On 11/18/06, Ran <[EMAIL PROTECTED]> wrote:
The way the undo-redo is described in the wiki involves triggers to insert
the information of the change in each table to other table which logs the
changes. This will have a price in performance. It also complicates things
when triggers are already used for other things.
So I wonder if journals might be used to implement undo:
If I understand it correctly, for each transaction there is journal that
keeps the information so the transaction could be rolled back. If the
journals are kept somewhere could they be used to rollback _successful_
transactions?
Transactions let you "undo" whatever you have done since the transaction
started. You decide if it was "successful" or not and either commit
the transaction
to make it permanent or roll it back to undo it.
In the case of large numbers of insertions it's faster to put them
into a transaction
and commit them than to do them separately. Performance in most installations is
very quick since the database generally ends up in operating system cache.
I do recall the author of mysql writing "he had no intention of implementing
transaction since it was much slower and proper design eliminated the need
for them." Sqlite has a much lower code overhead than mysql so I imagine
it's just as fast or faster in most cases.
-----------------------------------------------------------------------------
To unsubscribe, send email to [EMAIL PROTECTED]
-----------------------------------------------------------------------------
