newdb: Fix replacing formats in a book creating orphans

When replacing formats in a book with a very long title+authors on
windows, newdb could leave behind the old format file, because the
filename shortening algorithm has changed. Handle that case.

src/calibre/db/backend.py

@@ -1337,7 +1337,7 @@ class DB(object):
                         if wam is not None:
                             wam.close_handles()
 
-    def add_format(self, book_id, fmt, stream, title, author, path):
+    def add_format(self, book_id, fmt, stream, title, author, path, current_name):
         fmt = ('.' + fmt.lower()) if fmt else ''
         fname = self.construct_file_name(book_id, title, author, len(fmt))
         path = os.path.join(self.library_path, path)
@@ -1345,6 +1345,22 @@ class DB(object):
         if not os.path.exists(path):
             os.makedirs(path)
         size = 0
+        if current_name is not None:
+            old_path = os.path.join(path, current_name + fmt)
+            if old_path != dest:
+                # Ensure that the old format file is not orphaned, this can
+                # happen if the algorithm in construct_file_name is changed.
+                try:
+                    # rename rather than remove, so that if something goes
+                    # wrong in the rest of this function, at least the file is
+                    # not deleted
+                    os.rename(old_path, dest)
+                except EnvironmentError as e:
+                    if getattr(e, 'errno', None) != errno.ENOENT:
+                        # Failing to rename the old format will at worst leave a
+                        # harmless orphan, so log and ignore the error
+                        import traceback
+                        traceback.print_exc()
 
         if (not getattr(stream, 'name', False) or not samefile(dest, stream.name)):
             with lopen(dest, 'wb') as f:

src/calibre/db/cache.py

@@ -1201,7 +1201,8 @@ class Cache(object):
             title = self._field_for('title', book_id, default_value=_('Unknown'))
             author = self._field_for('authors', book_id, default_value=(_('Unknown'),))[0]
             stream = stream_or_path if hasattr(stream_or_path, 'read') else lopen(stream_or_path, 'rb')
-            size, fname = self.backend.add_format(book_id, fmt, stream, title, author, path)
+
+            size, fname = self.backend.add_format(book_id, fmt, stream, title, author, path, name)
             del stream
 
             max_size = self.fields['formats'].table.update_fmt(book_id, fmt, fname, size, self.backend)

src/calibre/db/tests/add_remove.py

@@ -7,7 +7,7 @@ __license__   = 'GPL v3'
 __copyright__ = '2013, Kovid Goyal <kovid at kovidgoyal.net>'
 __docformat__ = 'restructuredtext en'
 
-import os
+import os, glob
 from io import BytesIO
 from tempfile import NamedTemporaryFile
 from datetime import timedelta
@@ -285,3 +285,19 @@ class AddRemoveTest(BaseTest):
         af(db.has_format(1, 'ORIGINAL_FMT1'))
         ae(set(fmts), set(db.formats(1, verify_formats=False)))
     # }}}
+
+    def test_format_orphan(self):  # {{{
+        ' Test that adding formats does not create orphans if the file name algorithm changes '
+        cache = self.init_cache()
+        path = cache.format_abspath(1, 'FMT1')
+        base, name = os.path.split(path)
+        prefix = 'mushroomxx'
+        os.rename(path, os.path.join(base, prefix + name))
+        cache.fields['formats'].table.fname_map[1]['FMT1'] = prefix + os.path.splitext(name)[0]
+        old = glob.glob(os.path.join(base, '*.fmt1'))
+        cache.add_format(1, 'FMT1', BytesIO(b'xxxx'), run_hooks=False)
+        new = glob.glob(os.path.join(base, '*.fmt1'))
+        self.assertNotEqual(old, new)
+        self.assertEqual(len(old), len(new))
+        self.assertNotIn(prefix, cache.fields['formats'].format_fname(1, 'FMT1'))
+    # }}}