On Thu, Mar 17, 2011 at 8:05 PM, Richard Hipp <d...@sqlite.org> wrote:
> On Thu, Mar 17, 2011 at 12:30 AM, Igor Tandetnik <itandet...@mvps.org > >wrote: > > > Afriza N. Arief <afriza...@gmail.com> wrote: > > > What is the correct way of detecting existing transaction? > > > > sqlite3_get_autocommit > > > > Correct. Or, you could just always use SAVEPOINT which works the same as > BEGIN if you are not already in a transaction. > > My understanding is that using [SAVEPOINT's automatic] BEGIN is more prone to deadlocks than BEGIN IMMEDIATE if there are many read-then-write transactions; and SAVEPOINT only allows automatic BEGIN DEFERRED. This is because multiple transactions may hold the SHARED locks and multiple of them may then tries to get RESERVED in the same transaction. I am not using WAL and want to rely on sqlite3_busy_timeout() instead of using more complex error handling. This understanding leads me to always use BEGIN IMMEDIATE for read-then-maybe-write transactions as well as read-then-always-write transactions. Is my understanding correct? Regards, Afriza N. Arief _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
