Alfranio Correia wrote:
> Hi all,
> Find some comments in-line.
[ snip ]
> >> 3) N T...: the beginning part is non-transactional, followed by
> >> transactional
> >> statements.
> > In this case, the beginning part of non-transactional statements should
> > be logged separately and written to binlog directly in all modes. There
> > is nothing unsafe while binlogging the first non-transactional part. And
> > then the reset part T... will be either a T...T or TN..., which are
> > handled by 1) or 4).
> I really don't like this idea of handling this case differently.
> However, this is my personal opinion.
I think we have to do this to be compatible with the old behavior as
much as possible.