[OE-core] [PATCH v2 0/7] run-postinsts refactoring
Mark Hatle
mark.hatle at windriver.com
Thu Jun 6 18:22:23 UTC 2013
On 6/6/13 1:09 PM, Laurentiu Palcu wrote:
>
>
> On 06/06/2013 07:43 PM, Mark Hatle wrote:
>> Comments below....
>>
>> On 6/6/13 2:41 AM, Laurentiu Palcu wrote:
>>> (cover letter only)
>>>
>>> Changes in v2:
>>> * addressed an RPM issue that happened on AB. Interestingly, it happened only on
>>> certain distros. Our internal AB (running on Ubuntu server 12.04), didn't show
>>> any issues. So, what was the problem? We use a scriptlet wrapper to run the
>>> pre/post install scriptlets. What I missed in the previous patchset was that I
>>> didn't use the scriptlet wrapper when running the pre/post remove scriptlets
>>> and the context in which they ran was not correct. Hence, the update-rc.d used
>>> was the host's...
>>> * addresses Ross's request to move run-postinsts recipe out of dpkg directory,
>>> since it's generic now;
>>>
>>> Paul, Mark would you please review the RPM changes please?
>>>
>>> Thanks,
>>> Laurentiu
>>>
>>> Changes in v1:
>>> Hi all,
>>>
>>> My work at #4484 revealed that the package managers deb/ipk/rpm handle removal
>>> of PM meta-data (when the PM is not deployed) and the delayed postinstalls
>>> execution in their own way. Currently we have:
>>> * run-postinsts (for deb/ipk) and rpm-postinsts (for rpm) for running the delayed
>>> postinstalls on target when the PM is not part of the image;
>>> * opkg removes the meta-data and some uneeded packages whilst deb/rpm remove
>>> only the meta-data;
>>> * both opkg and deb have no way of running the delayed postinstalls on target,
>>> if package-management is not part of IMAGE_FEATURES, because the meta-data
>>> was deleted (and it would have been useless anyway since the PM is not installed);
>>>
>>> That being said, this patchset tries to create a more unified solution for all
>>> PM backends.
>>>
>>> rpm-postinsts recipe is replaced by run-postinsts but I'll keep it in oe-core
>>> for a while (a couple of weeks maybe), just in case the current solution does not
>>> satisfy the RPM users (I doubt it though).
>>>
>>> I did tests with core-image-minimal using sysvinit/systemd and with/without delayed
>>> postinstalls. The results are as expected.
>>>
>>> Thanks,
>>> Laurentiu
>>>
>>> The following changes since commit a62aed41f2d8f874f7ae24d0e5be5dbc66ea2199:
>>>
>>> lrzsz: check locale.h in configure (2013-06-04 15:55:46 +0100)
>>>
>>> are available in the git repository at:
>>>
>>> git://git.yoctoproject.org/poky-contrib lpalcu/run_postinsts_refactor_v2
>>> http://git.yoctoproject.org/cgit.cgi/poky-contrib/log/?h=lpalcu/run_postinsts_refactor_v2
>>>
>>> Laurentiu Palcu (7):
>>> rootfs_*.bbclass: add some helper functions
>>> image.bbclass: remove unneeded files from the image
>>> core-image-minimal: do not remove packaging files in the rootfs
>>> postprocess
>>> run-postinsts: make it generic
>>> rootfs_rpm.bbclass: switch to using run-postinsts
>>> package_rpm.bbclass: handle pre/post remove scriptlets
>>> rpm: replace rpm-postinsts dependency with run-postinsts
>>>
>>> meta/classes/image.bbclass | 33 ++++++++++++++
>>> meta/classes/package_rpm.bbclass | 16 ++++---
>>> meta/classes/rootfs_deb.bbclass | 23 +++++++++-
>>> meta/classes/rootfs_ipk.bbclass | 39 ++++++++--------
>>> meta/classes/rootfs_rpm.bbclass | 29 ++++++++++--
>>> meta/recipes-core/images/core-image-minimal.bb | 2 -
>>> .../dpkg/run-postinsts/run-postinsts | 36 ---------------
>>> .../dpkg/run-postinsts/run-postinsts.awk | 30 -------------
>>> meta/recipes-devtools/rpm/rpm_5.4.9.bb | 2 +-
>>> .../run-postinsts/run-postinsts/run-postinsts | 47 ++++++++++++++++++++
>>> .../{dpkg => run-postinsts}/run-postinsts_1.0.bb | 7 ++-
>>> 11 files changed, 159 insertions(+), 105 deletions(-)
>>> delete mode 100755 meta/recipes-devtools/dpkg/run-postinsts/run-postinsts
>>> delete mode 100644 meta/recipes-devtools/dpkg/run-postinsts/run-postinsts.awk
>>> create mode 100755 meta/recipes-devtools/run-postinsts/run-postinsts/run-postinsts
>>> rename meta/recipes-devtools/{dpkg => run-postinsts}/run-postinsts_1.0.bb (68%)
>>>
>>
>>> +rootfs_remove_unneeded () {
>>> + if ${@base_contains("IMAGE_FEATURES", "package-management", "false", "true", d)}; then
>>> + if [ -z "$(delayed_postinsts)" ]; then
>>> + # All packages were successfully configured.
>>> + # update-rc.d, base-passwd, run-postinsts are no further use, remove them now
>>> + remove_run_postinsts=false
>>> + if [ -e ${IMAGE_ROOTFS}${sysconfdir}/init.d/run-postinsts ]; then
>>> + remove_run_postinsts=true
>>> + fi
>>> + rootfs_remove_packages update-rc.d base-passwd ${ROOTFS_BOOTSTRAP_INSTALL}
>>> +
>>> + # Need to remove rc.d files for run-postinsts by hand since opkg won't
>>> + # call postrm scripts in offline root mode.
>>> + if $remove_run_postinsts; then
>>> + update-rc.d -f -r ${IMAGE_ROOTFS} run-postinsts remove
>>> + fi
>>
>> The above runs unconditionally for all package types. Will this cause an issue
>> w/ RPM or deb? (I don't think it will, but I'm slightly worried about RPM and
>> it's removal scripts.)
> I believe I missed to remove the opkg from the comment since I pasted
> this piece of code from rootfs_ipk.bbclass. However, during my tests, I
> saw no issues with rpm/deb. Is there anything in particular you think I
> should check for RPM to rule out your concerns?
Simply that the update-rc.d will be run by the RPM rule, and rerun in the above.
As long as that doesn't cause a problem, I'm fine with the change.
--Mark
> Thanks,
> Laurentiu
>>
>>> + else
>>> + # Some packages were not successfully configured, save them only
>>> + # if we have run-postinsts script present. Otherwise, they're
>>> + # useless
>>> + if [ -e ${IMAGE_ROOTFS}${sysconfdir}/init.d/run-postinsts ]; then
>>> + save_postinsts
>>> + fi
>>> + fi
>>> +
>>> + # Since no package manager is present in the image the metadata is not needed
>>> + remove_packaging_data_files
>>> + fi
>>> +}
>>> +
>>
>> Everything else looks fine.
>>
>> Reviewed-by: Mark Hatle <mark.hatle at windriver.com>
>>
More information about the Openembedded-core
mailing list