On 9/22/2026 10:35 AM, Daniel P. Berrangé wrote: > On Tue, Sep 22, 2026 at 10:16:27AM -0700, Pierrick Bouvier wrote: >> On 9/22/2026 9:54 AM, Daniel P. Berrangé wrote: >>> On Tue, Sep 22, 2026 at 09:15:02AM -0700, Pierrick Bouvier wrote: >>>> On 9/21/2026 4:14 PM, Pierrick Bouvier wrote: >>>>> latest update of libvirt-ci changed packages, but images were not >>>>> refreshed. >>>>> >>>>> Signed-off-by: Pierrick Bouvier <[email protected]> >>>>> --- >>>>> tests/docker/dockerfiles/debian.docker | 1 - >>>>> tests/docker/dockerfiles/ubuntu2404.docker | 1 - >>>>> 2 files changed, 2 deletions(-) >>>>> >>>>> diff --git a/tests/docker/dockerfiles/debian.docker >>>>> b/tests/docker/dockerfiles/debian.docker >>>>> index 77869f71546..2d03ff03c8d 100644 >>>>> --- a/tests/docker/dockerfiles/debian.docker >>>>> +++ b/tests/docker/dockerfiles/debian.docker >>>>> @@ -73,7 +73,6 @@ RUN export DEBIAN_FRONTEND=noninteractive && \ >>>>> libpcre2-dev \ >>>>> libpipewire-0.3-dev \ >>>>> libpixman-1-dev \ >>>>> - libpmem-dev \ >>>>> libpng-dev \ >>>>> libpulse-dev \ >>>>> librbd-dev \ >>>>> diff --git a/tests/docker/dockerfiles/ubuntu2404.docker >>>>> b/tests/docker/dockerfiles/ubuntu2404.docker >>>>> index bb8f4e4e53f..440699a5005 100644 >>>>> --- a/tests/docker/dockerfiles/ubuntu2404.docker >>>>> +++ b/tests/docker/dockerfiles/ubuntu2404.docker >>>>> @@ -73,7 +73,6 @@ RUN export DEBIAN_FRONTEND=noninteractive && \ >>>>> libpcre2-dev \ >>>>> libpipewire-0.3-dev \ >>>>> libpixman-1-dev \ >>>>> - libpmem-dev \ >>>>> libpng-dev \ >>>>> libpulse-dev \ >>>>> librbd-dev \ >>>> >>>> Dropping this patch, as it's incorrect. >>>> In libvirt-ci, libpmem-dev is a dependency of targets debian-13 and >>>> ubuntu2404, so not sure why it removes it here. Seems like a bug in >>>> libvirt-ci itself. >>> >>> Most likely you ran invoked lcitool from an aarch64 host OS, so it >>> filtered out libpmem-dev which is only valid on x86_64 targets. >>> The non-cross containers we have defined here are assumed to be >>> for x86_64 targets. >>> >>> We should change the tests/lcitool/refresh script in QEMU to pass >>> '--host-arch x86_64' to lcitool, rather than defaulting to the >>> developers' machine's arch. >>> >> >> Thanks for the sharing the reason, I'll add this option. >> >> Out of curiosity, what's the reason to differentiate generation based on >> host arch? >> Wouldn't that be better to be explicit if you want to generate an arch >> specific Dockerfile (and use FROM docker.io/arch/... instead of >> docker.io/library)? > > I'm not familiar with docker.io/arch/ but this all just grew up > organically over time, and since most usage was x86_64 host based > there was not much need to think about other scenarios. >
Basically, it's (official) docker images per arch, following docker platform naming: - docker.io/amd64 - docker.io/aarch64 https://github.com/docker-library/official-images#architectures-other-than-amd64 Advantage is that image are provided for a specific platform, which avoid some of the issue multi plaform images have. For instance, on aarch64 host: $ podman run --platform linux/amd64 docker.io/library/debian:latest true $ podman run docker.io/library/debian:latest true WARNING: image platform (linux/amd64) does not match the expected platform (linux/arm64) # podman/docker now select the amd64 version by default, which can break build of other containers. It's a design issue with docker cli IMHO. > If we want the tests/docker/dockerfiles to be portable to non-x86 > hosts, then we'd need to either have many dockerfiles there for > each arch, or drop packages that are arch specific from the > install list. > I'll add the --host-arch x86_64 option to lcitool as recommended, which should be enough for our need. > With regards, > Daniel Regards, Pierrick
