fix urljoin bug causing multiple slashes to be merged into one

archivebox/parsers/generic_html.py

@@ -40,9 +40,20 @@ def parse_generic_html_export(html_file: IO[str], root_url: Optional[str]=None,
         parser.feed(line)
         for url in parser.urls:
             if root_url:
-                # resolve relative urls /home.html -> https://example.com/home.html
+                url_is_absolute = (url.lower().startswith('http://') or url.lower().startswith('https://'))
-                url = urljoin(root_url, url)
+                # url = https://abc.com                       => True
-                # TODO: fix double // getting stripped by urljoin bug https://github.com/python/cpython/issues/96015
+                # url = /page.php?next=https://example.com    => False
+
+                if not url_is_absolute:                       # resolve it by joining it with root_url
+                    relative_path = url
+
+                    url = urljoin(root_url, relative_path)    # https://example.com/somepage.html + /home.html
+                                                              # => https://example.com/home.html
+
+                    # special case to handle bug around // handling, crucial for urls that contain sub-urls
+                    # e.g. https://web.archive.org/web/https://example.com
+                    if did_urljoin_misbehave(root_url, relative_path, url):
+                        url = fix_urljoin_bug(url)
 
             for archivable_url in find_all_urls(url):
                 yield Link(
@@ -57,3 +68,74 @@ def parse_generic_html_export(html_file: IO[str], root_url: Optional[str]=None,
 KEY = 'html'
 NAME = 'Generic HTML'
 PARSER = parse_generic_html_export
+
+
+#### WORKAROUND CODE FOR https://github.com/python/cpython/issues/96015 ####
+
+def did_urljoin_misbehave(root_url: str, relative_path: str, final_url: str) -> bool:
+    """
+    Handle urljoin edge case bug where multiple slashes get turned into a single slash:
+    - https://github.com/python/cpython/issues/96015
+    - https://github.com/ArchiveBox/ArchiveBox/issues/1411
+
+    This workaround only fixes the most common case of a sub-URL inside an outer URL, e.g.:
+       https://web.archive.org/web/https://example.com/some/inner/url
+
+    But there are other valid URLs containing // that are not fixed by this workaround, e.g.:
+       https://example.com/drives/C//some/file
+    """
+
+    # if relative path is actually an absolute url, cut off its own scheme so we check the path component only
+    relative_path = relative_path.lower()
+    if relative_path.startswith('http://') or relative_path.startswith('https://'):
+        relative_path = relative_path.split('://', 1)[-1]
+
+    # TODO: properly fix all double // getting stripped by urljoin, not just ://
+    original_path_had_suburl = '://' in relative_path
+    original_root_had_suburl = '://' in root_url[8:]     # ignore first 8 chars because root always starts with https://
+    final_joined_has_suburl = '://' in final_url[8:]     # ignore first 8 chars because final always starts with https://
+
+    urljoin_broke_suburls = (
+        (original_root_had_suburl or original_path_had_suburl)
+        and not final_joined_has_suburl
+    )
+    return urljoin_broke_suburls
+
+
+def fix_urljoin_bug(url: str, nesting_limit=5):
+    """
+    recursively replace broken suburls .../http:/... with .../http://...
+
+    basically equivalent to this for 99.9% of cases:
+      url = url.replace('/http:/',  '/http://')
+      url = url.replace('/https:/', '/https://')
+    except this handles:
+        other schemes besides http/https     (e.g. https://example.com/link/git+ssh://github.com/example)
+        other preceding separators besides / (e.g. https://example.com/login/?next=https://example.com/home)
+        fixing multiple suburls recursively
+    """
+    input_url = url
+    for _ in range(nesting_limit):
+        url = re.sub(
+            r'(?P<root>.+?)'                             # https://web.archive.org/web
+            + r'(?P<separator>[-=/_&+%$#@!*\(\\])'       # /
+            + r'(?P<subscheme>[a-zA-Z0-9+_-]{1,32}?):/'  # http:/
+            + r'(?P<suburl>[^/\\]+)',                    # example.com
+            r"\1\2\3://\4",
+            input_url,
+            re.IGNORECASE | re.UNICODE,
+        )
+        if url == input_url:
+            break                                        # nothing left to replace, all suburls are fixed
+        input_url = url
+
+    return url
+
+
+# sanity check to make sure workaround code works as expected and doesnt introduce *more* bugs
+assert did_urljoin_misbehave('https://web.archive.org/web/https://example.com', 'abc.html', 'https://web.archive.org/web/https:/example.com/abc.html') == True
+assert did_urljoin_misbehave('http://example.com', 'https://web.archive.org/web/http://example.com/abc.html', 'https://web.archive.org/web/http:/example.com/abc.html') == True
+assert fix_urljoin_bug('https:/example.com') == 'https:/example.com'   # should not modify original url's scheme, only sub-urls
+assert fix_urljoin_bug('https://web.archive.org/web/https:/example.com/abc.html') == 'https://web.archive.org/web/https://example.com/abc.html'
+assert fix_urljoin_bug('http://example.com/link/git+ssh:/github.com/example?next=ftp:/example.com') == 'http://example.com/link/git+ssh://github.com/example?next=ftp://example.com'
+