Implement RFC 3349, mixed utf8 literals.

Specifically:
- Allow unicode chars in b"" and br"" literals. This is done by changing
  `Mode::allow_unicode_chars` to succeed on `ByteStr` and `RawByteStr`.
- Allow unicode escapes in b"" literals. This is done by changing
  `Mode::allow_unicode_escapes` to succeed on `ByteStr`.

Byte string literals can already have high bytes (`\x80`..`\xff`).
Because they now also support unicode chars, they can now be mixed utf8,
so we use `unescape_mixed`/`cook_mixed` instead of
`unescape_unicode`/`cook_unicode` to process them.

A new type `Rfc3349`, is used to implement the feature gating. Values of
that type are threaded through the unescaping code to track whether
rules from rfc3349 are required for unescaping to succeed.

Test changes:

- tests/ui/mixed-utf8-literals/basic.rs: new `check-pass` UI test with
  various literals exercising the new forms.

- tests/ui/attributes/key-value-non-ascii.rs: changed from a byte string
  literal to a byte literal; we just need some kind of problem with a
  literal to preserve the test's intent.

- tests/ui/parser/raw/raw-byte-string-literals.rs: moved the raw byte
  string literal with a non-ASCII char to `basic.rs`.

- tests/ui/parser/byte-string-literals.rs: similar.

- tests/ui/parser/issues/issue-23620-invalid-escapes.rs: moved one case
  fully to `basic.rs`, and one partially.

- tests/ui/parser/unicode-control-codepoints.rs: left the code
  unchanged, but the errors are now about mixed-utf8-literals being
  feature gated.

- tests/ui/suggestions/multibyte-escapes.rs: moved one case to
  `basic.rs`.

- compiler/rustc_lexer/src/unescape/tests.rs: various adjustments
  - two cases that previously failed now succeed
  - added some more cases for the newly supported syntax

I wasn't sure how to handle rust-analyzer in general, so I just allowed
mixed utf8 literals everywhere without complaint.

Author: nnethercote

compiler/rustc_ast/src/util/literal.rs

@@ -3,7 +3,7 @@
 use crate::ast::{self, LitKind, MetaItemLit, StrStyle};
 use crate::token::{self, Token};
 use rustc_lexer::unescape::{
-    byte_from_char, unescape_byte, unescape_char, unescape_mixed, unescape_unicode, MixedUnit, Mode,
+    unescape_byte, unescape_char, unescape_mixed, unescape_unicode, MixedUnit, Mode,
 };
 use rustc_span::symbol::{kw, sym, Symbol};
 use rustc_span::Span;
@@ -49,7 +49,8 @@ impl LitKind {
 
         // For byte/char/string literals, chars and escapes have already been
         // checked in the lexer (in `cook_lexer_literal`). So we can assume all
-        // chars and escapes are valid here.
+        // chars and escapes are valid here, and ignore `Rfc3349` return
+        // values.
         Ok(match kind {
             token::Bool => {
                 assert!(symbol.is_bool_lit());
@@ -84,7 +85,7 @@ impl LitKind {
                     // Force-inlining here is aggressive but the closure is
                     // called on every char in the string, so it can be hot in
                     // programs with many long strings containing escapes.
-                    unescape_unicode(
+                    _ = unescape_unicode(
                         s,
                         Mode::Str,
                         &mut #[inline(always)]
@@ -108,8 +109,11 @@ impl LitKind {
             token::ByteStr => {
                 let s = symbol.as_str();
                 let mut buf = Vec::with_capacity(s.len());
-                unescape_unicode(s, Mode::ByteStr, &mut |_, c| match c {
-                    Ok(c) => buf.push(byte_from_char(c)),
+                _ = unescape_mixed(s, Mode::ByteStr, &mut |_, c| match c {
+                    Ok(MixedUnit::Char(c)) => {
+                        buf.extend_from_slice(c.encode_utf8(&mut [0; 4]).as_bytes())
+                    }
+                    Ok(MixedUnit::HighByte(b)) => buf.push(b),
                     Err(err) => {
                         assert!(!err.is_fatal(), "failed to unescape string literal")
                     }
@@ -125,7 +129,7 @@ impl LitKind {
             token::CStr => {
                 let s = symbol.as_str();
                 let mut buf = Vec::with_capacity(s.len());
-                unescape_mixed(s, Mode::CStr, &mut |_span, c| match c {
+                _ = unescape_mixed(s, Mode::CStr, &mut |_span, c| match c {
                     Ok(MixedUnit::Char(c)) => {
                         buf.extend_from_slice(c.encode_utf8(&mut [0; 4]).as_bytes())
                     }

