The snapshot generation logic was moved earlier in the build (before
the cleaning logic) which caused the preclean to nuke the freshly
generated cache. Move all the preclean logic back up before that.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Due to https://bugs.gentoo.org/722108 , selinux stable builds cannot
possibly succeed right now. We are wasting lots of hours in update_seed
since that part cannot be cached. We can re-enable them when
https://bugs.gentoo.org/714376 is fixed (stabilization).
Signed-off-by: Ben Kohler <bkohler@gentoo.org>
Added a small debug echo so that post_build.log shows some information
about what's being uploaded. Might add destination info later.
Signed-off-by: Ben Kohler <bkohler@gentoo.org>
Use an easier-to-read "upload" function in place of "${cmd[@]}" array
syntax. Added a helper var UPLOAD_DEST to be set in advance so that
upload function takes only a list of filenames to be uploaded.
Signed-off-by: Ben Kohler <bkohler@gentoo.org>
Previously when we were copying locally there was no (major) problem
with wildcarding a long list of files to be transferred after every
single spec finishes. Now that we are transferring to a remote host,
this results in a lot of wasted transfer time & bandwidth.
I have adjusted the post_build function to only transfer the new files
known to be produced by each spec (which has files meant to be
published).
Signed-off-by: Ben Kohler <bkohler@gentoo.org>
Previously when we were copying locally there was no (major) problem
with wildcarding a long list of files to be transferred after every
single spec finishes. Now that we are transferring to a remote host,
this results in a lot of wasted transfer time & bandwidth.
I have adjusted the post_build function to only transfer the new files
known to be produced by each spec (which has files meant to be
published).
Signed-off-by: Ben Kohler <bkohler@gentoo.org>
Most architecture's build systems are hosted at OSUOSL. The place the
stages need to end up is ultimately masterreleases.gentoo.org, also at
OSUOSL. For some unknown reason, instead of rsync'ing from one system to
another in the same building, we were sending the stages first (slowly,
I might add) to nightheron in France, before sending them back to
OSUOSL.
robbat2 has added a releng-incoming.gentoo.org DNS record that currently
points to nightheron. This will allow us to switch the record in the
future and save the stages from needing to circumnavigate the globe
before reaching the mirrors.
Signed-off-by: Matt Turner <mattst88@gentoo.org>
We're using UserKnownHostsFile=/dev/null to force ssh to use the
fingerprint from DNSSEC, so StrictHostKeyChecking=no just prevents ssh
from asking us to confirm the fingerprint.
Signed-off-by: Matt Turner <mattst88@gentoo.org>
This causes an interactive prompt:
Are you sure you want to continue connecting (yes/no/[fingerprint])?
The plan is to get the SSH fingerprint via DNSSEC, so I'll revert this
change until that is in place and wired up.
This reverts commit 63174346f2.
Signed-off-by: Matt Turner <mattst88@gentoo.org>
Most architecture's build systems are hosted at OSUOSL. The place the
stages need to end up is ultimately masterreleases.gentoo.org, also at
OSUOSL. For some unknown reason, instead of rsync'ing from one system to
another in the same building, we were sending the stages first (slowly,
I might add) to nightheron in France, before sending them back to
OSUOSL.
robbat2 has added a releng-incoming.gentoo.org DNS record that currently
points to nightheron. This will allow us to switch the record in the
future and save the stages from needing to circumnavigate the globe
before reaching the mirrors.
Signed-off-by: Matt Turner <mattst88@gentoo.org>
catalyst-auto-ppc64.conf is no longer used since ppc and ppc64 builds
are done from catalyst-auto-ppc.conf. Repurpose (and rename) this file
to be used by ppc64le.
Signed-off-by: Matt Turner <mattst88@gentoo.org>
- Update kconfig for alpha, now including virtio for block/scsi and
networking, and make all qla drivers modules.
- Add linux-firmware to installer CD.
- Rearrange some variables in tools/catalyst-auto-alpha.conf
- Remove cifs-utils (huge Samba 4 dep)
- Add btrfs-progs
- Remove "static" USE flag, since it causes a USE dep conflict in pam.
Signed-off-by: Tobias Klausmann <klausman@gentoo.org>
Signed-off-by: Matt Turner <mattst88@gentoo.org>
As has been done in other arches' specs, allow a list of different
stage3 tarball extensions. Also disable building of multilib stages.
Signed-off-by: Ben Kohler <bkohler@gentoo.org>
Otherwise we write the last_success_file when the first build succeeds
and prevent the second from executing. Both of these sets of builds
should probably be executed from the same catalyst-auto file...
Commit ed258c751b (change
datestamp->timestamp) broke a lot of builders. The symlinks
and upload logic were all based on the datestamp. Update it
to use the new timestamp format.
Commit b7e493b4e4 (run trigger_post_build
after every successful spec) broke a lot of builders. The configs were
written to be called only after everything was done, but by calling it
after the first build (stage1 only), all the other artifacts we expected
did not yet exist.
Change post_build to pass the current set & spec name down so the config
can figure out what is available at that point in time.
All configs defining update_symlinks run it in pre_build, and all the
other configs will want to do the same. So make it the default in the
pre_build phase.
Mixing inline funcs and executable code makes it hard to follow and
shuffle ordering of operations. Put everything other than variables
into functions.
The delta here is large, but it's almost entirely indentation changes.
This also makes updating the script (like `git pull`) more resilient.