Improve performance of update 6007

Hi,

The logic is wrong. Instead of GREATEST (OR), GREATEST (OR), GREATEST (OR) it should be AND, IMP, AND. IMP stands for material implication which can be implemented as OR(NOT(x), y).

Also, doesn't this spew out lots of errors for the failed insert? Which is wrong also, as it should be nid, $print_display, $print_display_comment, $print_display_urllist and not nid, 0, 0, 1.

Finally, there's already some code in 6.x-1.12 to try to reduce this. I would love to have code that's more performant AND PostgreSQL compatible, but it must start by being correct.

Thanks for providing a patch!

João Ventura

While we're waiting for the performance enhancement I'd like to have a fix committed for PostgreSQL to update without error since that's more critical than performance. Here is a patch to do just that.

I would expect this issue to be reset to needs work after that commit. :)

sammys: that fix was committed more than a month ago (#881870: last update_n doesn't work in postgres)

update 6007 barfs on any site with a moderate number of nodes due to an out-of-memory error. I think we need to use batch API to get this to complete reliably. Some code comments would be great - I can't even figure out what the code is trying to do.

Even when I hard-code
ini_set('memory_limit', '256M');
at the beginning of the update, another memory error when update.php tries to output the results due to the massive number of queries run. I had to switch to db_query instead of udpate_sql().

I am so sorry, everyone..

Looking at this again, I realized the operations are AND, OR, AND (not AND, IMP, AND) as I believed in #2.

That sequence of operations is very simple to implement in standard SQL that is valid in MySQL and PostgreSQL, and after some other fixes which dispense with the INSERT operation, the 6007 updates can become a lot simpler..

If anyone still cares, feel free to try the dev.

João