Re: Differential Code Coverage report for Postgres

Hi,

A little belatedly, I've made this work and the report can be seen here:

https://coverage.postgresql.org/diff/

It refreshes once a day.

The script I used is attached -- mostly Nazir's, though I removed
unnecessary installation steps for our underlying OS and changed some
other things.

I'm open to suggestions for improvement. Examples just off the top of
my head:
- script output should be saved somewhere sensible and maybe
published on the website
- currently, worktrees are removed and recreated on each run.
Seems unnecessary.
- baseline is the tip of the previous stable branch. Seems odd to me.
Shouldn't it be the branch point from the current master?
- should we have one report per live branch, instead of just master?
- date bins are 1, 7, 30, 360. Is that good?
- we just run "meson test", but the cross-branch upgrade tests aren't
run, and those increase coverage significantly; need to find a way
to run those too. Can we use the olddump/oldinstall mechanism in
src/bin/pg_upgrade/t/002_pg_upgrade.pl?
- What other tests can we run? Should we set PG_TESTS_EXTRA?
- Two tests seem to be failing, but I didn't investigate why.
(Scrollback wasn't kept by tmux; see the first point above.)
- Would it make sense to create a buildfarm animal and use the
buildfarm client to run the tests instead of manual meson test?

--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/