Changelog
autopkgtest (3.18.1~ubuntu14.04.1) trusty-backports; urgency=medium
* No-change backport to trusty
autopkgtest (3.18.1) unstable; urgency=medium
* With --apt-pocket=POCKET=pkgname,... also include <release>-updates.
* cloud-vm-setup, adt-setup-vm: Disable /etc/cron.daily/apt.
* Detect if apt fails to download packages (as opposed to installing them),
retry up to three times, and eventually fail with exit code 16. This
avoids failing tests due to DNS/infrastructure problems and handling them
as tmpfails instead.
* Add test case for multiple Tests: with --apt-pocket=proposed=pkg
selection and nonexisting dependency.
* Properly keep track of apt pinnings across testbed resets.
* cloud-vm-setup, adt-setup-vm: Purge open-iscsi.
* Treat apt errors (exit code 100) from source download/extraction as
tmpfail.
* ssh-setup/nova: Stop adding apt sources for restricted/multiverse, as
current images now have them by default (see LP #1177432). This avoids
duplicate apt sources which cause stderr failure with some tests.
* Recognize negative architecture restrictions in test Depends:.
(LP: #1516959)
* tools/adt-buildvm-ubuntu-cloud: Counter-hack LP #1510345 harder by
checking /proc/cmdline instead of /etc/default/grub.
autopkgtest (3.18) unstable; urgency=medium
Behaviour changes/improvements:
* adt-virt-lxc: Drop --eatmydata option. Conceptually this isn't specific to
the LXC runner, it should always be used for package installation (when
available) with all runners, but never for tests.
* setup-commands/cloud-vm-setup, tools/adt-setup-vm: Install and use
eatmydata.
* Use eatmydata for --apt-upgrade and installing test dependencies, if
available.
* ssh-setup/nova: Provide custom implementation of "wait-reboot" using "nova
reboot --poll". This works more reliably on some clouds.
* ssh-setup/nova: Call "nova boot" with --debug to better be able to
diagnose failures.
* Extend --apt-pocket=POCKET option to accept an additional
"=pkgname1,src:srcname1,..." package list. An entry "src:srcname" expands
to all binary packages built by that source. If given, set up apt pinning
to use only those packages from POCKET, so that package updates in that
pocket can be tested independently from each other for better isolation.
(LP: #1503150)
* Add ssh-setup/maas: Acquire and Deploy a machine via MAAS to use as an
autopkgtest testbed. It assumes that MaaS is already set up, machines are
commissioned to it, and you added your ssh key to it.
Bug fixes:
* ssh-setup/nova: Drop "Warning: PoC", we've been using this in production
for a long time.
* tools/adt-setup-vm: Check if we already have a configuration for the
specific network interface that we are configuring, not just any
interface.
* tools/adt-buildvm-ubuntu-cloud: Counter-hack the Ubuntu livecd-rootfs
workaround for LP #1510345 by creating a configuration for ens3 and
removing the one for eth0.
autopkgtest (3.17.4) unstable; urgency=medium
* Fix --apt-pocket for apt sources with [options].
* Add tests/testarchive.py module for creating dynamic archives of mock
debs, and add a full integration test for --apt-pocket.
* Fix version in adt-run output (@version@ moved to adt_testbed.py).
* adt-setup-vm, cloud-vm-setup: Purge lxc/lxd.
* tests/testdesc: Skip Debian.test_builddeps_profiles test if dpkg is too
old to support build profiles.
autopkgtest (3.17.3) unstable; urgency=medium
* Set DEB_BUILD_OPTIONS=parallel=<#cpu> for builds and tests (as they
sometimes build packages). (LP: #1399177)
* If copying up files/artifacts through shared dir fails because of
inaccessible files, fall back to piping them through the auxverb.
* adt-build-lxc: Install dbus; a lot of software indirectly expects it
(e. g. calling systemctl as user).
* tools/adt-setup-vm, setup-commands/cloud-vm-setup: Purge
libcpan-changes-perl, it breaks some Perl autopkgtests.
* tools/adt-setup-vm, setup-commands/cloud-vm-setup: Purge git, it breaks
Ruby's autopkgtests.
* NullRunner.test_tmp_install: Explicitly require Json 1.0 GI version, to
silence warning with current pygobject.
autopkgtest (3.17.2) unstable; urgency=medium
* ssh-setup/nova: Use instance UUIDs instead of names after initial "nova
boot", to avoid name collisions. (LP: #1495788)
* adt-setup-vm: Drop "pin eth0 name" hack, this was fixed properly in
systemd 226-2.
* adt-setup-vm: Generalize ethernet ifupdown setup for net.ifnames.
* adt_binaries.py: Fix resetting of testbed (regression in 3.17).
autopkgtest (3.17.1) unstable; urgency=medium
* Add new private python modules to Makefile, to actually ship them.
* tests/adt-run: Fix test regression when $ADT_TEST* are not set.
autopkgtest (3.17) unstable; urgency=medium
* setup-commands/cloud-vm-setup: Provide fallback for Ubuntu 12.04 which
does not yet have /etc/default/grub.cfg.d/.
* adt-virt-ssh: Add --timeout-ssh option. Thanks Leo Arias for the initial
patch. (LP: #1488358)
* adt-run: Add --env option to pass arbitrary environment variables to the
test.
* adt-virt-lxc: In the auxverb, clean up leaked background processes which
share the same stdout/stderr, to avoid eternal hangs. (LP: #1488359)
* Move Testbed class and related logic into separate module
lib/adt_testbed.py, and clean up some unnecessary code interdependencies
and error handling. This will make it easier to write alternative CLI
frontents. This also handles test timeouts in a better way: There now is a
proper "summary" report as "FAIL timed out", and subsequent tests will now
run. (Closes: #741322)
* Drop the --leave-lang option. Please use the explicit option --set-lang
instead, to avoid weakly defined behaviour.
* Move Binaries class into separate module lib/adt_binaries.py.
* Testbed.execute(): Explicitly decode output of programs as UTF-8, instead
of relying on the host locale's encoding.
* tools/adt-setup-vm: Only configure eth0 in ifupdown if it actually exists,
and then make sure it does not get renamed any more.
* Add @builddeps@ build profile parsing fallback for libdpkg-perl < 1.04;
just ignore all build profiles.
