Implement normalize lexically

Author: ChrisDenton

library/std/src/path.rs

@@ -2148,6 +2148,13 @@ pub struct Path {
 #[stable(since = "1.7.0", feature = "strip_prefix")]
 pub struct StripPrefixError(());
 
+/// An error returned from [`Path::normalize_lexically`] if a `..` parent reference
+/// would escape the path.
+#[unstable(feature = "normalize_lexically", issue = "134694")]
+#[derive(Debug, PartialEq)]
+#[non_exhaustive]
+pub struct NormalizeError;
+
 impl Path {
     // The following (private!) function allows construction of a path from a u8
     // slice, which is only safe when it is known to follow the OsStr encoding.
@@ -2962,6 +2969,61 @@ impl Path {
         fs::canonicalize(self)
     }
 
+    /// Normalize a path, including `..` without traversing the filesystem.
+    ///
+    /// <div class="warning">
+    ///
+    /// This function always resolves `..` to the "lexical" parent.
+    /// That is "a/b/../c" will always resolve to `a/c` which can change the meaning of the path.
+    /// In particular, `a/c` and `a/b/../c` are distinct on many systems because `b` may be a symbolic link, so its parent isn’t `a`.
+    ///
+    /// </div>
+    ///
+    /// [`path::absolute`](absolute) is an alternative that preserves `..`.
+    /// Or [`Path::canonicalize`] can be used to resolve any `..` by querying the filesystem.
+    #[unstable(feature = "normalize_lexically", issue = "134694")]
+    pub fn normalize_lexically(&self) -> Result<PathBuf, NormalizeError> {
+        let mut lexical = PathBuf::new();
+        let mut iter = self.components().peekable();
+
+        // Find the root, if any.
+        let root = match iter.peek() {
+            Some(Component::ParentDir) => return Err(NormalizeError),
+            Some(p @ Component::RootDir) | Some(p @ Component::CurDir) => {
+                lexical.push(p);
+                iter.next();
+                lexical.as_os_str().len()
+            }
+            Some(Component::Prefix(prefix)) => {
+                lexical.push(prefix.as_os_str());
+                iter.next();
+                if let Some(p @ Component::RootDir) = iter.peek() {
+                    lexical.push(p);
+                    iter.next();
+                }
+                lexical.as_os_str().len()
+            }
+            None => return Ok(PathBuf::new()),
+            Some(Component::Normal(_)) => 0,
+        };
+
+        for component in iter {
+            match component {
+                Component::RootDir | Component::Prefix(_) => return Err(NormalizeError),
+                Component::CurDir => continue,
+                Component::ParentDir => {
+                    if lexical.as_os_str().len() == root {
+                        return Err(NormalizeError);
+                    } else {
+                        lexical.pop();
+                    }
+                }
+                Component::Normal(path) => lexical.push(path),
+            }
+        }
+        Ok(lexical)
+    }
+
     /// Reads a symbolic link, returning the file that the link points to.
     ///
     /// This is an alias to [`fs::read_link`].
@@ -3503,6 +3565,15 @@ impl Error for StripPrefixError {
     }
 }
 
+#[unstable(feature = "normalize_lexically", issue = "none")]
+impl fmt::Display for NormalizeError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.write_str("parent reference `..` points outside of base directory")
+    }
+}
+#[unstable(feature = "normalize_lexically", issue = "none")]
+impl Error for NormalizeError {}
+
 /// Makes the path absolute without accessing the filesystem.
 ///
 /// If the path is relative, the current directory is used as the base directory.

library/std/src/path/tests.rs

@@ -2079,3 +2079,51 @@ fn clone_to_uninit() {
     unsafe { a.clone_to_uninit(ptr::from_mut::<Path>(&mut b).cast()) };
     assert_eq!(a, &*b);
 }
+
+#[test]
+fn normalize_lexically() {
+    #[track_caller]
+    fn check(a: &str, b: Result<&str, NormalizeError>) {
+        assert_eq!(Path::new(a).normalize_lexically(), b.map(PathBuf::from));
+    }
+
+    // Relative paths
+    check("a", Ok("a"));
+    check("./a", Ok("./a"));
+    check("a/b/c", Ok("a/b/c"));
+    check("a/././b/./c/.", Ok("a/b/c"));
+    check("a/../c", Ok("c"));
+    check("./a/b", Ok("./a/b"));
+    check("a/../b/c/..", Ok("b"));
+
+    check("..", Err(NormalizeError));
+    check("../..", Err(NormalizeError));
+    check("a/../..", Err(NormalizeError));
+    check("a/../../b", Err(NormalizeError));
+    check("a/../../b/c", Err(NormalizeError));
+    check("a/../b/../..", Err(NormalizeError));
+
+    // Check we don't escape the root or prefix
+    #[cfg(unix)]
+    {
+        check("/..", Err(NormalizeError));
+        check("/a/../..", Err(NormalizeError));
+    }
+    #[cfg(windows)]
+    {
+        check(r"C:\..", Err(NormalizeError));
+        check(r"C:\a\..\..", Err(NormalizeError));
+
+        check(r"C:..", Err(NormalizeError));
+        check(r"C:a\..\..", Err(NormalizeError));
+
+        check(r"\\server\share\..", Err(NormalizeError));
+        check(r"\\server\share\a\..\..", Err(NormalizeError));
+
+        check(r"\..", Err(NormalizeError));
+        check(r"\a\..\..", Err(NormalizeError));
+
+        check(r"\\?\UNC\server\share\..", Err(NormalizeError));
+        check(r"\\?\UNC\server\share\a\..\..", Err(NormalizeError));
+    }
+}