[HACKERS] floating point representation

[Hiroshi Inoue]

Hi all, I have a question of PostgreSQL's floating point representation. create table t (fl1 float4, fl2 float4, fl3 float4); insert into t values (1.234567, 1.23457, 1.23457); select * from t; fl1 | fl2 | fl3 ---------+---------+--------- 1.23457 | 1.23457 | 1.23457 (1 row) select * from t where fl1=fl2; fl1 | fl2 | fl3 -----+-----+----- (0 rows) select * from t where t where fl2=fl3; fl1 | fl2 | fl3 ---------+---------+--------- 1.23457 | 1.23457 | 1.23457 (1 row) OK, fl1 != fl2 and fl2 == fl3 but copy t to stdout; 1.23457 1.23457 1.23457 The output of pg_dump is same. Then after restoring from the pg_dump output, we would get a tuple such that fl1==fl2==fl3. Is it reasonable ? In addtion this makes a client library like ODBC driver very unhappy with the handlig of floating point data. For example, once a floating point data like fl1(1.234567) was stored, MS-Access couldn't update the tuple any more. Is there a way to change the precision of floating point representation from clients ? Regards, Hiroshi Inoue