Fix mutating path of URL without authority (idempotency, empty path segments)

Author: theskim

url/src/lib.rs

@@ -2129,7 +2129,7 @@ impl Url {
         } else {
             self.host_end
         };
-        let suffix = self.slice(old_suffix_pos..).to_owned();
+        let mut suffix = self.slice(old_suffix_pos..).to_owned();
         self.serialization.truncate(self.host_start as usize);
         if !self.has_authority() {
             debug_assert!(self.slice(self.scheme_end..self.host_start) == ":");
@@ -2149,7 +2149,16 @@ impl Url {
                 write!(&mut self.serialization, ":{}", port).unwrap();
             }
         }
-        let new_suffix_pos = to_u32(self.serialization.len()).unwrap();
+        let mut new_suffix_pos = to_u32(self.serialization.len()).unwrap();
+
+        // Remove starting "/." for empty path segment followed by the host
+        if suffix.starts_with("/.//") {
+            let adjustment: usize = "/.".len();
+            suffix.drain(..adjustment);
+            // pathname should be "//p" not "p" given that the first segment was empty
+            new_suffix_pos -= adjustment as u32;
+        }
+
         self.serialization.push_str(&suffix);
 
         let adjust = |index: &mut u32| {

url/src/path_segments.rs

@@ -239,7 +239,7 @@ impl<'a> PathSegmentsMut<'a> {
         I::Item: AsRef<str>,
     {
         let scheme_type = SchemeType::from(self.url.scheme());
-        let path_start = self.url.path_start as usize;
+        let mut path_start = self.url.path_start as usize;
         self.url.mutate(|parser| {
             parser.context = parser::Context::PathSegmentSetter;
             for segment in segments {
@@ -253,7 +253,44 @@ impl<'a> PathSegmentsMut<'a> {
                 {
                     parser.serialization.push('/');
                 }
-                let mut has_host = true; // FIXME account for this?
+
+                let mut path_empty = false;
+
+                // Check ':' and then see if the next character is '/'
+                let mut has_host = if let Some(index) = parser.serialization.find(":") {
+                    if parser.serialization.len() > index + 1
+                        && parser.serialization.as_bytes().get(index + 1) == Some(&b'/')
+                    {
+                        let rest = &parser.serialization[(index + ":/".len())..];
+                        let host_part = rest.split('/').next().unwrap_or("");
+                        path_empty = rest.is_empty();
+                        !host_part.is_empty() && !host_part.contains('@')
+                    } else {
+                        false
+                    }
+                } else {
+                    false
+                };
+
+                // For cases where normalization is applied across both the serialization and the path.
+                // Append "/." immediately after the scheme (up to ":")
+                // This is done if three conditions are met.
+                // https://url.spec.whatwg.org/#url-serializing
+                // 1. The host is null
+                // 2. The url's path length is greater than 1
+                // 3. the first segment of the URL's path is an empty string
+                if !has_host && segment.len() > 1 && path_empty {
+                    if let Some(index) = parser.serialization.find(":") {
+                        if parser.serialization.len() == index + 2
+                            && parser.serialization.as_bytes().get(index + 1) == Some(&b'/')
+                        {
+                            // Append an extra '/' to ensure that "/./path" becomes "/.//path"
+                            parser.serialization.insert_str(index + ":".len(), "/./");
+                            path_start += "/.".len();
+                        }
+                    }
+                }
+
                 parser.parse_path(
                     scheme_type,
                     &mut has_host,
@@ -262,6 +299,7 @@ impl<'a> PathSegmentsMut<'a> {
                 );
             }
         });
+        self.url.path_start = path_start as u32;
         self
     }
 }

url/tests/expected_failures.txt

@@ -36,8 +36,6 @@
 <file:/.//p>
 <http://example.net/path> set hostname to <example.com:8080>
 <http://example.net:8080/path> set hostname to <example.com:>
-<non-spec:/.//p> set hostname to <h>
-<non-spec:/.//p> set hostname to <>
 <foo:///some/path> set pathname to <>
 <file:///var/log/system.log> set href to <http://0300.168.0xF0>
 <file://monkey/> set pathname to <\\\\>

url/tests/unit.rs

@@ -1419,3 +1419,55 @@ fn test_fuzzing_uri_failures() {
     assert_eq!(url.as_str(), "web+demo:////.dummy.path");
     url.check_invariants().unwrap();
 }
+
+#[test]
+fn test_can_be_a_base_with_set_path() {
+    use url::quirks;
+    let mut url = Url::parse("web+demo:/").unwrap();
+    assert!(!url.cannot_be_a_base());
+
+    url.set_path("//not-a-host");
+    assert_eq!(url.path(), "//not-a-host");
+
+    let segments: Vec<_> = url
+        .path_segments()
+        .expect("should have path segments")
+        .collect();
+
+    assert_eq!(segments, vec!["", "not-a-host"]);
+
+    url.set_query(Some("query"));
+    url.set_fragment(Some("frag"));
+
+    assert_eq!(url.as_str(), "web+demo:/.//not-a-host?query#frag");
+    quirks::set_hostname(&mut url, "test").unwrap();
+    assert_eq!(url.as_str(), "web+demo://test//not-a-host?query#frag");
+    url.check_invariants().unwrap();
+    quirks::set_hostname(&mut url, "").unwrap();
+    assert_eq!(url.as_str(), "web+demo:////not-a-host?query#frag");
+    url.check_invariants().unwrap();
+}
+
+#[test]
+fn test_can_be_a_base_with_path_segments_mut() {
+    let mut url = Url::parse("web+demo:/").unwrap();
+    assert!(!url.cannot_be_a_base());
+
+    url.path_segments_mut()
+        .expect("should have path segments")
+        .push("")
+        .push("not-a-host");
+
+    url.set_query(Some("query"));
+    url.set_fragment(Some("frag"));
+
+    assert_eq!(url.as_str(), "web+demo:/.//not-a-host?query#frag");
+    assert_eq!(url.path(), "//not-a-host");
+    url.check_invariants().unwrap();
+
+    let segments: Vec<_> = url
+        .path_segments()
+        .expect("should have path segments")
+        .collect();
+    assert_eq!(segments, vec!["", "not-a-host"]);
+}