PR #5531 fixed the UTF-8-validation gap described in #5529, but it was
merged onto the still-unmerged bson-sizes branch rather than develop, so
develop was left with the original bug for all affected formats. A
follow-up comment on #5529 reproduced this on develop and additionally
found that UBJSON (and, by the same code path, BJData) has the identical
gap, undocumented.
Port the same fix directly onto develop: extract the UTF-8 DFA decoder
out of serializer<>::decode() into a shared detail::decode()/is_valid_utf8()
in string_utils.hpp, and call it from binary_reader::get_string() - the
single choke point shared by all five binary readers - so malformed text
strings are rejected at decode time (parse_error.113) instead of only
failing later on dump() (type_error.316). Byte/binary payloads are
unaffected. Add matching decode-time tests for CBOR, MessagePack, BSON,
UBJSON, and BJData, and document the new behavior on all five binary
format pages (the two UBJSON/BJData pages didn't get this note in #5531).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017sdieJCn6BHxzRMaXP49sP
Claude