On 9/22/2026 12:01 PM, Daniel P. Berrangé wrote:
> On Tue, Sep 22, 2026 at 11:33:47AM -0700, Pierrick Bouvier wrote:
>> On 9/22/2026 11:29 AM, Daniel P. Berrangé wrote:
>>> On Tue, Sep 22, 2026 at 11:22:43AM -0700, Pierrick Bouvier wrote:
>>>> On 9/22/2026 11:04 AM, Daniel P. Berrangé wrote:
>>>>> On Tue, Sep 22, 2026 at 06:55:18PM +0100, Daniel P. Berrangé wrote:
>>>>>> On Mon, Sep 21, 2026 at 11:14:54PM +0000, Pierrick Bouvier wrote:
>>>>>>> We'll use this in next patch to switch FROM docker.io/library to
>>>>>>> docker.io/amd64. It would not make sense to add a new target to
>>>>>>> libvirt-ci for this, since it's just a variant of an existing target.
>>>>>>> Simply replace container registry with arch specific one.
>>>>>>
>>>>>> We shouldn't need to add new targets to lcitool for this,
>>>>>> rather it would be enhanced to include a list of arch
>>>>>> specific container images.
>>>>>>
>>>>>> eg in
>>>>>>
>>>>>> https://gitlab.com/libvirt/libvirt-ci/-/blob/master/lcitool/facts/targets/debian-13.yml?ref_type=heads
>>>>>>
>>>>>> We would extend:
>>>>>>
>>>>>>   containers:
>>>>>>     base: docker.io/library/debian:13-slim
>>>>>>
>>>>>> to allow for
>>>>>>
>>>>>>   containers:
>>>>>>     base: docker.io/library/debian:13-slim
>>>>>>     aarch64:  docker.io/amd64/debian:13-slim
>>>>>>     ...more...
>>>>>>
>>>>>> such that we default to the 'base' image name, unless
>>>>>> there is an arch specific name defined. That would be
>>>>>> a fairly quick extension to impl in lcitool.
>>>>>
>>>>> Quicker than I thought.... take a look at this, which
>>>>> I think will enable you to drop this patch and the
>>>>> next one and get the same end result:
>>>>>
>>>>>   https://gitlab.com/libvirt/libvirt-ci/-/merge_requests/588
>>>>>
>>>>
>>>> Very quick indeed, good job :)
>>>>
>>>> IIUC, when no host-arch is given, we keep on using base, but in case
>>>> user specifies -host-arch, in this case we use one of specialized image
>>>> (if available)?
>>>
>>> Actually it is indepedent of --host-arch.
>>>
>>> If --host-arch is omitted we default it to what uname() reports,
>>> and the result then goes into the container image lookup, which
>>> is optionally specialized per arch.
>>>
>>
>> It works for QEMU needs, but would that be a safe default for other
>> users of lcitool?
> 
> Yes, I think it ought to be OK.  If something unexpected crops up
> we'll deal with it as needed.
> 
>> Technically, you can write portable Dockerfiles between arch. Our
>> current debian-all-test is kind of an example of it (ignoring the change
>> this series does).
> 
> True, but lcitool isn't guaranteeing it can be done, as we allow
> output to filter per-arch, so there's always the risk of getting
> the libpmem situation - IIRC it hits with 'xen-devel' too as some
> arches lack Xen.
> 
> So realistically I think we would say lcitool container outputs
> are arch specific.
>

I'm good with this assumption. There are always small details that makes
it hard to have truly portable Dockerfiles.

I'll wait for PR to be merged on libvirt-ci, then update current series
with it for v2.

> With regards,
> Daniel

Thanks for the help on this,
Pierrick

Reply via email to