Merge pull request #317 from python/feature/faster-entry-points

Use normalized names to distinguish unique distributions for performance

CHANGES.rst

@@ -1,3 +1,10 @@
+v4.3.0
+=======
+
+* #317: De-duplication of distributions no longer requires
+  loading the full metadata for ``PathDistribution`` objects,
+  entry point loading performance by ~10x.
+
 v4.2.0
 =======
 

importlib_metadata/__init__.py

@@ -498,6 +498,11 @@ def name(self):
         """Return the 'Name' metadata for the distribution package."""
         return self.metadata['Name']
 
+    @property
+    def _normalized_name(self):
+        """Return a normalized version of the name."""
+        return Prepared.normalize(self.name)
+
     @property
     def version(self):
         """Return the 'Version' metadata for the distribution package."""
@@ -805,6 +810,12 @@ def read_text(self, filename):
     def locate_file(self, path):
         return self._path.parent / path
 
+    @property
+    def _normalized_name(self):
+        stem = os.path.basename(str(self._path))
+        name, sep, rest = stem.partition('-')
+        return name
+
 
 def distribution(distribution_name):
     """Get the ``Distribution`` instance for the named package.
@@ -859,7 +870,8 @@ def entry_points(**params) -> Union[EntryPoints, SelectableGroups]:
 
     :return: EntryPoints or SelectableGroups for all installed packages.
     """
-    unique = functools.partial(unique_everseen, key=operator.attrgetter('name'))
+    norm_name = operator.attrgetter('_normalized_name')
+    unique = functools.partial(unique_everseen, key=norm_name)
     eps = itertools.chain.from_iterable(
         dist.entry_points for dist in unique(distributions())
     )