Re: Teach pg_upgrade test to honor NO_TEMP_INSTALL

Thomas Munro <thomas(dot)munro(at)gmail(dot)com> writes:
> After these commits (and Tom's commit "Un-break pg_upgrade regression
> test."), cfbot broke:

> sh: 1: /usr/local/pgsql/bin/psql: not found

I can confirm that here: check-world passes as long as I've done
"make install" beforehand ... but of course that should not be
necessary. If I blow away the install tree, pg_upgrade's
check fails at

../../../src/test/regress/pg_regress --inputdir=. --bindir='/home/postgres/testversion/bin' --port=54464 --dlpath=. --max-concurrent-tests=20 --port=54464 --schedule=./parallel_schedule
(using postmaster on /tmp/pg_upgrade_check-Nitf3h, port 54464)
============== dropping database "regression" ==============
sh: /home/postgres/testversion/bin/psql: No such file or directory
command failed: "/home/postgres/testversion/bin/psql" -X -c "DROP DATABASE IF EXISTS \"regression\"" "postgres"

pg_regress is being told the wrong --bindir, ie
the final install location not the temp install.

(More generally, should we rearrange the buildfarm test
sequence so it doesn't run "make install" till after the
tests that aren't supposed to require an installed tree?)

regards, tom lane