---- Darren Duncan <[EMAIL PROTECTED]> wrote: > At 4:38 PM -0700 4/5/07, Darren Duncan wrote: > >To get this to work would basically involve having additional > >journal files, with the original one being for the parent-most > >transaction, and with an additional one for each transaction level, > >or some such arrangement; the extra ones could have file names like > >the original but numeric suffixes indicating the transaction level. > ><snip> > >Note that to maintain backwards compatability, the original journal > >file will still need all pre-change pages written to it too, but > >intermediate-level files don't necessarily need this, or it can be > >done, as the implementer wishes. > > Actually, I will clarify that any "additional" journal files do not > need to be on disk ... said pages could just be in RAM, unless there > are too many of them and they need to spill to disk ... only the > parent-most transaction actually needs a journal file on disk, under > the same circumstances that the current journal needs to be on disk > ... before writes to the main SQLite db file are being made. > > My more general point is that to support child transactions, the > pager layer would need to be updated to represent N ordered layers of > state for changed pages, where N is the number of nested transactions > (an ordinary SQL statement counting as 1), rather than just > representing 2 layers, current and original. > > I believe that this can be done fairly easily, and should be done, > assuming the pager is already well designed and doesn't make certain > assumptions. > > Moreover, I don't believe that the addition of this feature should > make the SQLite code base much larger, and it shouldn't cause much of > a performance hit, if any at all. > > -- Darren Duncan > > ----------------------------------------------------------------------------- > To unsubscribe, send email to [EMAIL PROTECTED] > ----------------------------------------------------------------------------- >
It sounds like this feature can be added under a compile time switch. Do we have any volunteers out there? Ray ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
