Preserve explicitly empty URI authorities

Author: ac1982

docs/source/release-notes/unreleased.rst

@@ -1,6 +1,7 @@
 2.x.y - 202z-aa-bb
 ------------------
 
-- *Add Items here*
+- Preserve an explicitly empty authority when parsing, normalizing, and
+  reassembling URI and IRI references, including ``file:///`` references.
 
 .. links below here

src/rfc3986/_mixin.py

@@ -360,7 +360,7 @@ def unsplit(self) -> str:
         result_list: list[str] = []
         if self.scheme:
             result_list.extend([self.scheme, ":"])
-        if self.authority:
+        if self.authority is not None:
             result_list.extend(["//", self.authority])
         if self.path:
             result_list.append(self.path)

src/rfc3986/iri.py

@@ -55,7 +55,7 @@ def __new__(
         ref = super().__new__(
             cls,
             scheme or None,
-            authority or None,
+            authority,
             path or None,
             query,
             fragment,

src/rfc3986/parseresult.py

@@ -158,7 +158,7 @@ def from_parts(
             authority += f":{port}"
         uri_ref = uri.URIReference(
             scheme=scheme,
-            authority=authority,
+            authority=authority if authority or host == "" else None,
             path=path,
             query=query,
             fragment=fragment,
@@ -344,7 +344,7 @@ def from_parts(
             authority += f":{int(port)}"
         uri_ref = uri.URIReference(
             scheme=scheme,
-            authority=authority,
+            authority=authority if authority or host == "" else None,
             path=path,
             query=query,
             fragment=fragment,
@@ -404,9 +404,9 @@ def from_string(
         )
 
     @property
-    def authority(self) -> bytes:
+    def authority(self) -> t.Optional[bytes]:
         """Return the normalized authority."""
-        return self.reference.authority.encode(self.encoding)
+        return compat.to_bytes(self.reference.authority, self.encoding)
 
     def copy_with(
         self,

src/rfc3986/uri.py

@@ -96,7 +96,7 @@ def __new__(
         ref = super().__new__(
             cls,
             scheme or None,
-            authority or None,
+            authority,
             path or None,
             query,
             fragment,
@@ -136,11 +136,17 @@ def normalize(self) -> "URIReference":
         """
         # See http://tools.ietf.org/html/rfc3986#section-6.2.2 for logic in
         # this method.
+        authority = self.authority
+        if authority:
+            authority = (
+                normalizers.normalize_authority(
+                    (self.userinfo, self.host, self.port)
+                )
+                or None
+            )
         return URIReference(
             normalizers.normalize_scheme(self.scheme or ""),
-            normalizers.normalize_authority(
-                (self.userinfo, self.host, self.port)
-            ),
+            authority,
             normalizers.normalize_path(self.path or ""),
             normalizers.normalize_query(self.query),
             normalizers.normalize_fragment(self.fragment),

tests/test_empty_authority.py

@@ -0,0 +1,62 @@
+import pytest
+
+from rfc3986 import iri_reference
+from rfc3986 import uri_reference
+from rfc3986 import urlparse
+from rfc3986.parseresult import ParseResult
+from rfc3986.parseresult import ParseResultBytes
+
+
+@pytest.mark.parametrize("parse", [uri_reference, iri_reference, urlparse])
+@pytest.mark.parametrize(
+    "text", ["foo:///", "file:///tmp/a", "//", "///a?x#y"]
+)
+def test_empty_authority_round_trip(parse, text):
+    reference = parse(text)
+    assert reference.authority == ""
+    assert reference.unsplit() == text
+    assert (
+        reference.copy_with(fragment="new").unsplit()
+        == text.split("#")[0] + "#new"
+    )
+
+
+@pytest.mark.parametrize("text", ["foo:///", "file:///tmp/a", "//"])
+def test_normalization_and_encoding_preserve_empty_authority(text):
+    assert uri_reference(text).normalize().unsplit() == text
+    assert iri_reference(text).encode().unsplit() == text
+    assert urlparse(text).encode().unsplit() == text.encode()
+
+
+@pytest.mark.parametrize("text", ["foo:/", "/tmp/a", "a", ""])
+def test_absent_authority_remains_absent(text):
+    reference = uri_reference(text)
+    assert reference.authority is None
+    assert reference.normalize().authority is None
+    assert reference.unsplit() == text
+
+
+@pytest.mark.parametrize("cls", [ParseResult, ParseResultBytes])
+def test_from_parts_distinguishes_empty_and_absent_host(cls):
+    absent = cls.from_parts(scheme="foo", path="/a")
+    empty = cls.from_parts(scheme="foo", host="", path="/a")
+
+    def as_string(value):
+        return value.decode() if isinstance(value, bytes) else value
+
+    assert as_string(absent.unsplit()) == "foo:/a"
+    assert as_string(empty.unsplit()) == "foo:///a"
+
+
+@pytest.mark.parametrize(
+    "relative, base, expected",
+    [
+        ("///new", "foo://example.com/old", "foo:///new"),
+        ("new", "foo://", "foo:///new"),
+        ("", "foo://", "foo://"),
+    ],
+)
+def test_resolve_with_empty_authority(relative, base, expected):
+    result = uri_reference(relative).resolve_with(base)
+    assert result.authority == ""
+    assert result.unsplit() == expected