On Thu, Jan 19, 2012 at 4:07 PM, Andrew Dunstan <andrew@dunslane.net> wrote:
> On 01/19/2012 03:49 PM, Robert Haas wrote:
>> In other words, let's decree that when the database encoding isn't
>> UTF-8, *escaping* of non-ASCII characters doesn't work. But
>> *unescaped* non-ASCII characters should still work just fine.
>
> The spec only allows unescaped Unicode chars (and for our purposes that
> means UTF8). An unescaped non-ASCII character in, say, ISO-8859-1 will
> result in something that's not legal JSON. See
> <http://www.ietf.org/rfc/rfc4627.txt?number=4627> section 3.
I understand. I'm proposing that we not care. In other words, if the
server encoding is UTF-8, it'll really be JSON. But if the server
encoding is something else, it'll be almost-JSON. And specifically,
the \uXXXX syntax won't work, and there might be some non-Unicode
characters in there. If that's not the behavior you want, then use
UTF-8.
It seems pretty clear that we're going to have to make some trade-off
to handle non-UTF8 encodings, and I think what I'm suggesting is a lot
less painful than disabling high-bit characters altogether. If we do
that, then what happens if a user runs EXPLAIN (FORMAT JSON) and his
column label has a non-Unicode character in there? Should we say, oh,
sorry, you can't explain that in JSON format? That is mighty
unfriendly, and probably mighty complicated and expensive to figure
out, too. We *do not support* mixing encodings in the same database,
and if we make it the job of this patch to fix that problem, we're
going to be in the same place for 9.2 that we have been for the last
several releases: nowhere.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company