I believe that there are many good points to have CRC facilities "built
int", and I failed to detect any arguments against it. In my domain (the
medical domain) we simply can't use data without "proof" of integrity
("proof" as in highest possible level of confidence within reasonable
effort)
Therefore, I propose defining new data types like "CRC32", "CRC64",
"RIPEMD", whatever (rather than pluggable arbitrary CRCs).
Similar as with the SERIAL data type, the CRC data type would generate
automatically a trigger function that calculates a CRC across a tuple
(omitting the CRC property of course, and maybe the OID as well) before each
update and store it in itself.
Is there anything wrong with this idea?
Can somebody help me by pointing me into the right direction to implement
it? (The person who implemeted the SERIAL data type maybe ?)
Regards,
Horst