Merge pull request #105 from goldmann/gh-104-broken-symlinks

Better handling of symlinks pointing to non-existing locations

docker_squash/image.py

@@ -581,12 +581,22 @@ def _squash_layers(self, layers_to_squash, layers_to_move):
                             skipped_hard_links.append(member)
                             continue
 
+                        # If we have a symlink to a location that does not exist in the
+                        # squashed tar (yet), we don't know if it's a file or a directory.
+                        # In case it's a directory - we cannot add any new files under
+                        # this directory because extration will fail at the import time.
+                        # Assume here the target is directory (if it's a file it won't
+                        # break) and skip all files that should be added under this directory.
+                        if member.issym() and member.linkname not in squashed_files:
+                            to_skip.append(os.path.join(member.name, "/"))
+                            continue
+
                         if member.isfile():
                             # Finally add the file to archive
                             squashed_tar.addfile(
                                 member, layer_tar.extractfile(member))
                         else:
-                            # Special cases: symlinks and hard links, and other files, we skip the file itself
+                            # Special case: other(?) files, we skip the file itself
                             squashed_tar.addfile(member)
 
                         # We added a file to the squashed tar, so let's note it

tests/test_integ_squash.py

@@ -737,6 +737,21 @@ def test_should_squash_correctly_hardlinks(self):
                 squashed_image.assertFileExists('usr/libexec/git-core/git-remote-ftp')
                 squashed_image.assertFileExists('usr/libexec/git-core/git-remote-http')
 
+    # https://github.com/goldmann/docker-squash/issues/104
+    def test_should_handle_symlinks_to_nonexisting_locations(self):
+        dockerfile = '''
+        FROM %s
+        RUN mkdir -p /var/log
+        RUN touch /var/log/somelog
+        RUN mv /var/log /var/log-removed && ln -sf /data/var/log /var/log
+        RUN rm -rf /var/log-removed
+        ''' % TestIntegSquash.BUSYBOX_IMAGE
+
+        with self.Image(dockerfile) as image:
+            with self.SquashedImage(image, 3, numeric=True) as squashed_image:
+                self.assertEqual(
+                    len(squashed_image.layers), len(image.layers) - 2)
+
     def test_should_squash_every_layer_from_an_image_from_docker_hub(self):
         dockerfile = '''
         FROM python:3.5.1-alpine