Untokenize might not yield an identical source when non normalized-unicode is used.

[aroberge]

When I attempted to use token_utils with the unnormalized_unicode transformations form ideas, it did not work correctly. I suspect that, somewhere, some normalization was applied to some strings.