Fix URL parsing to include delimiters before slashes

README.md

@@ -22,6 +22,7 @@ Let's look at a few cases:
 Seems simple enough. But then we also have these cases:
 
 * `https://en.wikipedia.org/wiki/Link_(The_Legend_of_Zelda)` should include the trailing paren
+* `http://example.com/./` should include the dot because the slash makes it a path
 * `http://üñîçøðé.com/ä` should also work for Unicode (including Emoji and Punycode)
 * `<http://example.com/>` should not include angle brackets
 

src/url.rs

@@ -207,7 +207,7 @@ fn find_url_end(s: &str, quote: Option<char>, iri_parsing_enabled: bool) -> Opti
     let mut curly = 0;
     let mut single_quote = false;
 
-    let mut previous_can_be_last = true;
+    let mut previous_is_url_char = true;
     let mut end = Some(0);
 
     if !s[0..].starts_with("/") && !s[0..].starts_with("?") {
@@ -232,9 +232,12 @@ fn find_url_end(s: &str, quote: Option<char>, iri_parsing_enabled: bool) -> Opti
                 false
             }
             '/' => {
-                // This may be part of an URL and at the end, but not if the previous character
-                // can't be the end of an URL
-                previous_can_be_last
+                // A slash can be the end of a URL if the previous character is a valid URL
+                // character. This means that delimiters like `!` or `.` before a `/` are
+                // included in the URL, e.g. `/!/`, but non-URL characters (like non-ASCII
+                // when IRI parsing is disabled) before a `/` are not.
+                // See https://github.com/robinst/linkify/issues/90
+                previous_is_url_char
             }
             '(' => {
                 round += 1;
@@ -288,7 +291,12 @@ fn find_url_end(s: &str, quote: Option<char>, iri_parsing_enabled: bool) -> Opti
         if can_be_last {
             end = Some(i + c.len_utf8());
         }
-        previous_can_be_last = can_be_last;
+        // Track whether the current character is a valid URL character (even if it can't
+        // be the last character). Delimiters like `!` are valid URL characters, but
+        // non-ASCII characters are not valid URL characters when IRI parsing is disabled.
+        // This matters for `/` above: a slash after a delimiter like `!` should extend
+        // the URL (e.g. `/!/`), but a slash after a non-URL character should not.
+        previous_is_url_char = c.is_ascii() || iri_parsing_enabled;
     }
 
     end

tests/url.rs

@@ -151,6 +151,30 @@ fn delimiter_at_end_no_protocol() {
     assert_urls_without_protocol("example.org/;", "|example.org/|;");
 }
 
+#[test]
+fn delimiter_followed_by_slash() {
+    // https://github.com/robinst/linkify/issues/90
+    assert_linked("http://example.org/!/", "|http://example.org/!/|");
+    assert_linked("http://example.org/test!/", "|http://example.org/test!/|");
+    assert_linked("http://example.org/./", "|http://example.org/./|");
+    assert_linked("http://example.org/,/", "|http://example.org/,/|");
+    assert_linked("http://example.org/:/", "|http://example.org/:/|");
+    assert_linked("http://example.org/;/", "|http://example.org/;/|");
+    assert_linked("http://example.org/?/", "|http://example.org/?/|");
+    // Delimiter followed by slash and more path
+    assert_linked("http://example.org/!/a", "|http://example.org/!/a|");
+    assert_linked("http://example.org/test!/a", "|http://example.org/test!/a|");
+}
+
+#[test]
+fn delimiter_followed_by_slash_no_protocol() {
+    assert_urls_without_protocol("example.org/!/", "|example.org/!/|");
+    assert_urls_without_protocol("example.org/test!/", "|example.org/test!/|");
+    assert_urls_without_protocol("example.org/./", "|example.org/./|");
+    assert_urls_without_protocol("example.org/,/", "|example.org/,/|");
+    assert_urls_without_protocol("example.org/;/", "|example.org/;/|");
+}
+
 #[test]
 fn matching_punctuation() {
     assert_linked("http://example.org/a(b)", "|http://example.org/a(b)|");