 previous error