Re: [PATCH v14 00/10] rpm: 4.20.1 -> 6.0.2

["Mathieu Dubois-Briand" <mathieu.dubois-briand@bootlin.com>]

On Sat Aug 15, 2026 at 4:16 AM CEST, Sumanth Gavini wrote:
> On 8/13/2026 6:12 AM, Mathieu Dubois-Briand wrote:
>> CAUTION: This email comes from a non Wind River email account!
>> Do not click links or open attachments unless you recognize the sender and know the content is safe.
>> 
>> On Thu Aug 13, 2026 at 1:48 AM CEST, Sumanth Gavini wrote:
>>> * V14:
>>>      - Dropped "rpm: filter /usr/bin/pkg-config from RPM auto-generated Requires"
>>>      (pkgconfigdeps.sh patch). Per Ross Burton's review, instead of suppressing
>>>      the /usr/bin/pkg-config emission, added:
>>>        RPROVIDES:${PN} += "pkgconfig /usr/bin/pkg-config"
>>>      to the pkgconf recipe so it satisfies the auto-generated Requires without
>>>      modifying the upstream rpm script.
>>>
>>>     - Removed /usr/bin/pkg-config from nativesdk-sdk-provides-dummy.bb.
>>>       RPM 6's internal dependency generator auto-detects file-level Provides
>>>       from installed files, so nativesdk-pkgconf already provides
>>>       /usr/bin/pkg-config via the wrapper script it installs. Keeping the
>>>       dummy entry caused a DNF conflict during do_populate_sdk:
>>>         nativesdk-sdk-provides-dummy conflicts with /usr/bin/pkg-config
>>>         provided by nativesdk-pkgconf
>>>
>>>     - Added "libarchive: Disable RPM filter auto-bidding" (0005 patch).
>>>       With both the RPM filter and RPM format reader registered, the filter
>>>       intercepted RPM files first and failed to parse RPM 6 packages
>>>       (debuginfod scanning still failed after the 0004 patch). Disabling the
>>>       filter's automatic bidding lets the format reader handle RPM files,
>>>       while archive_read_append_filter(ARCHIVE_FILTER_RPM) still works
>>>       explicitly, keeping test_read_append_rpm_filter passing.
>>>
>>>     - Updated 0004-Convert-RPM-reader-into-a-proper-format.patch to include
>>>       the test_read_format_huge_rpm.c changes from upstream PR #2846. This
>>>       test now correctly exercises the format reader instead of the filter.
>>>
>>>     - Updated patches 7,9 and 10 for the latest master changes; no code
>>>       changes; removed upstream modifications
>>>
>>>     Tests verified:
>>>       - oe-selftest -r debuginfod.Debuginfod.test_debuginfod_qemu
>>>       - oe-selftest -r debuginfod.Debuginfod.test_debuginfod_native
>>>       - oe-selftest -r sdk.SDKTests.test_sdk_manifests
>>>       - oe-selftest -r sdk.SDKTests.test_sdk_runqemu
>>>       - bitbake core-image-ptest-libarchive:do_testimage
>>>       - bitbake core-image-sato -c populate_sdk
>>>       - bitbake core-image-sato-sdk:do_rootfs
>>>
>> 
>> Hi Sumanth,
>> 
>> We again have some pkgconfig errors. I suspect it's the same issue as
>> for the previous series: the generated recipe signatures are the same as
>> before, so we are re-using old data from the sstate-cache.
>> 
>> https://autobuilder.yoctoproject.org/valkyrie/#/builders/3/builds/4439
>> https://autobuilder.yoctoproject.org/valkyrie/#/builders/8/builds/4413
>> 
>> I also note a new failure:
>> 
>> Error:
>>   Problem 1: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by acl-dev-2.4.0-r0.cortexa57 from oe-repo
>>   Problem 2: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by eudev-dev-3.2.14-r0.cortexa57 from oe-repo
>>   Problem 3: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by kmod-dev-34.2-r0.cortexa57 from oe-repo
>>   Problem 4: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by libz-dev-1.3.2-r0.cortexa57 from oe-repo
>>   Problem 5: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by openssl-dev-3.5.7-r0.cortexa57 from oe-repo
>>   Problem 6: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by util-linux-dev-2.42.2-r0.cortexa57 from oe-repo
>>   Problem 7: conflicting requests
>>    - nothing provides /usr/bin/pkg-config needed by xz-dev-5.8.3-r0.cortexa57 from oe-repo
>> (try to add '--skip-broken' to skip uninstallable packages or '--setopt=optional_metadata_types=filelists' to load additional filelists metadata)
>> 
>> https://autobuilder.yoctoproject.org/valkyrie/#/builders/36/builds/4371
>> https://autobuilder.yoctoproject.org/valkyrie/#/builders/61/builds/4269
>> https://autobuilder.yoctoproject.org/valkyrie/#/builders/73/builds/4286
>> https://autobuilder.yoctoproject.org/valkyrie/#/builders/110/builds/942
>> 
>> As usual, the detailed configuration can be found in the "Write config"
>> step.
>> 
>> Thanks,
>> Mathieu
>> 
>> --
>> Mathieu Dubois-Briand, Bootlin
>> Embedded Linux and Kernel engineering
>> https://bootlin.com
>> 
>
> Hi Mathieu,
>
> Looks like this is an sstate reuse issue. The pkgconfigdeps.sh fix is in 
> the rpm recipe (used by both rpm and rpm-native via BBCLASSEXTEND), and
> do_package_write_rpm does declare a proper dependency on 
> rpm-native:do_populate_sysroot. Despite that, I'm seeing the shared 
> sstate cache and hash-equivalence server serve prebuilt RPMs for zlib, 
> glib-2.0, etc. that predate the fix.
>
> I confirmed locally: disabling sstate mirrors and rebuilding from 
> scratch fixes it, no missing pkgconfig() Provides errors. Could you 
> retry with sstate disabled to confirm on your end?
>
> Thanks,

Hi Sumanth,

I did that already with the previous series, as stated in my comments
there. Yes, disabling sstate or changing PR values mask the issue.

Thanks,
Mathieu


-- 
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com