* ssh-setup/nova: Add workaroud for cloud-init bug writing a wrong
/etc/hosts. (LP #1494678)
* adt-run: Log running kernel after each boot. (part of LP #1491865)
* Track initial kernel version after test bed setup and when a test/reboot
changes the running kernel version. Log these, and create a new
"testinfo.json" in the output dir with this information. (LP: #1491865)
* If --env options are specified, add these as list of name=value strings as
"custom_environment" key to testinfo.json.
* Put vserver arguments into testinfo.json.
autopkgtest (3.16.3) unstable; urgency=medium
* adt-setup-vm: Use /etc/cloud.cfg.d/ to avoid conffile prompts on upgrades.
(LP: #1477626)
* Don't invoke QEMU with -localtime. QEMU defaults to and Linux prefers the
hw clock in UTC. This avoids large time jumps when doing NTP.
* adt-setup-vm: Start autopkgtest shell after all other SysV init
scripts/facilities, to avoid a too early start of tests while boot is
still in progress.
* doc/README.package-tests.rst: Use comma separator in examples.
* VirtSubproc.check_exec(): Actually error out on stderr as documented.
* adt-virt-ssh: Verify that sudo does not write errors, as that
interferes with auxverbs and generally is a sign of host mis-configuration
(typically wrong hostname).
* ssh-setup-nova: Set hostname to "adt". This avoids overly long host names
from defaulting to nova instance names (which can be very long).
* ssh-setup/nova: Fix parsing of IP address from "nova show" to avoid
tripping over instance names containing "network". (LP: #1481574)
* Fix error message and code for "invalid test depends" errors.
* tests/pep8: Ignore E402 ("module level import not at top of file"), as
that's impossible to satisfy; we have to set sys.path before.
* ssh-setup/nova: In cleanup(), wait until the instance gets deleted. This
avoids name collisions and instance quota overflow.
* adt-run: Always show summary at the end, which is particularly helpful for
multiple tests. (LP: #1484991)
* adt-run.1: Clarify that without any logging options adt-run only logs to
stderr. (LP: #1485661)
* adt-virt-lxc: If the executed command exits with 255, translate that to
253 in the auxverb wrapper, as 255 is the exit code for failures of the
auxverb itself. This avoids considering ordinary test failures as
temporary testbed failures.
* adt-run: Show host name, to more easily identify hosts with frequent
failures.
* adt-virt-lxc: Always call lxc-stop with --kill, as the containers are not
precious and normal shutdown might hang.
* adt-virt-lxc: In cleanup, stop LXC container before removing the shared
downtmp, to ensure that we always clean up the running container. Also
don't fail on errors. (LP: #1488879)
* ssh-setup/nova: Add missing apt sources for -updates/-security
restricted/multiverse pockets.
autopkgtest (3.16.2) unstable; urgency=medium
* Exit with code 4 ("test failure") if a test times out, instead of 16
("testbed failure").
* Exit with code 12 ("erroneous package") if test dependencies are
unsatisfiable and put badpkg/blame into summary, instead of exiting with
code 20 ("other unexpected failures including bad usage").
* Exit with code 12 ("erroneous package") and a proper error message if
debian/tests/<testname> does not exist, instead of code 20 and a confusing
message about chmod failing and *-stderr missing.
* Exit with code 12 ("erroneous package") instead of 20 if test dependency
package install fails.
* adt-virt-ssh: If the executed command exits with 255, translate that to
253 in the auxverb wrapper, as 255 is ssh's (and auxverb's) exit code if
something with the connection failed. This avoids considering ordinary
test failures as temporary testbed failures. (LP: #1475679)
autopkgtest (3.16.1) unstable; urgency=medium
* setup-commands/cloud-vm-setup: Fix installation of linux-generic on
Ubuntu.
* ssh-setup/nova: Clean up userdata temp file on failed nova boot.
* If the test fails with exit code 127 (usually "command not found"),
consider this a test failure (code 4), not a testbed failure (code 16).
* doc/README.package-tests.rst: Document that "Tests:" and similar fields
can be comma separated too, like everywhere else in Debian's control
files.
* ssh-setup/nova: Show name of created nova instance, so that it's easier to
map instances to a test.
* run-from-checkout: Exec adt-run to avoid keeping the shell wrapper process
around.
* ssh-setup/snappy: Fix check whether --image and --channel are specified
together. (LP: #1474735)
* ssh-setup/nova: Allow selecting net IDs by name.
autopkgtest (3.16) unstable; urgency=medium
Behaviour changes and improvements:
* ssh-setup/nova: Re-try nova boot up to three times if it fails, and sleep
5 minutes in between. This occasionally fails due to exceeding quota or
temporary glitches.
* Drop gpg-signing local archive and mark it as [trusted=yes] instead. This
gets rid of the requirement to generate a gpg key on first run. Drop
corresponding --gnupg-home option. (LP: #1472691)
* Drop long-obsolete --tmp-dir backwards compat option.
* ssh-setup/snappy: Adjust ubuntu-device-flash arguments to new
release/channel structure from Ubuntu 15.10, and default to rolling/edge.
Add new --release option.
Bug fixes:
* ssh-setup/nova: Silence "invalid command wait-reboot" warning.
* setup-commands/cloud-vm-setup: Purge cloud-init too. It sometimes causes
longer hangs on boot and might interfere with tests in other ways, and we
only need it for first-time initialization anyway.
* adt-virt-qemu: adt-buildvm-ubuntu-cloud: Don't use -enable-kvm QEMU option
if /dev/kvm does not exist. (Closes: #790650)
* Quiesce warnings from tar. Avoids "time stamp is in the future" log spew.
(LP: #1468868)
* Move apt sources setup from setup-commands/cloud-vm-setup to
ssh-setup/nova. --apt-pocket does not work with the former as that runs
too late.
* Use Dpkg::Deps to evaluate/reduce build dependencies for expanding
"@builddeps@". This handles build profiles. (Closes: #787093)
* Resolve build profiles when installing build dependencies to build tested
package.
* Adjust NullRunner.test_tmp_install_perl test case to use
libtest-requires-perl instead of libtest-tester-perl; the latter got
removed from Debian.
autopkgtest (3.15.1) unstable; urgency=medium
* adt-buildvm-ubuntu-cloud: Fix lsb_release fallback if python3-distro-info
is not available.
* adt-buildvm-ubuntu-cloud: Don't move the uninitialized image to final
location if QEMU failed.
* adt-buildvm-ubuntu-cloud: Check availability of genisoimage and access of
/dev/kvm before downloading image. (LP: #1466486)
* Get along with dpkg-query not existing, such as in latest Ubuntu Snappy
images. Skip generation of testbed-packages artifact in that case.
