BSON prefixes every document and array with its size. to_bson() therefore
walked the value twice: once to compute every size, once to write. When
the output is the vector that to_bson(j) returns, the writer now puts a
placeholder in front of each document and array and writes its size there
once the document is complete, so one walk is enough.
Every key and value is still checked as the size pass checked it, in the
same order, so the same exception is thrown; the partial output is
discarded with the vector. If a length would not fit BSON's 32-bit size
fields, the writer drops what it has written and writes the document in
two passes, which reports that error before writing anything, as before.
The two passes remain for output adapters (streams, strings, vectors
passed by reference), for error_handler_t::replace and ignore (which
change sizes), and for integer types wider than 64 bits.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
Niels Lohmann
a624682c1d