On 2018/03/23 3:42, Pavan Deolasee wrote: > A slightly improved version attached. Apart from doc cleanup based on > earlier feedback, fixed one assertion failure based on Rahila's report. > This was happening when target relation is referenced in the source > subquery. Fixed that and added a test case to test that situation. > > Rebased on current master.
I tried these patches (applied 0002 on top of 0001). When applying 0002, I got some apply errors:
The next patch would create the file src/test/isolation/expected/merge-delete.out, which already exists! Assume -R? [n]
I managed to apply it by ignoring the errors, but couldn't get make check to pass; attached regressions.diffs if you want to take a look.
Thanks. Are you sure you're using a clean repo? I suspect you'd a previous version of the patch applied and hence the apply errors now. I also suspect that you may have made a mistake while resolving the conflicts while applying the patch (since a file at the same path existed). The failures also seem related to past version of the patch.
I just checked with a freshly checked out repo and the patches apply correctly on the current master and regression passes too. http://commitfest.cputube.org/ also reported success overnight.
Btw, is 0001 redundant with the latest patch on ON CONFLICT DO UPDATE thread? Can I apply just 0002 on top of that patch? So, I tried that -- that is, skipped your 0001 and instead applied ON CONFLICT DO UPDATE patch, and then applied your 0002.
Yes. I should probably rebase my patch on your v9 or just include the relevant changes in the MERGE patch itself to avoid any dependency right now. Will check.