[Yocto-builds] buildbot failure in Yocto on crownbay

[Tom Zanussi]

On Tue, 2012-06-05 at 08:11 -0700, Scott Garman wrote:
> On 06/04/2012 10:29 PM, Tom Zanussi wrote:
> > On Mon, 2012-06-04 at 22:21 -0700, Scott Garman wrote:
> >> On 06/04/2012 10:16 PM, Tom Zanussi wrote:
> >>> On Mon, 2012-06-04 at 21:10 -0700, Scott Garman wrote:
> >>>> On 06/04/2012 08:58 PM, Tom Zanussi wrote:
> >>>>> On Mon, 2012-06-04 at 20:27 -0700, Scott Garman wrote:
> >>>>>> On 06/04/2012 01:28 PM, Tom Zanussi wrote:
> >>>>>>> On Fri, 2012-06-01 at 09:17 -0700, Scott Garman wrote:
> >>>>>>>> On 06/01/2012 09:12 AM, Paul Eggleton wrote:
> >>>>>>>>> On Friday 01 June 2012 00:09:13 elizabeth.flanagan at intel.com wrote:
> >>>>>>>>>> Build status: FAILURE
> >>>>>>>>>>
> >>>>>>>>>> Buildslave for this Build: ab09
> >>>>>>>>>> Complete logs for all build steps:
> >>>>>>>>>> http://autobuilder.yoctoproject.org:8010/builders/crownbay/builds/134 Build
> >>>>>>>>>> Reason: Triggerable(crownbay)
> >>>>>>>>>> Build Source Stamp: [branch denzil] 7552ccd06c6327b5f36ee0cceb432946f8e98d3f
> >>>>>>>>>
> >>>>>>>>> | error: Failed dependencies:
> >>>>>>>>> | 	libusb-0.1-4>= 0.1.3 is needed by gnupg-2.0.18-r1.core2
> >>>>>>>>> NOTE: package core-image-sato-sdk-1.0-r0: task do_rootfs: Failed
> >>>>>>>>>
> >>>>>>>>> Not sure what is going on here but we seem to be getting this a lot with the
> >>>>>>>>> denzil builds. Scott, any ideas?
> >>>>>>>>
> >>>>>>>> I'm planning to run some builds over the weekend, and I'll make sure
> >>>>>>>> core-image-sato-sdk is one of them.
> >>>>>>>>
> >>>>>>>
> >>>>>>> Were you able to see anything?
> >>>>>>>
> >>>>>>> I was unable to reproduce this here, but it seems to be blocking all the
> >>>>>>> meta-intel BSPs on autobuilder...
> >>>>>>
> >>>>>> I haven't been able to reproduce it building core-image-sato for crownbay.
> >>>>>>
> >>>>>> However, I have the feeling this is the rpm ordering bug that bit us for
> >>>>>> the useradd packages that rely on shadow utils and friends being installed.
> >>>>>>
> >>>>>
> >>>>> Is there a workaround for that?
> >>>>>
> >>>>> It seems only the autobuilder is seeing this problem, and everyone else
> >>>>> doing their own (from-scratch?) builds is able to get a good build...
> >>>>>
> >>>>> Considering that all we need at this point are images for QA to test, is
> >>>>> there something that can be done on autobuilder to get past this and get
> >>>>> some images built?
> >>>>>
> >>>>> One thing I noticed is that gnupg depends on libusb, but gnupg doesn't
> >>>>> have libusb in its DEPENDS.  I don't know if thats the source of the
> >>>>> problem, but shouldn't it regardless?
> >>>>>
> >>>>> Beth, can you try the following patch on one of the failing builds and
> >>>>> see if it helps?
> >>>>
> >>>> I'm not at all clear why gnupng would depend on libusb? Do you know?
> >>>>
> >>>> That said, it's definitely not in the dependency list, so if it's
> >>>> needed, we do need to add it.
> >>>>
> >>>
> >>> I'm not at all familiar with this recipe, but it's gnupg (GNU Privacy
> >>> Guard), not gnupng, and this is from its configure:
> >>>
> >>> # libusb allows us to use the integrated CCID smartcard reader driver.
> >>>
> >>> CCID can be turned off with --disable-ccid-driver, but I have no idea if
> >>> that's what we want gnupg to do.
> >>
> >> I can say with some confidence that disabling this option and not having
> >> to depend on libusb is the way to go.
> >>
> >
> > OK, great, if you can tell me why (something I can put in a commit
> > message), I'll submit the patch if Beth can give it a test on one (or
> > all) of the the failed builds.
> >
> > Beth, can you try the patch below on one (or all) of the failed builds
> > and let us know if it fixes things?
> 
> I'd just note that the CCID smartcard reader is a specific piece of 
> hardware that is unlikely to be used in a majority of our use cases. 
> Users who need this feature can add it back as needed, and will also 
> need to add libusb to DEPENDS.
> 

OK, I've pushed both fixes to the following branch.

http://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/log/?h=tzanussi/poky-denzil-gnupg-dependency-fix.v1

The first patch adds the libusb DEPENDS, while the second cancels that
out by removing the feature that requires it as per the explanation
above.

This is against poky/denzil; for the 1.2 meta-intel release, we'll need
to add these changes as bbappends to meta-intel, but this is probably
easier for Beth to apply to the build at this point.

Tom

> Scott
>