For all logging statements, shouldn't it make sens to have the id of the object being indexed ? I've not checked how indexers work so i'm not sure if this id makes sense and is available at this level, but IMHO having warnings and error messages in the logs which we cannot easily link to a known entity (an origin, a revision, etc.) is pretty frustrating.

unless i'm mistaken, the match var is not used, so why not

for line in lines:
  if self._re_spec_new.match*line):
    break

Another solution is using the dropwhile iterator (itertools), something like:

lines = dropwhile(lambda x: not self._re_spec_new.match(x), lines.split('\n'))

and move the 'emptyness' detection at the end of the function.

I don't like too much this kind of if isinstance(str)... What are the expected types for s? Can it be something else than str or None?
In which case I prefer if s is not None:

why forbid tuples or iterators?

When writing tests for this kind of feature, please also add limit cases. How about invalid spec files? what if the description (or any other field) is a (literal) multi line string/list/dict?

One test for the most basic and "friendly" data is not enough.