On Fri, 2022-06-10 at 11:43 +0100, Richard Purdie via
lists.openembedded.org wrote:
> On Thu, 2022-06-09 at 17:39 +0100, Jose Quaresma wrote:
> > The archiver doesn't support using multiconfig sharing the same TMPDIR.
> > When in the multiconfig and using the same TMPDIR: the two machines can make
> > the same copy at the same time to the same destination, which will not work.
> >
> > This issiue can be replicated with meta-ti layer:
> > > # use a multiconfig machine
> > > MACHINE = "am64xx-evm"
> > > # archive sources for target recipes in conf/local.conf
> > > INHERIT += "archiver"
> > > COPYLEFT_RECIPE_TYPES = "target"
> > > ARCHIVER_MODE[src] = "original"
> > > ARCHIVER_MODE[diff] = "1"
> > > # and run bitbake with:
> > > bitbake mc:k3r5:gcc-source-11.2.0 gcc-source-11.2.0 -c
> > > do_unpack_and_patch -f
> >
>
> I went digging. If I enable this as above (using master) and run
>
> $ bitbake mc:k3r5:gcc-source-12.1.0 gcc-source-12.1.0 -c do_unpack_and_patch
> -S none
>
> then:
>
> $ bitbake-diffsigs
> tmp/stamps/work-shared/gcc-12.1.0-r0.do_unpack_and_patch.sigdata.*
> NOTE: Starting bitbake server...
> Task dependencies changed from:
> ['ABIEXTENSION', 'ARCHIVER_MODE[diff-exclude]', 'ARCHIVER_MODE[diff]',
> 'ARCHIVER_MODE[src]', 'ARCHIVER_OUTDIR', 'ARCHIVER_TOPDIR',
> 'ARCHIVER_WORKDIR', 'B', 'BASEURI', 'EXTENDPE', 'GNU_MIRROR', 'HOST_ARCH',
> 'HOST_OS', 'HOST_SYS', 'HOST_VENDOR', 'INSANE_SKIP', 'LIBCEXTENSION',
> 'PATCHTOOL', 'PE', 'PF', 'PN', 'PR', 'PV', 'RECIPE_SYSROOT_NATIVE', 'S',
> 'SDE_DIR', 'SDE_FILE', 'SOURCEDIR', 'SRC_URI', 'SRC_URI[sha256sum]',
> 'STAGING_DIR_NATIVE', 'T', 'TARGET_ARCH', 'TARGET_OS', 'TARGET_SYS',
> 'TARGET_VENDOR', 'TUNE_ARCH', 'UNPACK_DIR', 'base_do_unpack',
> 'base_do_unpack[cleandirs]', 'create_diff_gz',
> 'create_source_date_epoch_stamp', 'do_patch', 'do_qa_patch', 'do_qa_unpack',
> 'do_unpack', 'do_unpack[cleandirs]', 'is_work_shared', 'patch_do_patch',
> 'src_patches', 'unpack_check_src_uri']
> to:
> ['ABIEXTENSION', 'ABIEXTENSION_32', 'ABIEXTENSION_64',
> 'ARCHIVER_MODE[diff-exclude]', 'ARCHIVER_MODE[diff]', 'ARCHIVER_MODE[src]',
> 'ARCHIVER_OUTDIR', 'ARCHIVER_TOPDIR', 'ARCHIVER_WORKDIR',
> 'ARMPKGSFX_ENDIAN_64', 'B', 'BASEURI', 'EXTENDPE', 'GNU_MIRROR', 'HOST_ARCH',
> 'HOST_OS', 'HOST_SYS', 'HOST_VENDOR', 'INSANE_SKIP', 'LIBCEXTENSION',
> 'PATCHTOOL', 'PE', 'PF', 'PN', 'PR', 'PV', 'RECIPE_SYSROOT_NATIVE', 'S',
> 'SDE_DIR', 'SDE_FILE', 'SOURCEDIR', 'SRC_URI', 'SRC_URI[sha256sum]',
> 'STAGING_DIR_NATIVE', 'T', 'TARGET_ARCH', 'TARGET_OS', 'TARGET_SYS',
> 'TARGET_VENDOR', 'TUNE_ARCH', 'TUNE_ARCH_32', 'TUNE_ARCH_64', 'UNPACK_DIR',
> 'base_do_unpack', 'base_do_unpack[cleandirs]', 'create_diff_gz',
> 'create_source_date_epoch_stamp', 'do_patch', 'do_qa_patch', 'do_qa_unpack',
> 'do_unpack', 'do_unpack[cleandirs]', 'is_work_shared', 'patch_do_patch',
> 'src_patches', 'unpack_check_src_uri']
> basehash changed from
> 005580a7a5a1ab1ff554d0a9e47c2ad4ba932f3fe185d8d349b06c83e5e58b6c to
> d7b2e696005a578d5958ef4e73f1fc83e22612a3ae4294fb0540b0fadcb6c50f
> List of dependencies for variable ABIEXTENSION changed from 'set()' to
> '{'ABIEXTENSION_64', 'ABIEXTENSION_32'}'
> List of dependencies for variable TUNE_ARCH changed from 'set()' to
> '{'TUNE_ARCH_32', 'TUNE_ARCH_64'}'
> Dependency on variable ABIEXTENSION_32 was added
> Dependency on variable ABIEXTENSION_64 was added
> Dependency on variable ARMPKGSFX_ENDIAN_64 was added
> Dependency on variable TUNE_ARCH_32 was added
> Dependency on variable TUNE_ARCH_64 was added
> Variable ABIEXTENSION value changed:
> @@ -1 +1,2 @@
> -eabi
> +${@bb.utils.contains('TUNE_FEATURES', 'aarch64', '${ABIEXTENSION_64}',
> '${ABIEXTENSION_32}', d)}
> +TUNE_FEATURES{aarch64} = Set
> Variable TUNE_ARCH value changed:
> @@ -1,2 +1,2 @@
> -${@bb.utils.contains('TUNE_FEATURES', 'bigendian', 'armeb', 'arm', d)}
> -TUNE_FEATURES{bigendian} = Unset
> +${@bb.utils.contains('TUNE_FEATURES', 'aarch64', '${TUNE_ARCH_64}',
> '${TUNE_ARCH_32}', d)}
> +TUNE_FEATURES{aarch64} = Set
>
> this tells us the real issue. The task is changing due to toolchain
> issues. These stamps should be the same and aren't and this is the real
> issue that needs fixing.
To further follow up, there are two issues:
a) the task uses ${B} which for gcc includes HOST_SYS and TARGET_SYS.
We can just set that in the gcc-source recipe to B =
"${WORKDIR}/build".
b) the archiver uses:
ARCHIVER_OUTDIR = ${ARCHIVER_TOPDIR}/${TARGET_SYS}/${PF}/
which includes TARGET_SYS and between the two different MACHINE values,
this changes from 'arm-poky-linux-gnueabi' to 'aarch64-poky-linux'.
This leads to the task running twice, once for each multiconfig.
The first is easily solved as I mentioned. The question in the second
case is whether the task should run twice and where the output should
go since it is common to both builds. I do wonder if the archiver
shouldn't set:
ARCHIVER_OUTPUT:pn-gcc-source = "${ARCHIVER_TOPDIR}/allarch/${PF}"
Cheers,
Richard
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#166797):
https://lists.openembedded.org/g/openembedded-core/message/166797
Mute This Topic: https://lists.openembedded.org/mt/91650171/21656
Group Owner: openembedded-core+ow...@lists.openembedded.org
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub
[arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
