On 3/10/25 5:02 AM, Daniel P. Berrangé wrote:
On Sat, Mar 08, 2025 at 05:02:32PM +0900, Akihiko Odaki wrote:
> Builds for Linux without libnl is broken since commit
> 582f0966f9b9e2148d8887d072364e2a91aed000 so make libnl mandatory for
> Linux.
>
> Another option is to fix such builds, but it can bring more problems
> than benefits. libnl is available on all supported distributions so the
> only practical situation to lack libnl is that a developer does not have
> a development package of libnl installed. Enabling builds in such a
> situation make developers use builds divergent from CI and end-user
> scenarios, which has some implications.
>
> Builds without libnl lack several features such as SR-IOV,
> bridge/tap, switchdev, macvlan, veth, and vlan. Some part of libvirt
> may expect these features are readily available on Linux, and the lack
> of libnl may cause a build failure like one mentioned earlier or runtime
> errors even harder to debug.
Those features are only relevant if doing a daemon build for certain
virt drivers. For a client only build, or certain virt drivers, it is
valid to omit libnl while building on Linux, so we should not make
this mandatory.
Ah, right. I hadn't though of client-only builds the other day when I
agreed with the suggestion to make it mandatory. Seems like it would be
uncommon for someone to do a Linux build for just client, but I suppose
if we can allow for it we may as well.
In that case, I guess the comments in my first reply to the "util: Check
libnl function availability" patch apply.