Thanks for the patch, I left more comments on the review page.

IMHO it would be better to list the differences in a bullet list and expand later, rather than having a section for the parser and one for the generator.
AFAIU the differences are:
 * top-level scalar values are accepted/generated;
 * inf and nan are accepted/generated;
 * unicode strings (rather than utf-8) are produced/consumed;
 * duplicate keys are accepted, and the only the last one is used;

You can then add examples and explain "workarounds", either inline or after the list (I don't think it's necessary to add the snippets you posted in the last message though).


> Sidenote: Is there any way to get notified of these reviews?

In theory you should get notifications, in practice it doesn't always work.  We are still working on make it better.