Re: Surrogate pairs in UTF-8

Martijn van Oosterhout <kleptog(at)svana(dot)org> writes:
> On Fri, Jan 16, 2015 at 08:16:47AM -0600, Dave Rosckes wrote:
>> I have written a test program using postgres that creates a string with a
>> surrogate pair. I then insert that string into a varchar property in a
>> table.
>>
>> I then execute a select statement to pull the string out. But when I
>> evaluate the string the lead char of the pair is correct, but the following
>> pair value is mangled. I run this exact same code using DB2 and it works
>> just fine.
>>
>> Is this a postgres limitation, or is there a specific way surrogate pairs
>> need to be handled?

> Sounds odd. Can you provide actual queries showing the problem (and
> server version).

Surrogate pairs are illegal in UTF-8, per its specification at
http://www.faqs.org/rfcs/rfc3629.html
You're supposed to encode the underlying code point, not a surrogate pair
(those are a UTF-16ism). So if what you passed in was actually a
surrogate pair, it should have failed encoding validity check, or possibly
have gotten converted to the underlying single Unicode character depending
on exactly what code path is involved.

regards, tom lane