Re: [OE-core] [PATCH 1/6] package_rpm.bbclass: Handle posttrans scriptlets

["Richard Purdie" <richard.purdie@linuxfoundation.org>]

On Mon, 2021-08-23 at 15:14 +0200, Zoltan Boszormenyi via lists.openembedded.org
wrote:
> It's documented at www.rpm.org, Red Hat and SuSE.
> 
> https://docs.fedoraproject.org/en-US/packaging-guidelines/Scriptlets/
> https://en.opensuse.org/openSUSE:Packaging_scriptlet_snippets
> https://rpm-packaging-guide.github.io/
> 
> It's available since RPM 4.4. RPM 4.16.1.3 is in Hardknott.
> 
> A short Google search showed that *maybe* dpkg also supports it,
> or at least there are(were?) plans to implement it:
> https://wiki.debian.org/i18n/TranslationDebsProposals
> 
> With postinst, the time when the scriptlet runs may depend
> on the order of the packages in the upgrade transaction and
> may run with the older binaries in place.
> 
> The posttrans scriptlet runs at the very end of the
> install/upgrade transaction, just like the OPKG "intercept"
> scripts.
> 
> I successfully tested it long ago with a kernel upgrade
> that was running dracut on the new version, plus a dracut
> upgrade in the same session. Dissecting the initramfs proved
> that with postinst, the old dracut version generated it,
> while with posttrans, the new one did.
> 
> Who would want to use it? Anyone who needs some finalizer
> script to be run at the end of an RPM/DNF transaction.
> 
> I started on Angstrom and I know opkg doesn't implement this
> and Yocto inherited it. This patch is basically a heads up
> that there's another way to do the same thing. Maybe opkg
> implements this some day, maybe not. Anyway, Yocto in general
> will need more changes before the main package.bbclass can
> start to use it for all package formats.
> 

There is a bigger issue here which is how can anyone use this without making
their recipe rpm specific? I'm very reluctant to add functionality like this
unless we have a way to emulate it or fall back on the other package formats.

The intercept pieces are used by both rpm and opkg and actually do something
slightly different, they stack common calls until the end of the transaction.
This means for example we can call "ldconfig" once rather than in every
postinst.

Cheers,

Richard