On Thu, 21 Nov 2024 at 08:30, Guillaume Lelarge <guillaume@lelarge.info> wrote: > OK, I'm fine with this. v4 patch attached with one plan showing read, written, and dirtied buffers.
Today I spent more time polishing this patch. There were a few cases in the docs that displayed EXPLAIN ANALYZE output that you'd not adjusted to include the buffers output or adjusted to do BUFFERS OFF. I think I've got all these now. Tom went to some effort to fix some outdated EXPLAIN outputs for v17 in 5caa05749, so I think we owe it to him not to let these go out of date so soon after that change.
You're right and I completely forgot to check the whole documentation. I just looked at perform.sgml which was the obvious file for explain plans. Anyway, sorry about this, and thanks a lot for your work on this patch.
I also was thinking again about what Robert mentioned about auto_explain.log_buffers should now also be on by default. I'm less certain than him about this change. It seems like a separate consideration that we could apply many of the same arguments for the main change to. In any case, I extracted that change from the 0001 patch and put it in a 0002 patch as it doesn't seem like something that should be a sidenote in the commit message. I felt doing that increases the chances that it would be overlooked in the release notes.
I was very close to pushing 0001 today, but will hold off until tomorrow to see if anyone has final comments.
No more comments. I'm fine with both patches.
For 0002, I'd really like to see a bit more justification for it. For the record, I'm not against 0002, it's just that my personal arguments for wanting 0001 don't apply to 0002.
I guess consistency is the key word here. But I agree that 0001 is the one that's really important to me.