Re: [PATCH] Include error log of failing task in output

[Alexander Smirnov <asmirnov@ilbers.de>]

On 03/02/2018 05:42 PM, Jan Kiszka wrote:
> On 2018-03-02 14:57, Alexander Smirnov wrote:
>>
>>
>> On 03/02/2018 04:24 PM, Jan Kiszka wrote:
>>> On 2018-03-02 14:04, Alexander Smirnov wrote:
>>>>
>>>>
>>>> On 03/02/2018 03:46 PM, Jan Kiszka wrote:
>>>>> On 2018-03-02 13:43, Alexander Smirnov wrote:
>>>>>>
>>>>>>
>>>>>> On 02/26/2018 04:41 PM, Jan Kiszka wrote:
>>>>>>> From: Jan Kiszka <jan.kiszka@siemens.com>
>>>>>>>
>>>>>>> Particularly helpful in CI environment, but it also saves many manual
>>>>>>> log dumps.
>>>>>>>
>>>>>>> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
>>>>>>> ---
>>>>>>>      meta/conf/isar-bitbake.conf | 2 ++
>>>>>>>      1 file changed, 2 insertions(+)
>>>>>>>
>>>>>>> diff --git a/meta/conf/isar-bitbake.conf
>>>>>>> b/meta/conf/isar-bitbake.conf
>>>>>>> index b49386c..8a1d86b 100644
>>>>>>> --- a/meta/conf/isar-bitbake.conf
>>>>>>> +++ b/meta/conf/isar-bitbake.conf
>>>>>>> @@ -42,6 +42,8 @@ BB_STAMP_POLICY ?= "full"
>>>>>>>        BB_NUMBER_THREADS ?= "${@bb.utils.cpu_count()}"
>>>>>>>      +BBINCLUDELOGS ??= "yes"
>>>>>>> +
>>>>>>
>>>>>> Is there any specific reason of using the weakest assignment here? The
>>>>>
>>>>> Because all tuneable confs are included after this statement.
>>>>
>>>> In my understanding, tuneable confs should not contain "?" marks in
>>>> assignment, because they specify concrete configuration. :-) That's what
>>>> I see, for example, in local.conf file.
>>>>
>>>>>
>>>>>> whole file contains "?=" only, for me this looks enough here too.
>>>>>
>>>>> That's likely a bug to be fixed separated.
>>>>
>>>> What is your overall policy for assignment in this case?
>>>
>>> Upstream: Look at oe-core's bitbake.conf. It seems to do weak defaults
>>> pretty consistently for stuff that might be set via ?= in other confs.
>>>
>>
>> Yeah I saw it, but who could overwrite BBINCLUDELOGS? For me this option
>> has the same level as BB_NUMBER_THREADS or PARALLEL_MAKE. So IMHO the
>> only local.conf file is the place to overwrite it.
>>
>> I've looked into OE/Yocto bitbake.conf, and I think there is the
>> following logic that in my opinion makes sense:
>>
>>   - Global build system settings are mostly defined using "?=", because
>> they should be overwritten in some global file like local.conf only.
>> Tuning this parameter in machine/*.conf is definitely bad idea.
>>
>>   - Settings, related to produced results, like DISTRO, MACHINE, LDFLAGS
>> etc. are defined using "??=", because such things could be overloaded
>> from machine/distro config files, from generic recipes etc.
>>
>> So I still propose to use "?=" here.
> 
> 
> Is there a well defined ordering when multiple ?= follow each other?
> Which one wins, the first or the last?
> 

The first one. But user should not use "?=" for such variables like: 
BBINCLUDELOGS, BB_NUMBER_THREADS etc. If you overwrite default system 
settings, you completely understand what you are doing and hard 
assignment should be used.

> Often you have the desire to provide a default in some included config
> in case some other include does not define a final value. We are at the
> top level here, so we should step back from such things and use a weak
> default.

Don't really understand the usecase of cascading exactly BBINCLUDELOGS 
several times. This option doesn't affect build content, so the only one 
place to overwrite it - is your local.conf. So there are no other 
included configs for now, and I believe in future also.

Anyway, I could stay with this.

Alex