compiler/rustc_ast_passes/src/feature_gate.rs

@@ -508,6 +508,7 @@ pub fn check_crate(krate: &ast::Crate, sess: &Session, features: &Features) {
             }
         };
     }
+    gate_all!(mixed_utf8_literals, r#"mixed utf8 b"" and br"" literals are experimental"#);
     gate_all!(
         if_let_guard,
         "`if let` guards are experimental",

compiler/rustc_feature/src/unstable.rs

@@ -523,6 +523,8 @@ declare_features! (
     /// standard library until the soundness issues with specialization
     /// are fixed.
     (unstable, min_specialization, "1.7.0", Some(31844)),
+    /// Allows mixed utf8 b"" and br"" literals.
+    (unstable, mixed_utf8_literals, "CURRENT_RUSTC_VERSION", Some(116907)),
     /// Allows qualified paths in struct expressions, struct patterns and tuple struct patterns.
     (unstable, more_qualified_paths, "1.54.0", Some(86935)),
     /// Allows the `#[must_not_suspend]` attribute.

compiler/rustc_lexer/src/unescape.rs

@@ -85,7 +85,7 @@ impl EscapeError {
 ///
 /// Values are returned by invoking `callback`. For `Char` and `Byte` modes,
 /// the callback will be called exactly once.
-pub fn unescape_unicode<F>(src: &str, mode: Mode, callback: &mut F)
+pub fn unescape_unicode<F>(src: &str, mode: Mode, callback: &mut F) -> Rfc3349
 where
     F: FnMut(Range<usize>, Result<char, EscapeError>),
 {
@@ -94,16 +94,17 @@ where
             let mut chars = src.chars();
             let res = unescape_char_or_byte(&mut chars, mode);
             callback(0..(src.len() - chars.as_str().len()), res);
+            Rfc3349::Unused // rfc3349 not relevant for `Mode::{Char,Byte}`
         }
-        Str | ByteStr => unescape_non_raw_common(src, mode, callback),
+        Str => unescape_non_raw_common(src, mode, callback),
         RawStr | RawByteStr => check_raw_common(src, mode, callback),
         RawCStr => check_raw_common(src, mode, &mut |r, mut result| {
             if let Ok('\0') = result {
                 result = Err(EscapeError::NulInCStr);
             }
             callback(r, result)
         }),
-        CStr => unreachable!(),
+        ByteStr | CStr => unreachable!(),
     }
 }
 
@@ -142,18 +143,19 @@ impl From<u8> for MixedUnit {
 /// a sequence of escaped characters or errors.
 ///
 /// Values are returned by invoking `callback`.
-pub fn unescape_mixed<F>(src: &str, mode: Mode, callback: &mut F)
+pub fn unescape_mixed<F>(src: &str, mode: Mode, callback: &mut F) -> Rfc3349
 where
     F: FnMut(Range<usize>, Result<MixedUnit, EscapeError>),
 {
     match mode {
+        ByteStr => unescape_non_raw_common(src, mode, callback),
         CStr => unescape_non_raw_common(src, mode, &mut |r, mut result| {
             if let Ok(MixedUnit::Char('\0')) = result {
                 result = Err(EscapeError::NulInCStr);
             }
             callback(r, result)
         }),
-        Char | Byte | Str | RawStr | ByteStr | RawByteStr | RawCStr => unreachable!(),
+        Char | Byte | Str | RawStr | RawByteStr | RawCStr => unreachable!(),
     }
 }
 
@@ -169,6 +171,15 @@ pub fn unescape_byte(src: &str) -> Result<u8, EscapeError> {
     unescape_char_or_byte(&mut src.chars(), Byte).map(byte_from_char)
 }
 
