Re: Adding CI to our tree

On Mon, Feb 28, 2022 at 02:58:02PM -0600, Justin Pryzby wrote:
> I still think that if "Build Docs" is a separate cirrus task, it should rebuild
> docs on every CI run, even if they haven't changed, for any patch that touches
> docs/. It'll be confusing if cfbot shows 5 green circles and 4 of them were
> built 1 day ago, and 1 was built 3 weeks ago. Docs are the task that runs
> quickest, so I don't think it's worth doing anything special there (especially
> without understanding the behavior of changesInclude()).
>
> Also, to allow users to view the built HTML docs, cfbot would need to 1) keep
> track of previous CI runs; and 2) logic to handle "skipped" CI runs, to allow
> showing artifacts from the previous run. If it's not already done, I think the
> first half is a good idea on its own. But the 2nd part doesn't seem desirable.

Maybe changesInclude() could work if we use this URL (from cirrus'
documentation), which uses the artifacts from the last successful build.
https://api.cirrus-ci.com/v1/artifact/github/justinpryzby/postgres/Documentation/html_docs/html_docs/00-doc.html?branch=citest-cirrus2

That requires knowing the file being modified, so we'd have to generate an
index of changed files - which I've started doing here.

> However, I realized that we can filter on cfbot with either of these:
> | $CIRRUS_CHANGE_TITLE =~ '^\[CF...'
> | git log -1 |grep '^Author: Commitfest Bot <cfbot(at)cputube(dot)org>'
> If we can assume that cfbot will continue submitting branches as a single
> patch, this resolves the question of a "base branch", for cfbot.

I don't know what you think of that idea, but I think I want to amend my
proposal: show HTML and coverage artifacts for HEAD~1, unless set otherwise by
an environment var. Today, that'd do the right thing for cfbot, and also for
any 1-patch commits.

> These patches implement that idea, and make "code coverage" and "HTML diffs"
> stuff only run for cfbot commits. This still needs another round of testing,
> though.

The patch was missing a file due to an issue while rebasing - oops.

BTW (regarding the last patch), I just noticed that -Og optimization can cause
warnings with gcc-4.8.5-39.el7.x86_64.

be-fsstubs.c: In function 'be_lo_export':
be-fsstubs.c:522:24: warning: 'fd' may be used uninitialized in this function [-Wmaybe-uninitialized]
if (CloseTransientFile(fd) != 0)
^
trigger.c: In function 'ExecCallTriggerFunc':
trigger.c:2400:2: warning: 'result' may be used uninitialized in this function [-Wmaybe-uninitialized]
return (HeapTuple) DatumGetPointer(result);
^
xml.c: In function 'xml_pstrdup_and_free':
xml.c:1205:2: warning: 'result' may be used uninitialized in this function [-Wmaybe-uninitialized]
return result;

--
Justin