regex: some small cleanup in 'strip.rs'

We also utilize bstr's methods to get rid of some helpers we had written
by hand.

crates/regex/Cargo.toml

@@ -11,7 +11,7 @@ repository = "https://github.com/BurntSushi/ripgrep/tree/master/crates/regex"
 readme = "README.md"
 keywords = ["regex", "grep", "search", "pattern", "line"]
 license = "Unlicense OR MIT"
-edition = "2018"
+edition = "2021"
 
 [dependencies]
 aho-corasick = "1.0.2"

crates/regex/src/error.rs

@@ -1,8 +1,3 @@
-use std::error;
-use std::fmt;
-
-use crate::util;
-
 /// An error that can occur in this crate.
 ///
 /// Generally, this error corresponds to problems building a regular
@@ -32,7 +27,7 @@ impl Error {
         }
     }
 
-    pub(crate) fn generic<E: error::Error>(err: E) -> Error {
+    pub(crate) fn generic<E: std::error::Error>(err: E) -> Error {
         Error { kind: ErrorKind::Regex(err.to_string()) }
     }
 
@@ -68,18 +63,23 @@ pub enum ErrorKind {
     InvalidLineTerminator(u8),
 }
 
-impl error::Error for Error {}
+impl std::error::Error for Error {}
+
+impl std::fmt::Display for Error {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        use bstr::ByteSlice;
 
-impl fmt::Display for Error {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self.kind {
             ErrorKind::Regex(ref s) => write!(f, "{}", s),
             ErrorKind::NotAllowed(ref lit) => {
-                write!(f, "the literal '{:?}' is not allowed in a regex", lit)
+                write!(f, "the literal {:?} is not allowed in a regex", lit)
             }
             ErrorKind::InvalidLineTerminator(byte) => {
-                let x = util::show_bytes(&[byte]);
-                write!(f, "line terminators must be ASCII, but '{}' is not", x)
+                write!(
+                    f,
+                    "line terminators must be ASCII, but {} is not",
+                    [byte].as_bstr()
+                )
             }
         }
     }

crates/regex/src/lib.rs

@@ -14,5 +14,4 @@ mod matcher;
 mod multi;
 mod non_matching;
 mod strip;
-mod util;
 mod word;

crates/regex/src/strip.rs

@@ -1,5 +1,7 @@
-use grep_matcher::LineTerminator;
-use regex_syntax::hir::{self, Hir, HirKind};
+use {
+    grep_matcher::LineTerminator,
+    regex_syntax::hir::{self, Hir, HirKind},
+};
 
 use crate::error::{Error, ErrorKind};
 
@@ -15,7 +17,26 @@ use crate::error::{Error, ErrorKind};
 ///
 /// If the given line terminator is not ASCII, then this function returns an
 /// error.
-pub fn strip_from_match(
+///
+/// Note that as of regex 1.9, this routine could theoretically be implemented
+/// without returning an error. Namely, for example, we could turn
+/// `foo\nbar` into `foo[a&&b]bar`. That is, replace line terminators with a
+/// sub-expression that can never match anything. Thus, ripgrep would accept
+/// such regexes and just silently not match anything. Regex versions prior to 1.8
+/// don't support such constructs. I ended up deciding to leave the existing
+/// behavior of returning an error instead. For example:
+///
+/// ```text
+/// $ echo -n 'foo\nbar\n' | rg 'foo\nbar'
+/// the literal '"\n"' is not allowed in a regex
+///
+/// Consider enabling multiline mode with the --multiline flag (or -U for short).
+/// When multiline mode is enabled, new line characters can be matched.
+/// ```
+///
+/// This looks like a good error message to me, and even suggests a flag that
+/// the user can use instead.
+pub(crate) fn strip_from_match(
     expr: Hir,
     line_term: LineTerminator,
 ) -> Result<Hir, Error> {
@@ -23,23 +44,20 @@ pub fn strip_from_match(
         let expr1 = strip_from_match_ascii(expr, b'\r')?;
         strip_from_match_ascii(expr1, b'\n')
     } else {
-        let b = line_term.as_byte();
-        if b > 0x7F {
-            return Err(Error::new(ErrorKind::InvalidLineTerminator(b)));
-        }
-        strip_from_match_ascii(expr, b)
+        strip_from_match_ascii(expr, line_term.as_byte())
     }
 }
 
-/// The implementation of strip_from_match. The given byte must be ASCII. This
-/// function panics otherwise.
+/// The implementation of strip_from_match. The given byte must be ASCII.
+/// This function returns an error otherwise. It also returns an error if
+/// it couldn't remove `\n` from the given regex without leaving an empty
+/// character class in its place.
 fn strip_from_match_ascii(expr: Hir, byte: u8) -> Result<Hir, Error> {
-    assert!(byte <= 0x7F);
-    let chr = byte as char;
-    assert_eq!(chr.len_utf8(), 1);
-
-    let invalid = || Err(Error::new(ErrorKind::NotAllowed(chr.to_string())));
-
+    if !byte.is_ascii() {
+        return Err(Error::new(ErrorKind::InvalidLineTerminator(byte)));
+    }
+    let ch = char::from(byte);
+    let invalid = || Err(Error::new(ErrorKind::NotAllowed(ch.to_string())));
     Ok(match expr.into_kind() {
         HirKind::Empty => Hir::empty(),
         HirKind::Literal(hir::Literal(lit)) => {
@@ -50,7 +68,7 @@ fn strip_from_match_ascii(expr: Hir, byte: u8) -> Result<Hir, Error> {
         }
         HirKind::Class(hir::Class::Unicode(mut cls)) => {
             let remove = hir::ClassUnicode::new(Some(
-                hir::ClassUnicodeRange::new(chr, chr),
+                hir::ClassUnicodeRange::new(ch, ch),
             ));
             cls.difference(&remove);
             if cls.ranges().is_empty() {

crates/regex/src/util.rs

@@ -1,30 +0,0 @@
-/// Converts an arbitrary sequence of bytes to a literal suitable for building
-/// a regular expression.
-#[allow(dead_code)]
-pub fn bytes_to_regex(bs: &[u8]) -> String {
-    use regex_syntax::is_meta_character;
-    use std::fmt::Write;
-
-    let mut s = String::with_capacity(bs.len());
-    for &b in bs {
-        if b <= 0x7F && !is_meta_character(b as char) {
-            write!(s, r"{}", b as char).unwrap();
-        } else {
-            write!(s, r"\x{:02x}", b).unwrap();
-        }
-    }
-    s
-}
-
-/// Converts arbitrary bytes to a nice string.
-pub fn show_bytes(bs: &[u8]) -> String {
-    use std::ascii::escape_default;
-    use std::str;
-
-    let mut nice = String::new();
-    for &b in bs {
-        let part: Vec<u8> = escape_default(b).collect();
-        nice.push_str(str::from_utf8(&part).unwrap());
-    }
-    nice
-}