[bitbake-devel] [PATCH 2/2] bitbake: build.py: check dependendent task for addtask

[Robert Yang]

Hi RP,

After more investigations, these warnings are caused by recrdeptask,
the rm_work calls:

addtask after 'deps + recrdeptask' before do_build

Then we will get the warnings, and we can divide these warnings into 3 kinds:

1) deltask only removes task from 'deps', but doesn't remove it from
    'recrdeptask', so there are warnings for native recipes like dependent
    task do_package_write_rpm, although native.bbclass has called
    "deltask do_package_write_rpm', such a recrdeptask doesn't make any sense
    for native recipes. But we can't make deltask remove recrdeptask since we may
    need keep recrdeptask sometimes, for example, the image recipes don't need
    do_package_write_rpm, but their do_rootfs tasks require all the dependent
    rpms are done. I think that we need a command such as delrecrdeptask to fix
    it, but I'm not sure whether it is worth or not.

2) base.bbclass has set 'do_build[recrdeptask] += "do_deploy"', but do_deploy
    doesn't exist for most of the recipes, so we get the warnings of not
    do_deploy task, I don't know why do_deploy should be a recrdeptask, and
    maybe we need set it in rootfs_xxx.bbclass or image.bbclass, or set it
    conditionally in a func.

3) gcc-source doesn't have do_populate_sysroot or do_populate_lic, but
    do_build_without_rm_work always depends on them:

    bb.build.addtask('do_build_without_rm_work', '', ' '.join(deps), d)

    so we get warnings that gcc-source has no dependent task do_populate_sysroot
    or do_populate_lic. I think that we can make do_build_without_rm_work
    depend on them conditionally to fix the problem.

Any comments are appreciated.

// Robert


On 5/5/19 1:56 PM, Robert Yang wrote:
> 
> 
> On 5/2/19 3:42 AM, Richard Purdie wrote:
>> On Wed, 2019-05-01 at 18:48 +0200, Nicolas Dechesne wrote:
>>> On Wed, May 1, 2019 at 5:30 PM Burton, Ross <ross.burton at intel.com>
>>> wrote:
>>>> On Wed, 1 May 2019 at 11:51, Jacob Kroon <jacob.kroon at gmail.com>
>>>> wrote:
>>>>> I'm getting a lot of these warnings after removing tmp/ and
>>>>> rebuilding
>>>>> my image from sstate cache.
>>>>> FWIW I use rm_work.bbclass, and regularly use the
>>>>> sstate-cache-management,sh script to prune old cache.
>>>>
>>>> There's a patch I hacked up in master-next that shows where these
>>>> warnings are coming from.  Yes, this patch produces two warnings
>>>> per
>>>> recipe with rm_work enabled.
>>>
>>> argh... i hit the same issue and came up with the exact same
>>> conclusion.. i should have checked the mailing list earlier today ;)
>>>
>>> I am getting thousands of warnings like that with rm_work. I had
>>> modified bitbake so that it shows the 'task' name like Ross, in ().
>>
>> I suspect we're going to need to rethink that patch, and/or perhaps
>> rm_work...
> 
> 
> Sorry for the warnings, I just came back from the International Labour Day
> Holiday. Something is wrong with rm_work, I will send a patch to fix it.
> 
>>
>> I did have concerns about it but was assured it was widely tested. We
>> also probably need an rm_work test case on the autobuilder (need a bug
>> for that so we don't forget?).
> 
> Rather than add a testcae on autobuilder, how about add a testcase to
> oe-selftest, please ?
> 
> // Robert
> 
>>
>> Cheers,
>>
>> Richard
>>
>>
>>