Allow '[' and ']' in URI paths.

This is strictly noncompliant and they should be encoded, but browsers
routinely send them unencoded, so we allow them to avoid trouble.

core/http/src/parse/uri/parser.rs

@@ -146,7 +146,7 @@ fn host<'a>(
     input: &mut RawInput<'a>,
 ) -> Result<'a, Extent<&'a [u8]>> {
     switch! {
-        peek(b'[') => enclosed(b'[', is_pchar, b']')?,
+        peek(b'[') => enclosed(b'[', is_host_char, b']')?,
         _ => take_while(is_reg_name_char)?
     }
 }

core/http/src/parse/uri/tables.rs

@@ -58,7 +58,11 @@ const USER_INFO_CHARS: [u8; 256] = char_table(&[
 ]);
 
 pub const PATH_CHARS: [u8; 256] = char_table(&[
-    &REG_NAME_CHARS, &[b':', b'@', b'/']
+    &REG_NAME_CHARS, &[b':', b'@', b'/'],
+
+    // NOTE: these are _not_ accepted in RFC 7230/3986. However, browsers
+    // routinely send these unencoded, so allow them to support the real-world.
+    &[b'[',  b']'],
 ]);
 
 const QUERY_CHARS: [u8; 256] = char_table(&[
@@ -72,6 +76,9 @@ const QUERY_CHARS: [u8; 256] = char_table(&[
 #[inline(always)]
 pub const fn is_pchar(&c: &u8) -> bool { PATH_CHARS[c as usize] != 0 }
 
+#[inline(always)]
+pub const fn is_host_char(c: &u8) -> bool { is_pchar(c) && *c != b'[' && *c != b']' }
+
 #[inline(always)]
 pub const fn is_scheme_char(&c: &u8) -> bool { SCHEME_CHAR[c as usize] != 0 }
 

core/http/src/parse/uri/tests.rs

@@ -143,7 +143,7 @@ fn single_byte() {
         "@" => Authority::new("", "", None),
     );
 
-    assert_no_parse!("[", "]");
+    assert_no_parse!["\\", "^"];
 }
 
 #[test]

core/lib/tests/uri-percent-encoding-issue-808.rs

@@ -31,7 +31,7 @@ mod tests {
 
     #[test]
     fn uri_percent_encoding_redirect() {
-        let expected_location = vec!["/hello/John%5B%5D%7C%5C%25@%5E"];
+        let expected_location = vec!["/hello/John[]%7C%5C%25@%5E"];
         let client = Client::debug(rocket()).unwrap();
 
         let response = client.get("/raw").dispatch();