Andrew Dunstan <andrew@dunslane.net> writes:
> It's not clear to me how we should represent a unicode null. i.e. given
> a json of '["foo\u0000bar"]', I get that we'd store the element as
> 'foo\x00bar', but what is the result of
> (jsonb '["foo\u0000bar"')->>0
> It's defined to be text so we can't just shove a binary null in the
> middle of it. Do we throw an error?
Yes, that is what I was proposing upthread. Obviously, this needs some
thought to ensure that there's *something* useful you can do with a field
containing a nul, but we'd have little choice but to throw an error if
the user asks us to convert such a field to unescaped text.
I'd be a bit inclined to reject nuls in object field names even if we
allow them in field values, since just about everything you can usefully
do with a field name involves regarding it as text.
Another interesting implementation problem is what does indexing do with
such values --- ISTR there's an implicit conversion to C strings in there
too, at least in GIN indexes.
Anyway, there is a significant amount of work involved here, and there's
no way we're getting it done for 9.4.1, or probably 9.4.anything. I think
our only realistic choice right now is to throw error for \u0000 so that
we can preserve our options for doing something useful with it later.
regards, tom lane
