I conservatively made it a link to /usr/bin/true as the glibc's package %post still references it, and if that somehow gets executed, it's simplest to just make it a no-op.

any reason to do if ! test -f rather than:

if test -f ${x}; then
    ln -srf /usr/bin/true ${x}
fi

It's how I write most of my loops in C to avoid excessive indentation, but I agree it obscures things a bit here.

Updated :arrow_up:

are we waiting for https://src.fedoraproject.org/rpms/glibc/pull-request/4# to be merged or are we doing this because it isn't merged yet?

That one is unlikely to be merged as is soon, so we should merge this. If glibc ever drops the post upgrade bit, then this will just be a safe no-op.

LGTM

This breaks any ability to verify the rpms inside of a running Atomic Host. It makes me kinda sad. What real problems does it fix?

rpm -Va is (mostly) obsoleted by ostree fsck.

This saves us about 800k of disk space. The implementation source is also really ugly code that IMO has no business running as root on system updates.

This breaks any ability to verify the rpms inside of a running Atomic Host. It makes me kinda sad. What real problems does it fix?

I didn't think about that. If https://src.fedoraproject.org/rpms/glibc/pull-request/4# gets merged then we can remove this and rpm -V will work again, so that's nice.

I'm guessing neither of you have actually tried running rpm -Va today on an rpm-ostree host...this PR is just one of hundreds of issues. The timestamp issue is the biggest, but also the /var handling we impose, etc. Really in the end what we need to do is intercept some rpm commands like -Va and redirect them.