Auto merge of #90559 - rusticstuff:optimize-bidi-detection, r=davidtwco

Optimize bidi character detection.

Should fix most of the performance regression of the bidi character detection (#90514), to be confirmed with a perf run.

compiler/rustc_ast/src/lib.rs

@@ -16,6 +16,7 @@
 #![feature(nll)]
 #![feature(min_specialization)]
 #![recursion_limit = "256"]
+#![feature(slice_internals)]
 
 #[macro_use]
 extern crate rustc_macros;
@@ -25,6 +26,7 @@ pub mod util {
     pub mod comments;
     pub mod literal;
     pub mod parser;
+    pub mod unicode;
 }
 
 pub mod ast;

compiler/rustc_ast/src/util/unicode.rs

@@ -0,0 +1,35 @@
+pub const TEXT_FLOW_CONTROL_CHARS: &[char] = &[
+    '\u{202A}', '\u{202B}', '\u{202D}', '\u{202E}', '\u{2066}', '\u{2067}', '\u{2068}', '\u{202C}',
+    '\u{2069}',
+];
+
+#[inline]
+pub fn contains_text_flow_control_chars(s: &str) -> bool {
+    // Char   - UTF-8
+    // U+202A - E2 80 AA
+    // U+202B - E2 80 AB
+    // U+202C - E2 80 AC
+    // U+202D - E2 80 AD
+    // U+202E - E2 80 AE
+    // U+2066 - E2 81 A6
+    // U+2067 - E2 81 A7
+    // U+2068 - E2 81 A8
+    // U+2069 - E2 81 A9
+    let mut bytes = s.as_bytes();
+    loop {
+        match core::slice::memchr::memchr(0xE2, &bytes) {
+            Some(idx) => {
+                // bytes are valid UTF-8 -> E2 must be followed by two bytes
+                let ch = &bytes[idx..idx + 3];
+                match ch {
+                    [_, 0x80, 0xAA..=0xAE] | [_, 0x81, 0xA6..=0xA9] => break true,
+                    _ => {}
+                }
+                bytes = &bytes[idx + 3..];
+            }
+            None => {
+                break false;
+            }
+        }
+    }
+}

compiler/rustc_lint/src/context.rs

@@ -16,9 +16,9 @@
 
 use self::TargetLint::*;
 
-use crate::hidden_unicode_codepoints::UNICODE_TEXT_FLOW_CHARS;
 use crate::levels::{is_known_lint_tool, LintLevelsBuilder};
 use crate::passes::{EarlyLintPassObject, LateLintPassObject};
+use ast::util::unicode::TEXT_FLOW_CONTROL_CHARS;
 use rustc_ast as ast;
 use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::sync;
@@ -602,7 +602,7 @@ pub trait LintContext: Sized {
                     let spans: Vec<_> = content
                         .char_indices()
                         .filter_map(|(i, c)| {
-                            UNICODE_TEXT_FLOW_CHARS.contains(&c).then(|| {
+                            TEXT_FLOW_CONTROL_CHARS.contains(&c).then(|| {
                                 let lo = span.lo() + BytePos(2 + i as u32);
                                 (c, span.with_lo(lo).with_hi(lo + BytePos(c.len_utf8() as u32)))
                             })

compiler/rustc_lint/src/hidden_unicode_codepoints.rs

@@ -1,4 +1,5 @@
 use crate::{EarlyContext, EarlyLintPass, LintContext};
+use ast::util::unicode::{contains_text_flow_control_chars, TEXT_FLOW_CONTROL_CHARS};
 use rustc_ast as ast;
 use rustc_errors::{Applicability, SuggestionStyle};
 use rustc_span::{BytePos, Span, Symbol};
@@ -37,11 +38,6 @@ declare_lint! {
 
 declare_lint_pass!(HiddenUnicodeCodepoints => [TEXT_DIRECTION_CODEPOINT_IN_LITERAL]);
 
-crate const UNICODE_TEXT_FLOW_CHARS: &[char] = &[
-    '\u{202A}', '\u{202B}', '\u{202D}', '\u{202E}', '\u{2066}', '\u{2067}', '\u{2068}', '\u{202C}',
-    '\u{2069}',
-];
-
 impl HiddenUnicodeCodepoints {
     fn lint_text_direction_codepoint(
         &self,
@@ -57,7 +53,7 @@ impl HiddenUnicodeCodepoints {
             .as_str()
             .char_indices()
             .filter_map(|(i, c)| {
-                UNICODE_TEXT_FLOW_CHARS.contains(&c).then(|| {
+                TEXT_FLOW_CONTROL_CHARS.contains(&c).then(|| {
                     let lo = span.lo() + BytePos(i as u32 + padding);
                     (c, span.with_lo(lo).with_hi(lo + BytePos(c.len_utf8() as u32)))
                 })
@@ -131,7 +127,7 @@ impl HiddenUnicodeCodepoints {
 impl EarlyLintPass for HiddenUnicodeCodepoints {
     fn check_attribute(&mut self, cx: &EarlyContext<'_>, attr: &ast::Attribute) {
         if let ast::AttrKind::DocComment(_, comment) = attr.kind {
-            if comment.as_str().contains(UNICODE_TEXT_FLOW_CHARS) {
+            if contains_text_flow_control_chars(&comment.as_str()) {
                 self.lint_text_direction_codepoint(cx, comment, attr.span, 0, false, "doc comment");
             }
         }
@@ -142,7 +138,7 @@ impl EarlyLintPass for HiddenUnicodeCodepoints {
         let (text, span, padding) = match &expr.kind {
             ast::ExprKind::Lit(ast::Lit { token, kind, span }) => {
                 let text = token.symbol;
-                if !text.as_str().contains(UNICODE_TEXT_FLOW_CHARS) {
+                if !contains_text_flow_control_chars(&text.as_str()) {
                     return;
                 }
                 let padding = match kind {

compiler/rustc_parse/src/lexer/mod.rs

@@ -1,6 +1,7 @@
 use rustc_ast::ast::{self, AttrStyle};
 use rustc_ast::token::{self, CommentKind, Token, TokenKind};
 use rustc_ast::tokenstream::{Spacing, TokenStream};
+use rustc_ast::util::unicode::contains_text_flow_control_chars;
 use rustc_errors::{error_code, Applicability, DiagnosticBuilder, FatalError, PResult};
 use rustc_lexer::unescape::{self, Mode};
 use rustc_lexer::{Base, DocStyle, RawStrError};
@@ -137,12 +138,8 @@ impl<'a> StringReader<'a> {
         // Opening delimiter of the length 2 is not included into the comment text.
         let content_start = start + BytePos(2);
         let content = self.str_from(content_start);
-        let span = self.mk_sp(start, self.pos);
-        const UNICODE_TEXT_FLOW_CHARS: &[char] = &[
-            '\u{202A}', '\u{202B}', '\u{202D}', '\u{202E}', '\u{2066}', '\u{2067}', '\u{2068}',
-            '\u{202C}', '\u{2069}',
-        ];
-        if content.contains(UNICODE_TEXT_FLOW_CHARS) {
+        if contains_text_flow_control_chars(content) {
+            let span = self.mk_sp(start, self.pos);
             self.sess.buffer_lint_with_diagnostic(
                 &TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
                 span,