And that is what we did in our application before this discussion even started. Don't know what Petr is doing in his. I think it is more of an interesting, mostly academic, discussion about alternative techniques; probably a very low priority issue to the SA code base.
On Nov 26, 5:56 pm, Michael Bayer <[EMAIL PROTECTED]> wrote: > executemany() syntax is very efficient and I dont really understand > how the column/string thing is that much of an issue other than an > small inconvenience and a slight failure of the API to be > consistent...all you have to do is convert the dict keys to be > column.key. > > On Nov 26, 2008, at 5:52 PM, MikeCo wrote: > > > > > Oops, I stand corrected. seehttp://pastebin.com/fe4a38d6 > > > At least for SQLite, my loop solution is many times slower than the > > insert many syntax. I would be curious to see results run against > > different database engines. I don't have quick access to them right > > now. > > Still, unless there are very large numbers of inserts there should be > > little, if any, noticeable difference. > > > -- > > Mike > > > On Nov 26, 4:50 pm, MikeCo <[EMAIL PROTECTED]> wrote: > >> You probably don't want to do the inserts one by one because of the > >> commit overhead, or needing to rollback on failure of on insert. You > >> can still get multiple inserts in one transaction. Add this to the > >> example posted athttp://pastebin.com/fd0653b0tosee three inserts in > >> one transaction. > > >> print '****', '5. column attributes in .values()' > >> data = [{MyTable.c.col1:itm.x} for itm in items] > >> print '****', data > >> trans = conn.begin() # < - - start a transaction to wrap many > >> inserts > >> for d in data: > >> conn.execute(MyTable.insert().values(d)) > >> trans.commit() # < - - end transaction > > >> Wrapping the insert loop in a transaction should give you the same or > >> very nearly the same performance as executing the "insert many" > >> operation, and only adds 4 or 5 simple lines of code. The exception > >> might be if you have a large batch process inserting thousands or > >> tens > >> of thousands of records; in that case, you will need to design a > >> solid > >> commit strategy anyway. with or without SQLAlchemy. > > >> -- > >> Mike > > >> On Nov 26, 10:42 am, "Michael Bayer" <[EMAIL PROTECTED]> > >> wrote: > > >>> Petr Kobalíèek wrote: > > >>>> So, I can use that way in inserting one row, but can't when > >>>> inserting > >>>> multiple rows ? It is correct ? > > >>> you can only use string keys as the arguments to the execute() > >>> method. > >>> this applies to one row or many. columns as keys can be used for > >>> the > >>> values argument/generative method on an insert() or update() > >>> construct. > > >>> to be quite honest there was never an intention for columns as > >>> keys to not > >>> work properly when sent as keys as parameters to .execute(), there > >>> just > >>> seems to be a lack of testing in this case, so feel free to file a > >>> ticket > >>> for this. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "sqlalchemy" group. To post to this group, send email to sqlalchemy@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sqlalchemy?hl=en -~----------~----~----~----~------~----~------~--~---
