Re: Rework the way multixact truncations work

On Tue, Dec 01, 2015 at 05:07:15PM -0500, Robert Haas wrote:
> I think it's weird to back a commit out only to put a bunch of very similar
> stuff back in.

I agree with that. If the original patches and their replacements shared 95%
of diff lines in common, we wouldn't be having this conversation. These
replacements redo closer to 50% of the lines, so the patches are not very
similar by verbatim line comparison. Even so, let's stipulate that my
proposal is weird. I'd rather be weird than lose one of the benefits I
articulated upthread, let alone all of them.

My post-commit review of RLS woke me up to the problems of gradually finishing
work in-tree. By the time I started that review in 2015-06, the base RLS
feature already spanned twenty-two commits. (That count has since more than
doubled.) Reviewing each commit threatened to be wasteful, because I would
presumably find matters already fixed later. I tried to cherry-pick the
twenty-two commits onto a branch, hoping to review the overall diff as "git
diff master...squash-rls", but that yielded complex merge conflicts. Where
the conflicts were too much, I reviewed entire files instead. (Granted, no
matter how this thread ends, I do not expect an outcome that opaque.) Hackers
have been too reticent to revert and redo defect-laden commits. If doing that
is weird today, let it be normal.

> Even figuring out what you've actually changed here seems
> rather hard. I couldn't get github to give me a diff showing your
> changes vs. master.

If you, an expert in the 2015-09-26 commits, want to study the key changes I
made, I recommend perusing these outputs:

git remote add nmisch_github https://github.com/nmisch/postgresql.git
git fetch nmisch_github
git diff nmisch_github/mxt0-revert nmisch_github/mxt1-disk-independent
git log -p --reverse --no-merges nmisch_github/mxt2-cosmetic..nmisch_github/mxt3-main

For the overall diff vs. master that you sought:

git remote add community git://git.postgresql.org/git/postgresql.git
git remote add nmisch_github https://github.com/nmisch/postgresql.git
git fetch --multiple community nmisch_github
git diff community/master...nmisch_github/mxt4-rm-legacy

If anyone not an author or reviewer of the 2015-09-26 commits wishes to review
the work, don't read the above diffs; the intermediate states are
uninteresting. Read these four:

git remote add nmisch_github https://github.com/nmisch/postgresql.git
git fetch nmisch_github
git diff nmisch_github/mxt0-revert nmisch_github/mxt1-disk-independent
git diff nmisch_github/mxt1-disk-independent nmisch_github/mxt2-cosmetic
git diff nmisch_github/mxt2-cosmetic nmisch_github/mxt3-main
git diff nmisch_github/mxt3-main nmisch_github/mxt4-rm-legacy

Thanks,
nm