1. I have removed the dependency on count(DISTINCT ...) by using an anti-join instead. The code is much clearer now IMHO.
It is much cleaner! I
2. That instead of selecting operators at execution time we save which operators to use in pg_constraint.
This is a clever approach. If you have any updates on this please let me know.
I am still reviewing your changes. I have split your changes from my work to be able to isolate the changes and review them carefully. And to help others review the changes.
Changelist:
- v11 (compatible with current master 2021, 03, 20, commit e835e89a0fd267871e7fbddc39ad00ee3d0cb55c) * rebase
* split andreas and joel's work
On Tue, Mar 16, 2021 at 1:27 AM Andreas Karlsson <andreas@proxel.se> wrote:
On 3/15/21 4:29 PM, Mark Rofail wrote: > Actually, your fix led me in the right way, the issue was how windows > handle pointers.
Hi,
I have started working on a partially new strategy for the second patch. The ideas are:
1. I have removed the dependency on count(DISTINCT ...) by using an anti-join instead (this was implemented by Joel Jacobson with cleanup and finishing touches by me). The code is much clearer now IMHO.
2. That instead of selecting operators at execution time we save which operators to use in pg_constraint. This is heavily a work in progress in my attached patches. I am not 100% convinced this is the right way to go but I plan to work some on this this weekend to see how ti will work out.
Another thing I will look into is you gin patch. While you fixed it for simple scalar types which fit into the Datum type I wonder if we do not also need to copy types which are too large to fit into a Datum but I have not investigated yet which memory context the datum passed to ginqueryarrayextract() is allocated in.
