Georg Brandl answered this already on doc-sig:
On Wed, Mar 4, 2009 at 11:33, Jeffrey C. Jacobs
> The two reStructuredText files:
> This paragraph has a very funny **indent** after that word, right?
> his paragraph has a very funny **indent
> after that word, right?**
> are theoretically different. The first puts strong emphasis only on the
> word **indent**, which is followed by exactly 4 spaces, where as the
> other puts strong emphasis on the entire expression "indent after that
> word, right?", where there is a line feed between "indent" and "after".
> However, when publish_string is called to output the tree for both of
> these expressions, they both return:
> <document source="<string>">
> This paragraph has a very funny
> after that word, right?
> which is not different. As far as I can tell, the internal node structure
> is correct, it's just when the node structure is displayed in string form,
> the default function of publish_string. Since this output is a
> serialization of the node structure, it seems that the output to
> publish_string should not be ambiguous in terms of what it truly
> represents. Or, is there a better way to represent the internal doc tree
> unambiguously as a string?