Protect @ as safe character when cleaning URLs

Remote URLs that contain a link to a git repository and a tag
reference will have the `@` character converted into `%40`. This is
incorrect.

Fixes: pypa#6437

Signed-off-by: Nicolas Bock <[email protected]>

Author: nicolasbock

news/6440.bugfix

@@ -0,0 +1,2 @@
+Fix a regression that caused `@` to be quoted in pypiserver links.
+This interfered with parsing the revision string from VCS urls.

src/pip/_internal/index.py

@@ -336,6 +336,7 @@ class FoundCandidates(object):
     * `evaluator`: A CandidateEvaluator object to sort applicable candidates
         by order of preference.
     """
+
     def __init__(
         self,
         candidates,     # type: List[InstallationCandidate]
@@ -1061,7 +1062,9 @@ def _clean_link(url):
         path = urllib_request.pathname2url(
             urllib_request.url2pathname(result.path))
     else:
-        path = urllib_parse.quote(urllib_parse.unquote(result.path))
+        # In addition to the `/` character we protect `@` so that
+        # revision strings in VCS URLs are properly parsed.
+        path = urllib_parse.quote(urllib_parse.unquote(result.path), safe="/@")
     return urllib_parse.urlunparse(result._replace(path=path))
 
 

tests/unit/test_index.py

@@ -308,7 +308,10 @@ def test_request_retries(caplog):
         # URL with something that looks like a drive letter, but is
         # not. The `:` should be quoted.
         ("https://localhost.localdomain/T:/path/",
-         "https://localhost.localdomain/T%3A/path/")
+         "https://localhost.localdomain/T%3A/path/"),
+        # VCS URL containing revision string.
+        ("git+ssh://example.com/path to/[email protected]#egg=my-package-1.0",
+         "git+ssh://example.com/path%20to/[email protected]#egg=my-package-1.0")
     ]
 )
 def test_clean_link(url, clean_url):