Обсуждение: "no snapshot has been set" error

Hi all,
  I encountered a problem when I'm implementing my user-defined type. 
My  user-defined type is defined as

CREATE TYPE X
(   INTERNALLENGTH = 10   INPUT = X_in,   OUTPUT = X_out
);

In my X_in() function, I want to access other table to do some 
bookkeeping stuff. this table keeps track of the number of calls to 
X_in(). However, while SPI_connect() returns SPI_OK_CONNECT, whenever I 
execute SPI_exec( "SELECT count FROM XTABLE", 0), postgres stops 
executing and issues "Error: no snapshot has been set".

Does anyone here knows what happens?  Am I not allowed to use SPI 
functions in this situation?

Thank you for your advice and suggestions in advance.

Ruey-Lung Hsiao

Ruey-Lung Hsiao <rlhsiao@gmail.com> writes:
> In my X_in() function, I want to access other table to do some 
> bookkeeping stuff. this table keeps track of the number of calls to 
> X_in(). However, while SPI_connect() returns SPI_OK_CONNECT, whenever I 
> execute SPI_exec( "SELECT count FROM XTABLE", 0), postgres stops 
> executing and issues "Error: no snapshot has been set".

Are you sure you've diagnosed this correctly?  AFAICS it's not possible
to get into SPI without having set a snapshot.  Depending on what PG
version you're using (which you did not say, naughty naughty) there are
paths in interactive query entry that might try to execute datatype
input functions before setting the query snapshot.  But I don't believe
it can happen down inside SPI.  I'd be interested to see a complete
counterexample.
        regards, tom lane