[OE-core] [PATCH 00/38] Morty Next pull request

[Martin Jansa]

I think I've found the reason why it was failing for me.

glibc is now the only component in morty release which is using
TMPDIR/sysroots-components (normally used only after RSS):
oe-core$ git grep "\${STAGING_DIR}-components"
meta/recipes-core/glibc/glibc-locale.inc:LOCALETREESRC =
"${STAGING_DIR}-components/${PACKAGE_ARCH}/glibc-stash-locale"
meta/recipes-core/glibc/glibc-mtrace.inc:SRC =
"${STAGING_DIR}-components/${PACKAGE_ARCH}/glibc-stash-locale/scripts"
meta/recipes-core/glibc/glibc-package.inc:do_stash_locale[sstate-outputdirs]
= "${STAGING_DIR}-components/${PACKAGE_ARCH}/glibc-stash-locale"
meta/recipes-core/glibc/glibc-package.inc:do_stash_locale[sstate-fixmedir]
= "${STAGING_DIR}-components/${PACKAGE_ARCH}/glibc-stash-locale"
meta/recipes-core/glibc/glibc-scripts.inc:SRC =
"${STAGING_DIR}-components/${PACKAGE_ARCH}/glibc-stash-locale/scripts"

and the workspace of my jenkins jobs is still running test-dependencies
jobs for morty, which does only selective pruning of TMPDIR:
http://git.openembedded.org/openembedded-core/tree/scripts/test-dependencies.sh#n165

which left TMPDIR/sysroots-components files from qlibc.do_stash_locale
behind and when next build started in the same workspace it failed, because
it was trying to overwrite files already there.

Every time I wiped whole TMPDIR and restarted the jobs it worked for a
while and then it got broken again during first test-dependencies job in
each workspace.

I can obviously change test-dependencies.sh to
delete TMPDIR/sysroots-components as well and I'll send update for it after
a bit more testing.

Ross's oe-core/scripts/wipe-sysroot and maybe some other scripts/tools need
similar fix as well.

Regards,


On Fri, Apr 6, 2018 at 4:15 PM, Martin Jansa <martin.jansa at gmail.com> wrote:

> In this build I was starting with empty TMPDIR, there shouldn't be any old
> files, but will retry manually in different builds.
>
> On Fri, Apr 6, 2018 at 4:11 PM, Richard Purdie <richard.purdie@
> linuxfoundation.org> wrote:
>
>> On Fri, 2018-04-06 at 16:07 +0200, Martin Jansa wrote:
>> > > Is it safe to backport to morty, which doesn't have RSS?
>> >
>> > It doesn't work well for me, all the builds are now failing with:
>> >
>> > ERROR: glibc-2.24-r0 do_stash_locale_setscene: The recipe glibc is
>> > trying to install files into a shared area when those files already
>> > exist. Those files and their manifest location are:
>> >    TMPDIR/sysroots-components/aarch64/glibc-stash-locale/SUPPORTED
>> >  Matched in b''
>> >    TMPDIR/sysroots-components/aarch64/glibc-stash-
>> > locale/lib/libnss_nisplus.so.2
>> >  Matched in b''
>> >    TMPDIR/sysroots-components/aarch64/glibc-stash-locale/lib/ld-
>> > 2.24.so
>> >  Matched in b''
>> > ...
>>
>> I suspect using its own sstate task is ok in morty even if it didn't
>> have RSS. What may have broken is the existing TMPDIR upgrade path
>> though :(
>>
>> If the files are removed, I suspect things will then work out ok. There
>> were manifest handling changes in later versions which dealt with this
>> kind of issue better (removals before installations).
>>
>> Cheers,
>>
>> Richard
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20180411/6eddcc83/attachment-0002.html>