Re: [OE-core] [PATCH] archiver.bbclass: Fix archiver interaction with kernel recipes

[Phil Reid <preid@electromag.com.au>]

On 10/10/2024 15:17, Richard Purdie wrote:
> On Thu, 2024-10-10 at 13:25 +0800, Phil Reid via lists.openembedded.org
> wrote:
>> Changes to the logic of is_work_shared where made in
>> commit: 5fbb4ca8da4f4f1ea426275c45634802dcb5a575
>> "archiver.bbclass: Improve work-shared checking"
>>
>> The resuled in a change of the logic (simplifed here) from:
>>    inherits(gcc-source) or inherits(kernel) or (inherits(kernelsrc)
>> and srcin(work-shared))
>> to just
>>    srcin(work-shared)
>>
>> With INHERIT += "archiver" in the local.conf and a kernel recipe that
>> uses
>> KERNEL_PACKAGE_NAME. When KERNEL_PACKAGE_NAME is defined the kernel
>> source is not placed into work-shared, but the archiver ends up
>> deleting the source folder in the work dir, and the build
>> subsequently fails.
>>
>> Restore the previous logic while also mainting the referenced commits
>> intent
>> to consider all recipes that use work-shared. Logic is now
>>    inherits(gcc-source) or inherits(kernel) or srcin(work-shared)
>>
>> Signed-off-by: Phil Reid <preid@electromag.com.au>
>> ---
>>   meta/classes/archiver.bbclass | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
> 
> You're effectively reverting that commit whilst changing the logic so
> it looks slightly different.
> 
> Was there a problem with gcc-source archiving as I notice the patch
> adds gcc-source back too?
To be honest I didn't test without the gcc-source, I reverted the commit first after
finding that commit to be the cause and modified it so it consider it always considered
work-shared path and called it done and the old logic conditions.

> 
> The commit message isn't quite accurate as gcc-source isn't something
> which gets inherited.
Fair enough.

> 
> Does this only happen with kernel recipes which use
> KERNEL_PACKAGE_NAME? That might be the key missing detail which would
> allow us to reproduce the failure.
> 

I haven't seen the problem on a virtual/kernel.
When KERNEL_PACKAGE_NAME is specified the kernel source isn't put into work-shared.


How would you like to see this fixed?


-- 
Regards
Phil Reid
  
ElectroMagnetic Imaging Technology Pty Ltd
Development of Geophysical Instrumentation & Software
www.electromag.com.au
  
23 Junction Parade, Midland WA 6056, AUSTRALIA
Ph: +61 8 9250 8100
Fax: +61 8 9250 7100

Email: preid@electromag.com.au