+/// Used to indicate if rfc3349 (mixed-utf8-literals) was required for the
+/// literal to be valid. Once rfc3349 is stabilized this type can be removed.
+#[derive(Debug, PartialEq)]
+#[must_use]
+pub enum Rfc3349 {
+    Used,
+    Unused,
+}
+
 /// What kind of literal do we parse.
 #[derive(Debug, Clone, Copy, PartialEq)]
 pub enum Mode {
@@ -205,17 +216,25 @@ impl Mode {
 
     /// Are unicode (non-ASCII) chars allowed?
     #[inline]
-    fn allow_unicode_chars(self) -> bool {
+    fn allow_unicode_chars(self, rfc3349: &mut Rfc3349) -> bool {
         match self {
-            Byte | ByteStr | RawByteStr => false,
+            Byte => false,
+            ByteStr | RawByteStr => {
+                *rfc3349 = Rfc3349::Used;
+                true
+            }
             Char | Str | RawStr | CStr | RawCStr => true,
         }
     }
 
     /// Are unicode escapes (`\u`) allowed?
-    fn allow_unicode_escapes(self) -> bool {
+    fn allow_unicode_escapes(self, rfc3349: &mut Rfc3349) -> bool {
         match self {
-            Byte | ByteStr => false,
+            Byte => false,
+            ByteStr => {
+                *rfc3349 = Rfc3349::Used;
+                true
+            }
             Char | Str | CStr => true,
             RawByteStr | RawStr | RawCStr => unreachable!(),
         }
@@ -233,6 +252,7 @@ impl Mode {
 fn scan_escape<T: From<char> + From<u8>>(
     chars: &mut Chars<'_>,
     mode: Mode,
+    rfc3349: &mut Rfc3349,
 ) -> Result<T, EscapeError> {
     // Previous character was '\\', unescape what follows.
     let res: char = match chars.next().ok_or(EscapeError::LoneSlash)? {
@@ -262,13 +282,17 @@ fn scan_escape<T: From<char> + From<u8>>(
                 Ok(T::from(value as u8))
             };
         }
-        'u' => return scan_unicode(chars, mode.allow_unicode_escapes()).map(T::from),
+        'u' => return scan_unicode(chars, mode, rfc3349).map(T::from),
         _ => return Err(EscapeError::InvalidEscape),
     };
     Ok(T::from(res))
 }
 
-fn scan_unicode(chars: &mut Chars<'_>, allow_unicode_escapes: bool) -> Result<char, EscapeError> {
+fn scan_unicode(
+    chars: &mut Chars<'_>,
+    mode: Mode,
+    rfc3349: &mut Rfc3349,
+) -> Result<char, EscapeError> {
     // We've parsed '\u', now we have to parse '{..}'.
 
     if chars.next() != Some('{') {
@@ -296,7 +320,7 @@ fn scan_unicode(chars: &mut Chars<'_>, allow_unicode_escapes: bool) -> Result<ch
 
                 // Incorrect syntax has higher priority for error reporting
                 // than unallowed value for a literal.
-                if !allow_unicode_escapes {
+                if !mode.allow_unicode_escapes(rfc3349) {
                     return Err(EscapeError::UnicodeEscapeInByte);
                 }
 
@@ -322,18 +346,27 @@ fn scan_unicode(chars: &mut Chars<'_>, allow_unicode_escapes: bool) -> Result<ch
 }
 
 #[inline]
-fn ascii_check(c: char, allow_unicode_chars: bool) -> Result<char, EscapeError> {
-    if allow_unicode_chars || c.is_ascii() { Ok(c) } else { Err(EscapeError::NonAsciiCharInByte) }
+fn ascii_check(c: char, mode: Mode, rfc3349: &mut Rfc3349) -> Result<char, EscapeError> {
+    // We must check `is_ascii` first, to avoid setting `rfc3349` unnecessarily.
+    if c.is_ascii() || mode.allow_unicode_chars(rfc3349) {
+        Ok(c)
+    } else {
+        Err(EscapeError::NonAsciiCharInByte)
+    }
 }
 