(LP: #1469647)
* Adjust NullRunner.test_tmp_install test case to work with pygobject 3.16.
autopkgtest (3.15) unstable; urgency=medium
Behaviour changes and improvements:
* The specified --output-dir now must not exist or be empty. If it is
non-empty, adt-run now will refuse to use it instead of cleaning it. This
is safer, to avoid accidentally using the home directory or similarly
precious data. (LP: #1463561)
* Add --no-auto-control option to disable autodep8. (Closes: #788659)
* ssh-setup/nova: Add -e/--env option to append to /etc/environment. This is
mostly useful to configure $http_proxy and friends.
* ssh-setup/nova: Add --mirror option.
* Add setup-commands/cloud-vm-setup: --setup-commands script for cloud
instances to prepare them for running autopkgtests. This is similar to
adt-setup-vm, but won't configure serial console or networking.
Bug fixes:
* adt-setup-vm: Print a warning if calling update-grub fails, instead of
aborting. This fixes vmdebootstrap --grub, as that installs grub after
running the customization script. Adjust adt-virt-qemu.1 manpage to
recommend the --grub option of vmdebootstrap.
* testdesc.py, parse_rfc822(): Entirely ignore lines which only contain
whitespace after filtering out intra-line comments. They break
continuation line parsing in python-debian >= 0.1.23. (Closes: #784942)
* adt-setup-vm: Add inclusion of interfaces.d/ to /etc/network/interfaces.
This belongs into vmdebootstrap (#788792), but fix that locally for the
time being.
* adt-run.1: Document --testname.
autopkgtest (3.14.3) unstable; urgency=medium
* adt-virt-lxc: Don't try to set up a shared downtmp with unprivileged
(user) containers. This is full of permission errors due to the mapped
UIDs.
* tests/adt-run LxcRunner: Use $ADT_TEST_LXC as unprivileged user container
if it exists.
* adt-buildvm-ubuntu-cloud: Check at the beginning if QEMU command is
available and exit with proper error message if not. (LP: #1460066)
* If auxverb fails with exit code 126, 127, 254, or 255, consider this a
failure of auxverb/the testbed itself, not of the program run under it.
(As specified in doc/README.virtualisation-server.rst) (LP: #1462540)
autopkgtest (3.14.2) unstable; urgency=medium
* Add tests/testpkg-{simple,reboot,reboot-prepare}: Very simple stub
packages for manual testing.
* VirtSubproc, cmd_reboot(): Disconnect the backgrounded reboot command from
stdout/err, to avoid it getting killed by the ssh runner. (LP: #1454735)
autopkgtest (3.14.1) unstable; urgency=medium
* Go back to using the backgrounded reboot command as we had before 3.14.
Just calling "reboot" is prone to hangs and timeouts in case reboot is too
fast and shuts down our communication channel.
* Fix --override-control option in adt-run(1). (LP: #1453495)
* Fail with an error if --override-control file does not exist.
(LP: #1453503)
* ssh-setup/snappy: Add -b/--show-boot option to direct boot and console
messages to stdout. (LP: #1453154)
* adt-virt-qemu.1: Document --show-boot option.
* Add test case for using a @ command line include file for a virt runner.
* Strip leading and trailing space from argument lines in '@' included
command line argument files. (LP: #1453498)
autopkgtest (3.14) unstable; urgency=medium
Improvements:
* adt-virt-ssh: Add --reboot option to indicate that testbed can be
rebooted. This is useful when running without a setup script (which can
already declare capabilities).
* Simplify reboot implementation: Just call "reboot" as root in the testbed
to trigger rebooting, and replace the former hook_reboot() in adt-virt-*
with hook_wait_reboot() which merely waits for the testbed to go down and
come back up. This also replaces the "reboot" command for ssh setup
scripts to "wait-reboot" and in most cases it is not needed to implement
this at all (except for the "adb" script).
* Add /tmp/autopkgtest-reboot-prepare script for testbeds which support
rebooting. This saves the current state but leaves the actual reboot to
the test itself. This is useful if rebooting should happen with unusual
means (e. g. kexec) or as part of testing something else (like a system
upgrade). (LP: #1445819)
* Log complete command line invocation, for easier local reproduction,
comparison of log files, and clarifying which options were used.
(LP: #1449413)
Bug fixes:
* Update Vcs-Git: field for anonscm.debian.org.
* Drop obsolete XS-Testsuite: control field.
* adt-setup-vm: Fix script error if $ADT_SETUP_VM_POST_COMMAND is not set.
* README.click-tests.rst: Describe x-source manifest entry.
* Mark test dependencies as manually installed, so that they don't go away
if a test calls "apt-get autoremove".
* Set DEBIAN_FRONTEND=noninteractive for tests. This avoids stderr messages
about "unable to open stdin" and similar when tests call apt-get install.
autopkgtest (3.13) unstable; urgency=medium
* ssh-setup/adb: Add --keep-screen-active option to keep
"powerd-cli display" running after the tests. Use with caution!
(LP: #1441023)
* adt-buildvm-ubuntu-cloud: Add --post-command for running a shell command
after setup for additional customization.
* ssh-setup/adb: Quote tr arguments.
* click tests: Look into app/tests/autopilot/ for tests as well if no
explicit path is given. This is the default location of the Ubuntu cmake
templates.
autopkgtest (3.12.1) unstable; urgency=medium
[ Didier Roche ]
* setup-commands/ubuntu-touch-session: Use "upstart" instead of "init".
(LP: #1422681)
[ Martin Pitt ]
* adt-build-lxc: When removing the old root file system, remove btrfs
subvolumes in it.
* README.package-tests.rst: Document that '#' can be used for comments.
(Part of LP #1434543)
* Ignore debian/tests/control stanzas with unknown fields, as documented.
(LP: #1434543)
* adt-buildvm-ubuntu-cloud: Drop workaround for LP #1428495, it got fixed in
cloud-init a while ago.
* adt-buildvm-ubuntu-cloud: Fix cloud-init's prerm hacking to apply after
dist-upgrades.
autopkgtest (3.12) unstable; urgency=medium
* ssh-setup/adb: Re-enable ssh after reboot, in case an upgrade disables it.
* ssh-setup/adb: Try to create /userdata/.adb_onlock, to get adb after
rebooting without human interaction.
* Fix apt-get install --simulate version parsing with third-party sources.
