Re: [PATCH-for-5.2 2/2] gitlab-ci: Avoid running the EDK2 job when not necessary

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

On 1/18/21 9:30 AM, Thomas Huth wrote:
> On 17/01/2021 19.48, Philippe Mathieu-Daudé wrote:
>> On 11/11/20 10:18 AM, Philippe Mathieu-Daudé wrote:
>>> On 11/10/20 4:35 PM, Daniel P. Berrangé wrote:
>>>> On Tue, Nov 10, 2020 at 01:16:06PM +0100, Philippe Mathieu-Daudé wrote:
>>>>> The EDK2 jobs use the 'changes' keyword, which "makes it
>>>>> possible to define if a job should be created based on files
>>>>> modified by a Git push event." (see [1]). This keyword comes
>>>>> with a warning:
>>>>>
>>>>>    Caution:
>>>>>
>>>>>      In pipelines with sources other than the three above
>>>>>      changes can’t determine if a given file is new or old
>>>>>      and always returns true."
>>>>>
>>>>> In commit 922febe2af we moved the YAML config file from the
>>>>> repository root directory to the .gitlab-ci.d/ directory.
>>>>>
>>>>> We didn't respect the previous warning and disabled the
>>>>> 'changes' filter rule, as the files are now in a (directory)
>>>>> three above the YAML config file.
>>>>
>>>> This description is a bit wierd. I don't see how the location
>>>> in the directory tree has any relevance here.
>>>>
>>>> IIUC the caution docs quoted above are referring to what triggered
>>>> the pipeline. They're saying that if the trigger was not a "branch",
>>>> "merge request", or "external pull request", then the "changes" rule
>>>> always evaluates true.
>>>>
>>>> The "branch" source us a bit wierd though, as I'm not seeing
>>>> how gitlab figures out which commits are "new" to the pipeline
>>>> and thus whether the files were modified or not.
>>>>
>>>> Strangely qemu-project/qemu CI for master seems to be behaving
>>>> correctly and skipping the jobs.
>>>
>>> What about this one?
>>>
>>> https://gitlab.com/berrange/qemu/-/jobs/827459510
>>
>> Ping?
> 
> Daniel's comment sounded like there are some changes required here, at
> least to improve the commit message? So I was expecting a v2 here.
> Please clarify Daniel's concerns, then I can either pick up this version
> here or the v2 (in case you send one).

Sorry, this "ping" was for Daniel. I'm not sure how to continue,
so shared an example in his namespace of "incorrect behavior".

Meanwhile I'm tempted to put those jobs in Manual mode (2 months now),
until we settle this with Daniel.

Thanks,

Phil.