[yocto] [PATCH V4 03/10] utils.py: remove obsolete dependencies

[Robert Yang]

Hi Paul,

On 06/23/2017 11:03 AM, Robert Yang wrote:
>
>
> On 06/22/2017 07:54 PM, Paul Eggleton wrote:
>> On Thursday, 22 June 2017 12:47:31 PM CEST Robert Yang wrote:
>>> On 06/21/2017 05:06 PM, Paul Eggleton wrote:
>>>> So I do see the need to clear out old dependency values when your list of
>>>> dependencies is coming entirely from LAYERDEPENDS, however if you have only
>>>> manually added dependencies and LAYERDEPENDS is not complete (as is often the
>>>> case in the OE layer index), will this automatically remove those? If so I
>>>
>>> The dependencies will be removed if it is not in conf/layer.conf.
>>>
>>>> think we're going to have to come up with some mechanism to avoid that.
>>>> Probably the easiest one would be to avoid touching the dependencies unless
>>>> LAYERDEPENDS actually changes. Of course if that means we're going to store
>>>> the previous LAYERDEPENDS value somewhere to compare to, we'd need to avoid it
>>>> triggering initially when none of those values are set.
>>>
>>> Why not fix conf/layer.conf rather than fix it from database ? If
>>> conf/layer.conf is wrong, we would meet problems elsewhere.
>>
>> Up until now it hasn't caused any issues other than not getting an error
>> up front from bitbake when a dependencies are missing, so it can easily go
>> unnoticed (particularly when users follow manual instructions or use existing
>> configurations rather than an automated tool that pays attention to
>> LAYERDEPENDS).
>>
>>> The database
>>> should obey local.conf when mismatches, otherwise, this would confuse
>>> users a lot when they want to update LAYERDEPENDS.
>>
>> As I indicated, I agree that if LAYERDEPENDS gets updated then the
>> dependencies in the database should get updated. My concern is updating
>> them if LAYERDEPENDS has not changed and it simply happens to have
>> a difference to what is in the database (most importantly if it has less
>> dependencies specified).
>>
>>> I'm glad to fix oe-classic's LAYERDEPENDS issues if they are wrong.
>>
>> Well, it's not so much OE-Classic as every single layer in the index,
>> maintained by a number of different maintainers. I agree in the long term we
>> do want people to fix their dependencies. The trouble is we don't
>> even know the size of the problem at the moment, and if it's widespread
>> then we'll end up dropping dependencies potentially leading to errors during
>> parsing.
>>
>> How about this - could we start with a setting in the configuration that
>> specifies that dependencies should be updated automatically, and if not
>> set just gives a warning if the dependencies are not the same? We can
>> then decide how to handle it when we see the warnings we're getting in
>> the OE index.
>
> Sounds good to me, I will fix it in V5.

After more thinking, I will use:
# Remove layer dependencies if it is not in conf/layer.conf
REMOVE_LAYER_DEPENDENCIES = False

Currently it does can add dependencies from conf/layer.conf, but no remove,
so use REMOVE_LAYER_DEPENDENCIES makes it more clear.

// Robert

>
> // Robert
>
>>
>> Cheers,
>> Paul
>>