(LP: #1430017)
* ssh-setup-nova: Add --security-groups option. (LP: #1429862)
* adt-buildvm-ubuntu-cloud: Avoid non-blocking reads from the socket in
verbose mode, this sometimes causes hangs. Go back to blocking reads, but
don't wait between them.
* Fix UnicodeDecodeError on .deb package name reading. (LP: #1430773)
* adt-virt-ssh: Export $SUDO_ASKPASS to tests if sudo is available.
(LP: #1431421)
autopkgtest (3.11.1) unstable; urgency=medium
* Fix autopkgtest-reboot to also work when being called through sudo, and
for forking test scripts.
* Avoid failure if /var/cache/apparmor/click-ap.rules does not exist any
more after a reboot, like on snappy.
* adt-buildvm-ubuntu-cloud: Avoid cloud-init's prerm stopping cloud-init's
services while we are still running them. This makes it possible to purge
cloud-init from user-data, and avoids killing apt/dpkg underneath us. This
needs a cleaner solution, but is a good enough workaround for now.
(LP: #1427999)
* adt-buildvm-ubuntu-cloud: Don't wait between serial console reads while
we have data. Provides faster/smoother output in --verbose mode.
* ssh-setup/nova: Fix error message on missing keypair, the command is
"keypair-add", not "keypair-create". Thanks Thomi Richards! (LP: #1428433)
* adt-buildvm-ubuntu-cloud: Add workaround for recent cloud-init regression
that disables ssh (LP #1428495)
* adt-run: Suggest common reason for unsatisfiable test dependencies.
Suggest using a current image, or run apt-get update/-U (for apt-get) or
--setup-commands ro-apt-update (for temp dir install mode). (LP: #1425682)
* Quiesce confusing "failed to create symbolic link
/sbin/autopkgtest-reboot" warning on read-only testbeds.
autopkgtest (3.11) unstable; urgency=medium
Improvements:
* Add action argument --testname to run only one individual test from the
next package.
Bug fixes:
* Restore click AppArmor profiles after a test, if we modified them before
the test.
* When running a deb test in a click+AppArmor environment, update AppArmor
rules for all clicks for introspectability and temporary test
dependencies. This takes about a minute before the test, and another one
after the test, but is necessary for use cases like ubuntu-sanity-tests.
(LP: #1424889)
autopkgtest (3.10) unstable; urgency=medium
Improvements:
* adt-virt-lxc: Add reboot support for non-ephemeral containers.
* tests/run-parallel: Don't run all test classes in parallel, it creates too
much load. Serialize test classes which use the same runner.
Bug fixes:
* VirtSubproc.py, cmd_reboot(): Fix debug message.
* tests/adt-run: Avoid re-downloading ubuntu-device-flash images by
symlinking ~/.cache/ubuntuimages from the temp $HOME into the real user's
$HOME.
* adt-virt-lxc: Fix check for eatmydata when using --eatmydata.
* adt-virt-qemu: Synchronize ttyS0 output before writing eofcat.
* adt-virt-qemu, adt-virt-lxc: Asynchronously call reboot, to avoid eternal
hangs when it reboots too fast for us to notice that it finished.
autopkgtest (3.9.7) unstable; urgency=medium
* Include "Package-Type: deb" marked packages for testing. Fixes regression
with testing paramiko.
* adt-virt-qemu: Robustify setupup of shared /autopkgtest directory.
* VirtSubProc.expect(): Read data without delay as long as we have some.
* adt-virt-qemu: Add --show-boot option to show boot messages from serial
console.
* adt-setup-vm: Configure "console=ttyS0" boot option in grub on non-i386
architectures too, to fix serial consoles with systemd images. Create
/etc/default/grub.d/ if necessary.
* adt-setup-vm: Only configure grub if update-grub is available.
(Closes: #778907)
* adt-virt-qemu: Increase reboot and dpkg-reconfigure timeouts; on loaded
machines these might take longer than 5s.
* adt-run: Unless --leave-lang is given, unset $LANGUAGE and all $LC_* for
tests, to avoid clobbering the testbed locale with the host's.
(LP: #1423359)
autopkgtest (3.9.6) unstable; urgency=medium
* Skip test_tmp_install_imagemagick test case if python3-wand is not
available.
* adt-virt-ssh: Ignore errors when cleaning up the work dir. This races with
sshd cleaning up its control socket, don't stumble over that.
(LP: #1420355)
* Implement --timeout-factor, which was forgotten with the rewrite in
version 3.2. (LP: #1420540)
* ssh-setup/adb: Fix waiting for the desktop to boot: Lower D-Bus call
timeout to 5 seconds, and fix retry loop to time out afer 5 minutes of
real time instead of a fixed number of retries (as the duration of the
D-Bus call is not very reliable).
* virt-subproc/adt-virt-ssh: Don't try to exit the ssh muxer if we don't
have a ssh command yet. This happens when the setup script fails to open
the testbed.
autopkgtest (3.9.5) unstable; urgency=medium
* adt-run install_tmp(): If we have Python and Perl modules, don't specify
the Python modules in $PERL5LIB.
* Recognize udev rules from packages in temp dir install mode. Copy them to
/run/udev/rules.d/ and trigger them. (LP: #1417966)
* setup-commands/ubuntu-touch-session: Don't fail on udevadm failure, which
cannot work in schroots.
* adt-virt-qemu: Ensure that python3 or python is available (which is
required for the auxverb helper) and exit with a meaningful error message
if it isn't. Adjust the manpage accordingly. Fix adt-setup-vm to install
python3-minimal if no Python is already installed. Thanks to Christian
Kastner for the initial patch! (Closes: #777478)
* Support temp dir install mode for python[3]-wand's really broken way of
finding the imagemagick shared libraries. (LP: #1419024)
autopkgtest (3.9.4) unstable; urgency=medium
* Recognize modern "Package-Type:" control field in addition to
XC-Package-Type.
* ssh-setup/adb: Guard adb wait-for-device with a 5 min timeout.
(LP: #1414939)
* README.running-tests.rst: Fix mis-spelled ro-apt setup command.
