On 6 Sep 2013, at 4:08pm, Simon Slavin <slav...@bigfraud.org> wrote:
> Right. As I posted in my message that I had tested, this isn't being done > correctly. A conflict isn't a conflict until the write, and the write > doesn't happen until the COMMIT. Therefore conflict testing needs to happen > at the commit, for /all/ the changes in the transaction, whether they're all > the consequence of one UPDATE or even of separate commands within the same > transaction. On further thought, I may be wrong about that. If it was really true then if you explicitly declared a transaction with BEGIN ... END a SQLITE_CONSTRAINT result due to violations of UNIQUE could only be returned at the END. You wouldn't be able to tell which command within the transaction caused the conflict. I don't know whether that's the way things really should work. Or whether this is something the people who thought up SQL specified. Simon. _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
