cmd-diff: attempt to normalize metal diffs

The diffs contain a bunch of files with sha256 checksums in them, which
are unique on every build. Let's attempt to normalize the output
directories so the diffs will me less noisy and more meaningful.

Also fixup some permissions on some files because they can't be diffed
otherwise.

Author: dustymabe

src/cmd-diff

@@ -564,9 +564,14 @@ def diff_metal_helper(diff_from, diff_to):
 
 
             with tempfile.NamedTemporaryFile(suffix=".tar", delete=True) as tmp_tar:
+                # Exclude ostree/repo/objects. It just adds noise to the diff
+                excludes = ['*ostree/repo/objects/*']
                 g.tar_out("/", tmp_tar.name, xattrs=True, selinux=True, excludes=excludes)
-                # Extract the tarball.
-                runcmd(['tar', '-xf', tmp_tar.name, '-C', diff_dir])
+                # Extract the tarball. Normalize the output by replacing sha256sum hashes
+                # in filenames with XXXXXXXXXXXXXXXX so that we can get a real diff between
+                # two of the same files in different builds.
+                runcmd(['tar', '-xf', tmp_tar.name, '-C', diff_dir,
+                        '--transform', 's|[[:xdigit:]]{64}|XXXXXXXXXXXXXXXX|gx'])
 
         except Exception as e:
             print(f"Error in guestfs process for {image_path}: {e}", file=sys.stderr)
@@ -575,6 +580,13 @@ def diff_metal_helper(diff_from, diff_to):
             if g:
                 g.close()
 
+        # Some files like /etc/shadow and sudo have no read permissions so let's
+        # open it up so the difftool can access it.
+        runcmd(['find', diff_dir, '-type', 'f', '-perm', '000',
+               '-exec', 'chmod', '--verbose', '444', '{}', ';'])
+        runcmd(['find', diff_dir, '-type', 'f', '-perm', '111',
+               '-exec', 'chmod', '--verbose', '555', '{}', ';'])
+
     # Allow the caller to operate on these values
     return diff_dir_from, diff_dir_to