Re: Adding CI to our tree

On Thu, Feb 03, 2022 at 11:57:18AM -0800, Andres Freund wrote:
> On 2022-02-02 21:58:28 -0600, Justin Pryzby wrote:
> > FYI: I've rebased these against your cirrus/windows changes.
>
> Did you put then on a dedicated branch, or only intermixed with other changes?

Yes it's intermixed (because I have too many branches, and because in this case
it's useful to show the doc builds and coverage artifacts).

> > A recent cirrus result is here; this has other stuff in the branch, so you can
> > see the artifacts with HTML docs and code coverage.
>
> I'm a bit worried about the increased storage and runtime overhead due to the
> docs changes. We probably can make it a good bit cheaper though.

If you mean overhead of additional disk operations, it shouldn't be an issue,
since the git clone uses shared references (not even hardlinks).

If you meant storage capacity, I'm only uploading the *changed* docs as
artifacts. The motivation being that it's a lot more convenient to look though
a single .html, and not hundreds.

> What's the idea behind
> #echo 'deb http://deb.debian.org/debian bullseye main' >>/etc/apt/sources.list
> That's already in sources.list, so I'm not sure what this shows?

At one point I thought I needed it - maybe all I needed was "apt-get update"..

> > 9d0f03d3450 wip: cirrus: code coverage
>
> I don't think it's good to just unconditionally reference the master branch
> here. It'll do bogus stuff once 15 is branched off. It works for cfbot, but
> not other uses.

That's only used for filtering changed files. It uses git diff --cherry-pick
postgres/master..., which would *try* to avoid showing anything which is also
in master.

> Is looking at the .c files in the change really a reliable predictor of where
> code coverage changes? I'm doubtful. Consider stuff like removing the last
> user of some infrastructure or such. Or adding the first.

You're right that it isn't particularly accurate, but it's a step forward if
lots of patches use it to check/improve coverge of new code.

In addition to the HTML generated for changed .c files, it's possible to create
a coverage.gcov output for everything, which could be retrieved to generate
full HTML locally. It's ~8MB (or 2MB after gzip).

> > bff64e8b998 cirrus: upload html docs as artifacts
> > 80f52c3b172 wip: only upload changed docs
>
> Similar to the above.

Do you mean it's not reliable ? This is looking at which .html have changed
(not sgml). Surely that's reliable ?

> > 654b6375401 wip: vcsregress: add alltaptests
>
> I assume this doesn't yet work to a meaningful degree? Last time I checked
> there were quite a few tests that needed to be invoked in a specific
> directory.

It works - tap_check() does chdir(). But it's slow, and maybe should try to
implement a check-world target. It currently fails in 027_stream_regress.pl,
although I keep hoping that it had been fixed...
https://cirrus-ci.com/task/6116235950686208

(BTW, I just realized that that commit should also remove the recoverycheck
call.)

--
Justin