Re: [libvirt] [PATCHv4 13/15] build: switch --with-qemu default from yes to check
On Thu, Jul 19, 2018 at 07:38:15PM -0400, John Ferlan wrote:
On 07/18/2018 10:44 AM, Ján Tomko wrote:
> Unless explicitly requested, enable the QEMU driver
> only if the Jansson library is present.
>
> Signed-off-by: Ján Tomko <jtomko(a)redhat.com>
> ---
> m4/virt-driver-qemu.m4 | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
Perhaps it's obvious for someone else, but I think there some sort of
dependency missing.
Starting with this patch I found that my with-qemu
"went away".
I have:
jansson.x86_64 2.11-1.fc28
@fedora
But not:
jansson-devel x86_64 2.11-1.fc28 fedora
15 k
If I explicitly add --with-jansson onto the command line, then I get:
checking for JANSSON... no
configure: error: You must install the jansson >= 2.5 pkg-config module
to compile libvirt
error: configure failed
If I then install jansson-devel, the build succeeds.
Honestly I think we need to be much more in your face in this instance -
something isn't quite right and it eventually leads to some really
strange results because nothing in/for qemu is built, but you are left
with old build bits in your tree. Eventually something fails.
There was not that much discussion about it:
https://www.redhat.com/archives/libvir-list/2018-May/msg01321.html
I do not oppose reverting this bit and failing by default if we don't
have a JSON library (as Andrea mentioned, more drivers might possibly
require a working JSON implementation).
However after dropping old QEMU support, building QEMU driver without
JSON makes no sense. (There was some logic in virt-yajl.m4 that tried
to figure out whether lack of yajl should be fatal based on the version
of the QEMU executable present on the system).
This whole build/config system is a generally a mystery to me, so I
have
zero suggestions. Ironically if I had to build from source, I'd know
from libvirt.spec that jansson-devel was required, although there's no
>= version check so that probably should be fixed.
Right, the specfile still has:
%define min_rhel 6
Suffice to say digging into the config.log trying to figure why one's
QEMU is disabled is not an enjoyable or easy experience.
Yeah, yeah, we're developers we're supposed to be smart, we get what we
get... I'll bet some qemu devel will hit this some day and wonder how to
actually build libvirt with qemu because it's not obvious and it "used
to be" a default=yes.
I expected that people surprised by the lack of QEMU driver would pass
--with-qemu, which should have the proper error.
Jano
John
Off to go drown the rest of my frustration ;-)
Attachments:
- signature.asc (application/pgp-signature — 833 bytes)