* Add setup-commands/ro-apt-update: Shortcut for temporarily mounting a
readonly root file system writable and run apt-get. This allows tests to
download/unpack test dependencies on older images whose apt indexes are
out of date. (LP: #1408350)
* Export temp dir install mode variables in the test run script instead of
merely running it under that environment, so that we trump non-extending
LD_LIBRARY_PATHs in /etc/environment. (LP: #1407949)
* setup-commands/ubuntu-touch-session: Ensure that /dev/uinput is always
accessible, as we don't necessarily have local logind sessions during
tests. (part of LP #1415514)
* setup-commands/ubuntu-touch-session: Unset UPSTART_JOB (which crept into
the environment from a recent upstart version) when calling initctl, to
fix querying the session D-BUS address. (LP: #1415514)
autopkgtest (3.9.3) unstable; urgency=medium
[ Christian Kastner ]
* Clean up /etc/dpkg/dpkg.cfg.d/autopkgtest on upgrades if its contents are
just "force-unsafe-io", indicating that this file was generated by the
above bug.
autopkgtest (3.9.2) unstable; urgency=medium
* adt-setup-vm: Fix all other places which operated on the host instead of
the guest file system when specifying a directory argument.
autopkgtest (3.9.1) unstable; urgency=medium
[ Martin Pitt ]
* Revert installation of ubuntu-device-flash in adt-run autopkgtest, it's
too unstable (timeouts and grub errors due to qemu in qemu).
[ Christian Kastner ]
* adt-run: Fix detection of dpkg-dev in the testbed. (Closes: #775073)
* adt-setup-vm: Set umask to 0022 so that ordinary users in the testbed can
read generated files. (Closes: #775075)
* adt-setup-vm: Modify guest's dpkg config for enabling force-unsafe-io, not
host's, when a root directory is specified as argument. (Closes: #775076)
* adt-setup-vm: Make sure $proxy variable exists before using it. Fixes
crash when apt is not configured to use a proxy. (Closes: #775092)
autopkgtest (3.9) unstable; urgency=medium
Improvements:
* Add "snappy" ssh setup script to run tests in an Ubuntu Snappy
(http://www.ubuntu.com/snappy) testbed in QEMU. This uses a given image
with -i/--image, or constructs a temporary one using ubuntu-device-flash
(which caches its downloads and is fast).
* adt-run autopkgtest: Install ubuntu-device-flash and qemu-kvm if
available, to run the snappy ssh script tests under autopkgtest in Ubuntu.
* ssh setup script "reboot" commands can now simply "exit 10" if they want
adt-run to just call "reboot" through ssh, instead of having a custom
implementation.
Bug fixes:
* tools/adt-setup-vm: Avoid debconf prompts during dist-upgrade.
* adt-virt-ssh: Wait for ssh to become available before checking for sudo.
Fixes root capability detection for setup scripts which don't already wait
for ssh themselves.
* VirtSubproc.cmd_reboot: Avoid multi-line shell command so that this also
gets along with the ssh runner.
* Fix FileNotFoundError if built tree does not have a debian/changelog file.
Simply skip detection of source package version in this case.
autopkgtest (3.8.1) unstable; urgency=medium
* If the testbed does not have root privileges (e. g. missing sudo
password), install click packages with pkcon and skip the AppArmor rule
adjustment. This makes non-root click tests at least work for
non-Autopilot cases. (LP: #1384417)
* Make --shell-fail also apply to test dependency installation failure.
* Run root tests through "su" as well (if possible), to ensure that these
also get a proper PAM/logind session with all runners. (LP: #1393474)
* Skip the test_tmp_install tests if apt-get download does not work. This
might happen in some build environments which completely disable
networking or don't have working apt-get download for some reason.
(Closes: #769687)
* Don't put the log FIFO into the output directory. This avoids keeping it
in --output-dir, which fails if the output dir happens to be in the tested
tree itself. (LP: #1393426)
* adt-setup-vm: Don't purge man-db. It uninstalls way too much on images
which aren't just minimal environments.
* adt-setup-vm: Drop bogus "X-Start-Before" line in generated init.d script.
(Closes: #770517)
* Respect DEB_BUILD_OPTIONS=nocheck (side issue in #769687)
autopkgtest (3.8) unstable; urgency=medium
Improvements:
* adt-build-lxc: Configure unsafe dpkg I/O and disable apt's download of
package translations, like adt-setup-vm.
* adt-build-lxc: Use eatmydata for lxc-create if available, and install
eatmydata into containers. For Ubuntu, stop installing the default
$packages_template (which is ssh,vim).
* adt-virt-lxc: Pass any additional options/arguments (after --) verbatim to
lxc-start{,-ephemeral}.
* adt-virt-lxc: Stop running ephemeral containers with --keep-data by
default, as this makes them unnecessarily slow due to running on disk.
Environments with little RAM are instead advised to explicitly supply this
or --storage-type as extra arguments.
Bug fixes:
* Create autopkgtest-reboot command in /tmp, as /sbin/ might be read-only
(and /run is usually "noexec", so we can't put it there either). Write it
with "noclobber" to guard against symlink attacks.
* During reboot, if /var/cache/ isn't writable (as on Ubuntu touch images,
argh), fall back to saving the current temp dir in /home.
* adt-virt-ssh: Recreate the sudo wrapper after reboot, as it is in /tmp.
* ssh-setup/adb: Setup adb port forwarding again after rebooting.
* Fix error message if /tmp/autopkgtest-reboot already exists.
* Fix extra space in apt-get option, which breaks apt-get's option parsing
in some circumstances.
* adt-build-lxc: Fix hardcoded "utopic" release name for cache cleanup on
btrfs, use the actual release name from the command line. (LP: #1392589)
autopkgtest (3.7.1) unstable; urgency=medium
* NullRunner.test_tmp_install_perl: Use a package from Ubuntu main, to fix
FTBFS in Ubuntu's buildds.
* Fix AppArmor rule adjustment for a local click package.
* When executing a command in the testbed times out, don't wait for it
after killing if the auxverb includes "sudo". We can't kill those
processes, so we'd wait forever; we can only hope that they eventually
terminate themselves after we stop the testbed. Fixes non-working timeouts
for LXC runner with the --sudo option.
* Show "set -x" trace of AppArmor reconfiguration in debug mode.
* Use "install" instead of "short" time out for downloading and unpacking
dependencies in "temp dir" mode.
* ssh-setup/adb: Retry connecting with ssh a few times, the first time
doesn't always succeed immediately after enabling ssh.
* tools/adt-setup-vm: Prevent subsequent cloud-init runs from modifying the
apt sources again.
autopkgtest (3.7) unstable; urgency=medium
Improvements/cleanup:
* When not logging to a file, colorize adt-run debug/info/warning/error
messages to better tell them apart from from apt and test output.
* adt-virt-schroot: Simplify determination of schroot session location.
