bpo-30645: don't append to an inner loop path in imp.load_package() (GH-2268)

Bug didn't manifest itself when importing a module with source as .py files are always the first on the search path. The issue only showed up in bytecode-only packages where the calculated file path would be ``__init__.py/__init__.pyc``.

Patch by Alexandru Ardelean.

Lib/imp.py

@@ -203,8 +203,9 @@ def load_package(name, path):
         extensions = (machinery.SOURCE_SUFFIXES[:] +
                       machinery.BYTECODE_SUFFIXES[:])
         for extension in extensions:
-            path = os.path.join(path, '__init__'+extension)
+            init_path = os.path.join(path, '__init__' + extension)
-            if os.path.exists(path):
+            if os.path.exists(init_path):
+                path = init_path
                 break
         else:
             raise ValueError('{!r} is not a package'.format(path))

Misc/ACKS

@@ -57,6 +57,7 @@ Ankur Ankan
 Heidi Annexstad
 Ramchandra Apte
 Éric Araujo
+Alexandru Ardelean
 Alicia Arlen
 Jeffrey Armstrong
 Jason Asbahr

Misc/NEWS

@@ -428,6 +428,10 @@ Library
 - bpo-30508: Don't log exceptions if Task/Future "cancel()" method was
   called.
 
+- bpo-30645: Fix path calculation in `imp.load_package()`, fixing it for
+  cases when a package is only shipped with bytecodes. Patch by
+  Alexandru Ardelean.
+
 - bpo-11822: The dis.dis() function now is able to disassemble nested
   code objects.