On 12 June 2014 17:43, Patrick Ohly <[email protected]> wrote:

> How would that work when building in OBS (see question above)?

I'll describe what happens in production - plus some idea of how it
could be extended to devs.

1) Someone triggers a new build in OBS (so far this someone is a
release engineer, but it is possible to extend this also to devs)

2) the new build triggers the creation (in jenkins) of one or more
images, depending on the targets configured.

3) if said images can be built successfully, they are flashed and
tested (in a remote testing farm, which is part of the building
infrastructure).
Currently this happens with a fixed set of test cases, but it is
possible to make it conditional. For example to what comes from the
.spec file(s) of the package(s) that are new in the specific build.
And there could be some extra directive (like FULL_SMOKE_TEST /
UNIT_TEST / FLASH_ONLY)

4) results are published back to the OBS project that triggered the building

> What you describe sounds like it is targeting developers who have a
> development work station and a separate device running some Tizen image.

Currently we are targeting RE, so what we describe is necessarily
biased in that direction. It can be extended, though.

> In OBS, there is only a Tizen build chroot, but not a fully booted Tizen
> image.

That's why we have Jenkins and MIC to produce images.

I think the natural, evolutionary path would be to support those devs
who are also package maintainer.

So that they would be better equipped to ensure the quality of what
they accept. This could probably still be supported by the same
infrastructure, maybe with some ad-hoc increase of build/test
bandwith.

But for individual developers, they already have means of creating
packages and images, so they could just use the flashing tool and then
manually start the execution of the selected test cases.

Why waiting in a queue on a remote farm during tight development cycles?
Better use local resources and do a remote submission when reasonably
sure if its quality.

Even in such scenario, developers would use exactly the same test
framework and test cases.

Of course, depending on what one is working on, a real device might be
needed. Alternatively, one could use the emulator.

But yes, it was simpler for us to target first the real HW and avoid
the extra complexity dealing with possible quirks introduced by the
emulator scenario.

-- 
cheers, igor
_______________________________________________
Dev mailing list
[email protected]
https://lists.tizen.org/listinfo/dev

Reply via email to