* adt-virt-chroot: Drop support for dchroot arguments. dchroot has been
obsoleted by schroot for a long time now, and isn't being tested by the
developers. Allow relative chroot paths.
* adt-virt-ssh: Rewrite adt-ssh-wrapper to not call the readlink command on
every single /proc/*/fd/*. This is too slow on devices like Ubuntu Touch,
where it adds an extra ~ 5 seconds to every command.
* ssh-setup/adb: Drop --rw and --apt-update options. They stopped working
when adb switched to running as user, and can be done with
--setup-commands.
* Recognize Perl modules in temp dir install mode.
* Consider programs in the temp dir install tree for preparing the source
package. This makes it possible to run --source or --apt-source tests in
read-only environments without dpkg-dev installed. (LP: #1383878)
* With --apt-upgrade, retry apt-get update a third time after one minute.
With just a second try we still get occasional hash sum mismatches.
* Recognize Qt plugins in the temp dir install tree, so that autopilot tests
work on images which don't have autopilot-qt installed any more.
* Remove the "adt-satdep" package after test installation finished, to avoid
tests being confused about an unexpected package.
* ssh-setup/adb: Drop "autopilot --dbus-probe enable" call. It wastes some
1.5 minutes and is not needed any more, as we now reconfigure AppArmor for
the particular clicks that we want to test.
Bug fixes:
* adt-virt-ssh: Terminate the ssh connection muxer on cleanup. Fixes eternal
hangs on waiting for tee with the --output-dir option, as the muxer
inherits our stderr. (LP: #1384324)
* Fix crash when running from a git checkout, but git is not installed.
* ssh-setup/adb: Robustify powerd-cli invocation. It previously failed
randomly.
* ssh-setup/adb: Quiesce some noise from various phablet-config/gdbus/adb
commands.
* ssh-setup/adb: If the welcome wizard is already running, just kill that
instead of the entire session. In particular, this avoids killing adb on
both the host and the device.
autopkgtest (3.6) unstable; urgency=medium
[ Antonio Terceiro ]
* README.package-tests.rst: Make references to the control field consistent.
[ Martin Pitt ]
* When parsing .changes files, always register the debs before the dscs and
stop assuming Files: having this order already. (Closes: #764795)
* Drop the builtin synthesis of Perl/Ruby tests and use autodep8 instead.
Update the spec accordingly, and add a Recommends: autodep8.
(Closes: #763847)
* adt-build-lxc: Implement LXC cache cleanup for btrfs by removing the image
snapshots instead of rm -r'ing the cache dir, as the latter will fail.
* adt-setup-vm: Fix missing hashbang of generated /etc/init.d/autopkgtest.
(Regression from 3.5.5)
autopkgtest (3.5.5) unstable; urgency=medium
* adt-virt-ssh: Be more liberal when parsing the setup script's
capabilities, accept spaces and commas.
* adt-virt-ssh: Refactor code so that setup script's "revert" commands can
update/refresh the ssh config. Often revert translates to close and
re-open, which changes the IP. Update manpage and ssh-setup/SKELETON
accordingly.
* ssh-setup/nova: Make parsing commands work with mawk too, to avoid
depending on gawk.
* ssh-setup/nova: Implement a simple revert by a simple teardown/providing a
new instance. (This might become more efficient in the future with using
snapshots.)
* adt-buildvm-ubuntu-cloud: Put vmalloc kernel commandline config into
separate grub.d/ directory, to avoid subsequent cloud-init invocations
overwriting it.
* tools/adt-setup-vm: Add setup steps that adt-buildvm-ubuntu-cloud does, so
that vmdebootstrap generated VMs are functionally on par with the Ubuntu
cloud images: Set up serial console, dpkg unsafe-io, install some
additional packages, remove unnecessary packages, bump vmalloc on i386.
* adt-buildvm-ubuntu-cloud: Remove most of the setup steps from user-data,
and call adt-setup-vm instead. This keeps the knowledge of how to set up
VM testbeds in one single place.
autopkgtest (3.5.4) unstable; urgency=medium
adt-run fixes:
* Make apt-get source calls again show what's being downloaded, but not all
the other noise around it.
* Avoid copying down the tests tree into the testbed if it already exists
there. This is the common case for the first test in unbuilt trees, and
avoids one round of expensive large copying for the ssh and qemu runners.
* Fix relative --override-control paths: They should not be relative to the
source dir but relative to the current dir where adt-run is called.
ssh-setup/nova fixes:
* Add -N/--net-id argument for OpenStack deployments which require setting
an explicit --nic net-id.
* Stop using nova boot's --file option, it does not work any more (see
LP#1376176); use --user-data instead. Also set some go-faster apt options.
* Wait until ssh is available and cloud-init is done.
* Stop making --keyname mandatory. If not given, use the first one from
"nova keypair-list".
* Fix crash if ~/.ssh/known_hosts does not exist.
ssh-setup/adb fixes:
* Drop the sudo'ing for unlocking the screen. It stopped working and is not
necessary, now that we don't have images with adb running as root any
more.
* Fix unity8 restart after disabling first-time wizard to work on current
Ubuntu Touch images.
* Wait until com.canonical.UnityGreeter becomes available. This is a better
way to wait for the desktop to be started up, and also reliably unlocks
the greeter.
* During revert, don't fail if there are no configured networks.
* Add -b/--reboot option to reboot the device before running the test.
* Disable --reset option for now; it is broken as it does not restore the
password and developer mode. Point towards using --reboot, which should be
good enough for most cases in r/o mode.
adt-virt-ssh fixes:
* Implement reboot.
* Tests or builds sometimes leak background processes which might still be
connected to ssh's stdout/err; create and use a wrapper script that kills
these after the main program (build or test script) finishes, otherwise we
get eternal hangs as ssh waits until nothing is connected to its tty any
more.
Other fixes:
* adt-buildvm-ubuntu-cloud: Purge more unnecessary packages.
autopkgtest (3.5.3) unstable; urgency=medium
* ssh-setup/adb: Use the new "phablet-config network" to save/restore wifi
connections, which also works with non-root adb.
* Give adt-run's tee processes some time to catch up between printing
results and stderr, to avoid scrambled logs. (LP: #1373441)
* Fix NullRunner.test_timeout_no_output for older Python versions.
* tests/pyflakes: Skip test if pyflakes3 is not available (for backports).
* Bump Standards version to 3.9.6 (no changes necessary).
* NullRunner.test_timeout_no_output: Slightly relax time bound to make
failures less likely on loaded/slow machines.
