Re: [OE-core] State of bitbake world, wrong PACKAGE_ARCHs 2015-09-08

[Martin Jansa <martin.jansa@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 4400 bytes --]

On Wed, Sep 09, 2015 at 05:14:59PM +0200, Andreas Müller wrote:
> On Wed, Sep 9, 2015 at 4:55 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
> > On Wed, Sep 09, 2015 at 09:14:20AM +0200, Andreas Müller wrote:
> >> On Wed, Sep 9, 2015 at 1:10 AM, Martin Jansa <martin.jansa@gmail.com> wrote:
> >> > On Tue, Sep 08, 2015 at 11:39:20PM +0200, Andreas Müller wrote:
> >> >> On Tue, Sep 8, 2015 at 7:09 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
> >> >> > ERROR: polkit-group-rule-datetime different signature for task do_configure.sigdata between qemux86copy and qemuarm
> >> >> > runtaskdeps changed from ['polkitpolkit-group-rule-datetime.bb.do_patch', 'polkitpolkit_0.113.bb.do_populate_sysroot', 'shadowshadow_4.2.1.bb.do_populate_sysroot:virtual:native'] to ['polkitpolkit-group-rule-datetime.bb.do_patch', 'polkitpolkit_0.113.bb.do_populate_sysroot', 'shadowshadow_4.2.1.bb.do_populate_sysroot:virtual:native']
> >> >> > /home/jenkins/oe/world/shr-core/meta-openembedded/meta-oe/recipes-extended/polkit/polkit_0.113.bb.do_populate_sysroot with hash 284d6bbd28ea723c466e0c5c88a09a50
> >> >> I missed many of these - because I am not up to date on oe-core (did
> >> >> not want to loose midori for my images) - will change that soon. For
> >> >> this type of error: is it correct that it is caused by allarch recipes
> >> >> depending on machine/architecture specific recipes?
> >> >
> >> > close, the important part is few lines bellow:
> >> >
> >> > /home/jenkins/oe/world/shr-core/meta-openembedded/meta-oe/recipes-extended/polkit/polkit_0.113.bb.do_populate_sysroot
> >> > with hash
> >> > 284d6bbd28ea723c466e0c5c88a09a50
> >> >  changed to
> >> > /home/jenkins/oe/world/shr-core/meta-openembedded/meta-oe/recipes-extended/polkit/polkit_0.113.bb.do_populate_sysroot
> >> > with hash
> >> > 06cf7139c6588d047fa11444d88175ab
> >> > Hash for dependent task polkitpolkit_0.113.bb.do_populate_sysroot
> >> > changed from 284d6bbd28ea723c466e0c5c88a09a50 to
> >> > 06cf7139c6588d047fa11444d88175ab
> >> >
> >> > so allarch recipe polkitpolkit-group-rule-datetime is depending on
> >> > TUNE_PKGARCH-specific recipe polkit (on it's do_populate_sysroot task
> >> > which is expected to be different for diferent TUNE_PKGARCHs)
> >> >
> >> So to fix we could
> >>
> >> 1. either remove dependencies or
> >> 2. remove allarch or
> >> 3. ??
> >
> > 3a) add such dependency to SIGGEN_EXCLUDERECIPES_ABISAFE if it's
> > completely ABI safe (recipes depending on it shouldn't get rebuild after
> > any changes to it)
> > 3b) add such dependency to SIGGEN_EXCLUDE_SAFE_RECIPE_DEPS to exclude it
> > only for given recipe (in this case by saying that polkit-groups doesn't
> > need rebuild when polkit recipe is changed in any way)
> >
> >> For e.g the polkit-groups 1 seems not possible what about package
> >> feeds when removing allarch?
> >
> > I think in this case 3a) and 3b) don't apply as well, so easiest and
> > safest option is 2), package feeds should survive that, because new
> > package will have higher version (when PRserv is used) and also higher
> > package arch priority (all is lowest). Even the sysroot should now
> > survive ok, with new functionality which "uninstalls" staged file for
> > unreachable stamps.
> >
> > Also polkit-groups build doesn't take very long, so we don't mind to
> > build it once per TUNE_PKGARCH instead of building it as allarch only
> > once but with unsafe deps.
> >
> Thanks a lot and I agree with removing allarch for this case. Hope I
> have the courage to update oe-core this weekend :)
> 
> Ah - one further question for tests: The allarch signature change is
> detected only when building multiple machines - correct?

Yes, to detect the most obvious issues I'm using qemux86 and qemux86copy
MACHINEs (qemux86copy is added in one of oe-core/jansa/master patches).
This is good for detecting TUNE_PKGARCH depending on MACHINE_ARCH.

But to detect allarch depending on TUNE_PKGARCH recipe you need to
compare it between 2 MACHINEs with different TUNE_PKGARCH (I'm using
qemux86 and qemuarm).

Most of this testing logic is defined in:
openembedded-core/scripts/sstate-diff-machines.sh
if you pass --analyze parameter, that's what jenkins jobs are using to
generate this report.

Regards,

> 
> Andreas

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 188 bytes --]