Re: [PATCH] Teach pg_waldump to extract FPIs from the WAL

On Fri, Nov 04, 2022 at 09:16:29AM -0500, David Christensen wrote:
> On Nov 4, 2022, at 9:02 AM, Justin Pryzby <pryzby(at)telsasoft(dot)com> wrote:
> > On Fri, Nov 04, 2022 at 11:52:59AM +0900, Ian Lawrence Barwick wrote:
> >> 2022年5月3日(火) 8:45 David Christensen <david(dot)christensen(at)crunchydata(dot)com>:
> >>>
> >>> ...and pushing a couple fixups pointed out by cfbot, so here's v4.
> >>
> >> cfbot reports the patch no longer applies [1]. As CommitFest 2022-11 is
> >> currently underway, this would be an excellent time to update the patch.
> >
> > More important than needing to be rebased, the patch has never passed
> > its current tests on windows.
> >
> > As I recall, that's due to relying on "cp". And "rsync", which
> > shouldn't be assumed to exist by regression tests).
>
> I will work on supporting the windows compatibility here. Is there some list of guidelines for what you can and can’t use? I don’t have a windows machine available to develop on.

I think a lot (most?) developers here don't have a windows environment
available, so now have been using cirrusci's tests to verify. If you
haven't used cirrusci directly (not via cfbot) before, start at:
src/tools/ci/README

There's not much assumed about the build environment, and not much more
assumed about the test environment. Most of the portability is handled
by using C and perl. I think there's even no assumption that "tar" is
available (except maybe for building releases). This patch should avoid
relying on tools that aren't already required.

As a practical matter, cfbot needs to pass, not only to demonstrate that
the patch consistently passes tests, but also because if the patch were
merged while it failed tests in cfbot, it would cause every other patch
to start to fail, too.

> Was it failing on windows? I was attempting to skip it as I recall.

I don't see anything about skipping, and cirrus's logs from 2
commitfests ago were pruned. I looked at this patch earlier this year,
but never got around to replacing the calls to rsync and cp.

--
Justin