* Install clicks with --allow-unauthenticated to support local developer
generated packages. (LP: #1372640)
autopkgtest (3.5.2) unstable; urgency=medium
[ Martin Pitt ]
* Make apt-get source and package unpack output a bit less verbose.
* Avoid "grep: debian/tests/control: No such file or directory" warning for
packages with implicit test control. (Closes: #762009)
* Adjust "unpack into temp dir" mode for multi-archified typelib locations
introduced in gobject-introspection 1.41.4-1.
[ Niko Tyni ]
* Test perl packages that include test.pl, not just t/. (Closes: #762013)
autopkgtest (3.5.1) unstable; urgency=medium
* Fix apt cleanup to remove configuration files before apt-get update.
* When using local binaries, tell apt-get to only download our
autopkgtest.list, not all apt lists again.
* Move autopkgtest gpg directory from ~/.autopkgtest/gpg to
~/.cache/autopkgtest/. Check the former location for backwards
compatiblity.
* Invoke gpg with --no-random-seed-file to avoid VCS noise when using
tests/home/.
* run-from-checkout: Use --gnupg-home with our tests/home to avoid GPG key
generation.
* adt-run: Defer GPG key creation until binary publishing time. This avoids
key creation entirely for the common case of not using any local or built
binaries.
* tests: Use apt-doc instead of cpp-doc, as the latter is in Debian contrib
only. (Closes: #761417)
* When determining whether a testbed reset is necessary in between tests,
also consider if recommends were/will be installed. (Closes: #761420)
* README.package-tests.rst: Packages with implicit test control (ruby/perl)
should use a different Testsuite: value, namely
autopkgtest-pkg-{ruby,perl}, to avoid breaking compatibility with the
specification and to other implementations like sadt.
* autopkgtest-reboot: Switch from SIGPIPE to SIGKILL, as that is much more
reliable.
* Reorganize NullRunner tests so that most run also as normal user (and
hence during package build); only keep four tests in NullRunnerRoot which
actually do need root privileges.
* Improve NullRunner.test_timeout_no_output test case to ensure that the
test is properly killed after timing out.
* Fix invocation of tests so that they retain the default SIGINT handler.
This changes cleanup after a timeout, add new procps dependency for this.
(Closes: #761596)
autopkgtest (3.5) unstable; urgency=medium
Improvements:
* VirtSubProc.cmd_reboot(): Save/restore downtmp instead of creating a new
one. This is more useful for reboot scenarios within tests.
* Support tests that want to reboot the testbed (for runners which support
rebooting). Document this in doc/README.package-tests.rst.
* tests/testdesc: Add test case for Debian packages without a test control.
* If there is no Debian test control file, synthesize one for Ruby and Perl
packages, using gem2deb-test-runner and pkg-perl-autopkgtest. Thanks to
Antonio Terceiro!
* ssh-setup/adb: Unlock the greeter before running tests.
Bug fixes:
* qemu: Reconfigure tzdata after copying /etc/timezone, to make that change
actually active.
* qemu: Wait an additional 3 seconds after the login prompt appears on
ttyS0. Without that, the system time is wrong, causing bad file
timestamps. Unfortunately polling runlevel or the "hwclock" status file
doesn't help.
* Always keep the same downtmp even after resetting the testbed. Some built
trees/tests use absolute path names and break when being moved around
after a reset. This already worked in most cases by accident due to a bug,
but now clean up the code to be explicit about it.
* Expand '@' to quasi-versioned (>= 0~) dependencies to avoid satisfying
them by virtual packages. (Closes: #761003)
* Ensure that the build tree is world-readable. Fixes permission error when
running adt-run as non-root with a different --user and a tight umask.
(Closes: #761049)
* lxc: Ensure the container root directory is accessible to non-root users.
Works around lxc-start-ephemeral bug with tight umasks.
(https://launchpad.net/bugs/1367730)
* lxc: Fix lxc-clone option (--new, not --name).
autopkgtest (3.4.1) unstable; urgency=medium
* adt-buildvm-ubuntu-cloud: Use host's http[s] proxy in the VM, rewriting
localhost addresses accordingly.
* qemu: Accept any runlevel between 2 and 5 as "booted", to also work with
systemd. (Closes: #760060)
* adt-run: Read built tree changelogs with UTF-8 encoding. (Closes: #759752)
* adt-virt-lxc: Source /etc/profile as well, to get $PATH in Debian
environments.
autopkgtest (3.4) unstable; urgency=medium
* Produce a summary file by default when --output-dir is given.
* ssh-setup/adb: Add missing $ADBOPTS for reverting to factory state.
* Add support for retrieving a click source from bzr (from the manifest's
x-source/vcs-bzr key) if --click-source is not given.
* ssh-setup/adb: Use SystemImage D-BUS call for factory reset instead of
direct recovery commands, to also work as user.
* ssh-setup/adb: Add --password/-p option to specify a different sudo
password than "phablet".
* adt-run: Fix $USER in tests if testbed does not have root.
* adt-run: Source shell profiles if testbed does not have root. Fixes
running click tests on Ubuntu Touch without a sudo password.
* ssh-setup/adb: Try common sudo passwords for CI ("phablet" and "0000"); if
those do not work, and none is specified with --password, tests cannot run
as root (which should be fine for most cases).
* ssh-setup/adb: Stop using system-image-cli to detect an Ubuntu phone, as
that does not work as user. Check for unity8 instead, like in other parts
of the code.
* ssh-setup/adb: Fix cleaning up of first-time wizard for non-root adb.
autopkgtest (3.3.3) unstable; urgency=medium
* ssh-setup/adb: Always parse options, so that $ADBOPTS is also defined for
operations other than "open". Fixes failure on close when multiple devices
are present.
* adt-run manpage: Fix references to documentation. (Closes: #758016)
* README.package-tests.rst: Point out that dpkg-source >= 1.17.11 adds the
Testsuite: field automatically. Thanks to Guillem Jover!
* ssh-setup/adb: Use PropertyService to enable ssh, for future compatibility
with non-root adb.
* ssh-setup/adb: Protect against failures of ssh-keygen when pruning the
localhost fingerprint. It fails if ~/.ssh/known_hosts does not exist.
* When copying down, properly fail when there are unremovable files in an
existing directory, instead of ignoring them and making the subsequent
copytree() fail. This helps to get better error messages for #757396.
* Fix crash on undeletable files in source trees. (Closes: #757396)
* ssh-setup/adb: If there is no default ssh key, create an
~/.ssh/id_autopkgtest and use that.
