> From: Philipp Stephani <p.stephani2@gmail.com>
> Date: Fri, 22 Dec 2017 22:00:31 +0100
> Cc: Philipp Stephani <phst@google.com>
>
> * admin/merge-gnulib (GNULIB_MODULES): Add unistr modules.
>
> * lib/Makefile.in (.c.o): Fix output file for files in subdirectories.
> (${DEPDIR}/unistr, unistr/u8-check.o): Create missing deps directory.
>
> * src/json.c (json_check_utf8): New helper function.
> (lisp_to_json_toplevel_1, lisp_to_json): Use it.  To save a bit of
> time, check for invalid UTF-8 strings only after encountering an
> error, since Jansson already rejects them.
>
> * test/src/json-tests.el (json-serialize/invalid-unicode): Adapt
> expected error symbol.

I'm not sure I understand the need for this.  Emacs never does that
anywhere else, it just converts the invalid bytes to a special
character set.  An application that cares could then test for presence
of those bytes to see if anything like that happened.  What am I
missing?