Re: [Buildroot] [PATCH] utils/docker-run: allow explicitly setting of docker/podman/...

[Arnout Vandecappelle via buildroot <buildroot@buildroot.org>]

On 17/02/2025 22:53, Yann E. MORIN wrote:
> Arnout, All,
> 
> On 2025-02-17 21:54 +0100, Arnout Vandecappelle spake thusly:
>> On 16/02/2025 18:03, Yann E. MORIN wrote:
>>> Currently, our docker-run helper will decide on its own whether it
>>> should use docker (preferred) or podman (fallback). s introduced in
>>> 9a629f5129d6 (utils/docker-run: allow running with Podman). In case
>>> both are installed, it is not possible to exercise the podman case.
>>   To be honest, if you have both installed, then you probably want to use
>> podman because most likely "docker" is just an alias for podman...
> 
> That is probably a false assertion you are making above: podman
> impersonating docker is an additional package, at least in Fedora.
> And indeed, I have both podman and docker installed, and docker is
> definitely not podman.

  Hm, perhaps I should do the same as you and install both... CI uses docker, I 
use podman, and sometimes they don't behave the same (because docker runs as 
root). In the beginning my muscle memory was still sometimes accidentally typing 
"docker ..." but now I do the right thing anyway.


>>> Allow the user to force the one to use, by setting the BR2_DOCKER
>>> environment variable. If that is set and does not exist, exit with
>>> an explicit error message (rather than rely on the failure when
>>> eventually exec-ing the specified command).
>>
>>   Unfortunately, in commit 8aad67f1575f1d8a9c90ccd70302e9a6823cf794,
>> _someone_ already introduced the BR2_DOCKER environment variable in brmake
>> to tell it to run 'make' inside the container... Its use is not
>> incompatible, but it means that if you set BR2_DOCKER=podman in your
>> .profile, brmake will _always_ run in the container...
> 
> Meh. Use of unbuffer was one of the stupidest things I did. Indeed, when
> using brmake, one is not very concerned about seeing logs as soon as
> they are emitted, so one can very well accept that the '>>>' messages
> be slightly delayed. After all, they are only bufferred for so much (8K
> in glibc, that goes by relatively fast with all the output of a clasical
> build...)
> 
> I think it would have been much more elegant to actually allow use of
> "./utils/docker-run ./utils/brmake", as this is a construct that is
> already advertised to run our checks, "docker-run make check-package",
> or even just basically "./utils/docker-run make"...
> 
> I would wager that "BR2_DOCKER=y ./utils/brmake" is not in wide use.
> Would that be acceptable to revert 8aad67f1575f1 and drop use of
> unbuffer? I've just checked, and that works (but of course, any
> still-bufferred output could get lost on Ctrl-C, but that's probably
> a non-issue while using brmake...)

  Indeed, this feature was only introduced in 2024.11 (and very late in the 
cycle), and it wasn't documented other than in utils/readme.txt, so I think it's 
OK to drop it.

  Myself, I anyway always have to look up the exact wording of the environment 
variable, so I wouldn't miss it!

  Regards,
  Arnout

_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot