Re: [PATCH nft] build: disable --with-unitdir by default

[Phil Sutter <phil@nwl.cc>]

On Wed, Aug 27, 2025 at 04:46:15PM +0200, Pablo Neira Ayuso wrote:
> Hi,
> 
> Cc'ing Phil, Jan.
> 
> Excuse me my terse proposal description.
> 
> Extension: This is an alternative patch to disable --with-unitdir by
> default, to address distcheck issue.
> 
> I wonder also if this is a more conservative approach, this should
> integrate more seamlessly into existing pipelines while allowing
> distributors to opt-in to use this.
> 
> But maybe I'm worrying too much and it is just fine to change defaults
> for downstream packagers.

I think both is fine, packagers will know what to pass to configure and
what not.

Though I think you should also pass the --with-unitdir=<something>
option during distcheck builds, otherwise this patch inadvertently
disables the sanity check which sparked just this discussion in the
first place. :)

As Jan's patch shows, there is AM_DISTCHECK_CONFIGURE_FLAGS for that.

Thinking about the original issue, I wonder if the pkg-config method of
finding the unit install location is actually problematic: My
development build script calls configure with --prefix="$PWD/install"
and runs 'make install' as regular user. The only reason why I didn't
complain yet is probably because I don't have systemd installed and thus
'pkg-config systemd' fails.

So, I'm tempted to ask a fundamental question: Shouldn't --prefix be
applied to all installed files? I would say yes, but simply prefixing
the pkg-config-returned path by $prefix seems wrong, too as it is
definitely not where the unit file is expected to be. Also, --prefix
defaults to /usr/local and Fedora for instance passes --prefix=/usr when
building SRPMs.

OTOH Fedora (Rawhide) returns /usr/lib/systemd/system when requesting
systemdsystemunitdir. So maybe prefix $prefix only if it doesn't start
with $prefix already? (Seems silly, though.)

Cheers, Phil