 fn unescape_char_or_byte(chars: &mut Chars<'_>, mode: Mode) -> Result<char, EscapeError> {
     let c = chars.next().ok_or(EscapeError::ZeroChars)?;
+    let mut rfc3349 = Rfc3349::Unused;
     let res = match c {
-        '\\' => scan_escape(chars, mode),
+        '\\' => scan_escape(chars, mode, &mut rfc3349),
         '\n' | '\t' | '\'' => Err(EscapeError::EscapeOnlyChar),
         '\r' => Err(EscapeError::BareCarriageReturn),
-        _ => ascii_check(c, mode.allow_unicode_chars()),
+        _ => ascii_check(c, mode, &mut rfc3349),
     }?;
+
+    assert_eq!(rfc3349, Rfc3349::Unused); // rfc3349 not relevant for `Mode::{Char,Byte}`
+
     if chars.next().is_some() {
         return Err(EscapeError::MoreThanOneChar);
     }
@@ -342,12 +375,16 @@ fn unescape_char_or_byte(chars: &mut Chars<'_>, mode: Mode) -> Result<char, Esca
 
 /// Takes a contents of a string literal (without quotes) and produces a
 /// sequence of escaped characters or errors.
-fn unescape_non_raw_common<F, T: From<char> + From<u8>>(src: &str, mode: Mode, callback: &mut F)
+fn unescape_non_raw_common<F, T: From<char> + From<u8>>(
+    src: &str,
+    mode: Mode,
+    callback: &mut F,
+) -> Rfc3349
 where
     F: FnMut(Range<usize>, Result<T, EscapeError>),
 {
     let mut chars = src.chars();
-    let allow_unicode_chars = mode.allow_unicode_chars(); // get this outside the loop
+    let mut rfc3349 = Rfc3349::Unused;
 
     // The `start` and `end` computation here is complicated because
     // `skip_ascii_whitespace` makes us to skip over chars without counting
@@ -367,16 +404,17 @@ where
                         });
                         continue;
                     }
-                    _ => scan_escape::<T>(&mut chars, mode),
+                    _ => scan_escape::<T>(&mut chars, mode, &mut rfc3349),
                 }
             }
             '"' => Err(EscapeError::EscapeOnlyChar),
             '\r' => Err(EscapeError::BareCarriageReturn),
-            _ => ascii_check(c, allow_unicode_chars).map(T::from),
+            _ => ascii_check(c, mode, &mut rfc3349).map(T::from),
         };
         let end = src.len() - chars.as_str().len();
         callback(start..end, res);
     }
+    rfc3349
 }
 
 fn skip_ascii_whitespace<F>(chars: &mut Chars<'_>, start: usize, callback: &mut F)
@@ -409,12 +447,12 @@ where
 /// sequence of characters or errors.
 /// NOTE: Raw strings do not perform any explicit character escaping, here we
 /// only produce errors on bare CR.
-fn check_raw_common<F>(src: &str, mode: Mode, callback: &mut F)
+fn check_raw_common<F>(src: &str, mode: Mode, callback: &mut F) -> Rfc3349
 where
     F: FnMut(Range<usize>, Result<char, EscapeError>),
 {
     let mut chars = src.chars();
-    let allow_unicode_chars = mode.allow_unicode_chars(); // get this outside the loop
+    let mut rfc3349 = Rfc3349::Unused;
 
     // The `start` and `end` computation here matches the one in
     // `unescape_non_raw_common` for consistency, even though this function
@@ -423,16 +461,17 @@ where
         let start = src.len() - chars.as_str().len() - c.len_utf8();
         let res = match c {
             '\r' => Err(EscapeError::BareCarriageReturnInRawString),
-            _ => ascii_check(c, allow_unicode_chars),
+            _ => ascii_check(c, mode, &mut rfc3349),
         };
         let end = src.len() - chars.as_str().len();
         callback(start..end, res);
     }
+    rfc3349
 }
 
 #[inline]
-pub fn byte_from_char(c: char) -> u8 {
+pub(crate) fn byte_from_char(c: char) -> u8 {
     let res = c as u32;
-    debug_assert!(res <= u8::MAX as u32, "guaranteed because of ByteStr");
+    debug_assert!(res <= u8::MAX as u32, "guaranteed because of Byte");
     res as u8
 }