GitLab

Build Chromium in a separate project and install in Stage 2.

This reverts commit b0bf06c3.

Chromium, Firefox, and xmonad (+ all the Haskell stuff) are the most
time consuming part of the Stage 1 build.  They are also a place where a
lot of build failures happen.  Let those failures happen as early as
possible so that retries get back to it more quickly.

CI is jargon.

Let GHC be pulled in as a normal dependency of xmonad and other Haskell
packages.

The build was successful with '-j4', but it is not worth chasing for the
complication added to the Puppet configuration.  Will keep an eye on
QEMU changelogs for improvements to qemu-user forking/threading.

Just trying random stuff at this point.  The fact is this works outside
of a container environment, so I'm just looking at limitations put in
place by the container.

Don't know if this will help, but I'm out of ideas.  During Python
installation, a number of forks of:

/usr/bin/qemu-aarch64 ./python -E -Wi -OO
/var/tmp/portage/dev-lang/python-3.8.6-r1/image/usr/lib/python3.8/compileall.py
-j16 -d /usr/lib/python3.8 -f -x
bad_coding|badsyntax|site-packages|lib2to3/tests/data
/var/tmp/portage/dev-lang/python-3.8.6-r1/image/usr/lib/python3.8

are spawned and deadlock in

futex(0x550435f000, FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, 0, NULL,
FUTEX_BITSET_MATCH_ANY <unfinished ...>

Chromium is failing to compile reporting:

obj/components/feedback/libfeedback.a: error adding symbols: malformed archive

which https://github.com/OSSystems/meta-browser/issues/194 has led me to
believe is actually due to the max open file limit.  Checked
`/proc/<buildah-chroot-pid>/limits` inside the build container and, sure
enough, "1024."

Puppet was attempting to manage the package before eix-update had run.

The distcc bug is from Chromium being the first to compile after a bunch
of binary package installations and it doing a compilation check in
pkg_setup.  See: https://bugs.gentoo.org/581880.

This reverts commit 840890af.

Otherwise, somehow a state file gets set up owned by root and Portage,
notably Chromium, fails to use distcc.

Make FACTER_build a build argument.  It looks nicer and it doesn't hurt
to think of it as just another parameter to the build.

See also: stage0@9f790786

Parameterize the GitLab CI script and add the profile change to the
Containerfile.

Puppet does weird calculations with distcc that don't work when the CI
host is also a distcc server.

There is no need to pass build options from the CI host because the
algorithm Puppet uses to make /etc/portage/make.conf works just as well
inside the container.