This is perfect for my use case, thank you! The .fodt file is 150K, vs 117K for the .odt file. But after changing a word less than 20 of the 1983 lines in the file have changed.

  $ ls -hl todo.fodt
  -rw-rw-r-- 1 dotancohen dotancohen 150K Apr  7 16:41 todo.fodt
  $ ls -hl todo.odt
  -rwxrwxr-x 1 dotancohen dotancohen 117K Apr  6 12:01 todo.odt
  $ wc -l todo.fodt
  1983 todo.fodt
  $ cp todo.fodt todo-edit-one-line.fodt
  $ libreoffice todo-edit-one-line.fodt
  $ diff todo.fodt todo-2.fodt | grep "+ " | wc -l
  19

Wow, thanks for the Flat OpenDocument file type! Such kind of files are much better to treat with git (i.e. human readable diffs, for instance). However, unfortunately, these files tend to get very large (200kB ODS vs. 7MB FODS). What a pity the XML itself is not less verbose.

I dunno if that will help. I know Dia lets you store the uncompressed XML, but I have seen it sometimes have a 100% diff due to the serialiser moving *everything* around when all that happened was that I rearranged half the blocks on the diagram.