On 23 Nov 2013, at 11:35pm, Darren Duncan <darren@darrenduncan.net> wrote:

> On 2013.11.23 7:20 AM, Simon Slavin wrote:
>> Had the person who devised SQL thought it through, he'd have thought up savepoints instead of transactions and we wouldn't use transactions at all.
> 
> This is an interesting proposal, and makes a lot of sense to me, especially given that savepoints today don't have the precondition of a "transaction" being active to use them, so on their own "savepoint" is like a generalization of a "transaction". -- Darren Duncan

My guess is that the engine would treat the outmost level of savepoint specially: that the outmost level of savepoint is equivalent to a transaction.

I sometimes work in a computer language (not available to the public, sorry) which allows a set of tasks to be divided up by criteria that the programmer can assign, or distributed among different processors as soon as one becomes free.  These things are conventional and have been done previously.  But this language can do it with 'if' tasks.  You can parcel up a test into two parts (e.g. one proof for even values, another for odd values; one for current records, another for each archive) and the test automatically terminates when the first answer of 'not true' is returned.  And the programming structure reminds me of how SQL does SAVEPOINTs.

Simon.
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users