* Run autopilot tests with generating a subunit stream in
<output-dir>/artifacts/<testname>.subunit, and in verbose mode.
autopkgtest (3.3.2) unstable; urgency=medium
* When logging output to a file, properly close the "tee" processes' stdin
and wait for them to finish, instead of just killing them. This avoids
truncated logs on heavily loaded machines.
* Fix NullRunnerNoRoot.test_tmp_install test if gir1.2-json-1.0 is already
installed.
* Use long options for calling programs except for the most common ones, for
easier comprehension.
* qemu: Read all responses from ttyS1 after sending commands. Qemu 2.1 seems
to configure serial sockets differently, and not flushing responses
confuses it.
* qemu: Use "cache=unsafe" option to improve performance when the overlay is
not on a tmpfs.
* Further robustify cleaning up the chroots in the ChrootRunner tests.
* adt-build-lxc: When updating containers, move the new rootfs completely to
the original container. The rsync approach did not reliably update all
files.
autopkgtest (3.3.1) unstable; urgency=medium
* Accept comma separators in Tests:, Restrictions:, and Features: fields;
this is consistent with Depends: and avoids skipping tests. (LP: #1347958)
* Generalize implementation of getting an interactive shell in testbeds.
Most runners now don't need to provide a hook_shell() any more.
* Export install environment to interactive testbed shells, so that
unpacked test depends in the temp dir are accessible. (LP: #1339103)
* lxc: Don't fail on a nonexisting /etc/environment or /etc/default/locale.
(LP: #1348749)
* adb ssh script: Add --apt-update option to run "apt-get update" (with
temporarily switching to r/w), to run tests on older development images.
(LP: #1342838)
* Drop the very short and unnecessary install timeouts from the NullRunner
tests.
autopkgtest (3.3) unstable; urgency=medium
Improvements:
* Define "Classes:" (Debian) / "classes" (click manifest) test control field
in the specifications. For packages whose tests should run on particular/a
set of different hardware/platforms, this can specify a list of abstract
class names such as "desktop" or "graphics-driver". Consumers of
autopkgtest can then map these class names to particular
machines/platforms/policies.
* Fall back to searching for --setup-commands files in
/usr/share/autopkgtest/setup-commands/ if they are specified without path,
to reduce typing.
* Fall back to searching adt-virt-ssh's --setup-script in
/usr/share/autopkgtest/ssh-setup/ if they don't exist in the current
directory, to reduce typing.
Bug fixes:
* When running a test as user on a root testbed, run source extraction under
set -e, to fail early at the proper error. (LP: #1346388)
* Add test case for non-existing --apt-source package name.
* Add "installed" autopkgtest to run adt-run against a simple source tree in
$ADTTMP, to avoid using Python modules from the autopkgtest source tree.
This spots grave bugs like in 3.2.1.
autopkgtest (3.2.2) unstable; urgency=medium
* adt-run: Fix typo that broke --apt-source with "build-needed" tests.
autopkgtest (3.2.1) unstable; urgency=medium
* Ship forgotten adt_run_args module.
autopkgtest (3.2) unstable; urgency=medium
Improvements:
* Add --copy option to copy a file or directory into the testbed.
* Support specifying a click name like "com.example.myapp" with --click to
test an already installed click instead of a local *.click file.
(LP: #1338899)
* Introduce "autopilot_module" click test control field for specifying an
autopilot test module name. This allows tests to require additional test
dependencies, restrictions, etc. without having to specify the full
invocation of autopilot. (LP: #1338911)
* In "unpack into temp dir" mode, ignore Python 2.x test dependencies. They
are not exposed anyway as with $PYTHONPATH we can only serve one major
Python version, and we want Python 3.
* Port remaining runners from obsolete optparse to modern argparse. This
gives a consistent command line help and support for reading options from
a file with '@'.
* Port adt-run to argparse, and clean up the CLI:
- Drop rarely used and confusing options: --instantiate, --sources-tests,
--sources-no-tests, --built-binaries-filter,
--binaries{,-forbuilds,-fortests}. Add a single new --built-binaries as
the opposite of -B/--no-built-binaries.
- CLI parsing is now done in lib/adt_run_args.py, with separate unit
tests.
- This now allows any part of the command line (for adt-run itself or the
virt-server or both) to be read from a file, using the '@' prefix.
(Closes: #749241)
- Separate the options into different logical groups, in both --help and
the manpage.
* Make --override-control work for click manifests. (LP: #1342818)
* Introduce "Test-Command:" debian/tests/control field to run a shell
command instead of a "Tests" named executable. This avoids having to write
trivial one-liner scripts or allows wrapping calls to test executables.
(Closes: #741330)
Bug fixes:
* Set $USER environment variable also for tests that run as root.
(LP: #1337802)
* Add lintian override for ssh-setup/SKELETON, it's deliberately not
executable.
* Increase timeout for aa-clickhook workaround for LP#1337253, 100s is often
not enough.
* ssh-setup/adb: Disabling the first-time wizards sometimes fails due to adb
being brittle at that time. Retry up to 5 times.
* Optimize aa-clickhook workaround for LP #1337253 to only rebuild the
profile of the installed click app. This reduces the time from about a
minute to about a second.
* qemu, lxc: Use VirtSubproc subprocess wrappers to get timeout handling for
all commands.
* Rewrite build depends to not rely on builders being able to fall back to
alternatives (python >= 3.3 | python3-mock), to ease backporting.
* tests/pyflakes: Use pyflakes3.
* Only run pyflakes test during package build if pyflakes3 is available.
This unbreaks backports.
* Run "click info" as the same user as the test. (LP: #1342836)
* Keep <testname>-stderr output files for "allow-stderr" tests.
autopkgtest (3.1) unstable; urgency=low
Improvements:
* ssh-setup/adb: On Ubuntu phones, use phablet-config to disable first-time
wizards/intros. Implement reboot and revert. This requires phablet-config
now, either installed as package or from bzr checkout with
PHABLET_TOOLS_PATH=<checkout dir>.
* ssh-setup/adb: Save/restore NetworkManager system connections in revert.
* ssh-setup/adb: Add -r/--reset option to perform a factory reset before
running tests.
* Bring back --control-override option, with tests and documentation.
Bug fixes:
* Work around subprocess.Popen bug in Python 3.2.
* lib/testdesc.py: Change exceptions to be Python 3.2 compatible.
* Kill the actual test on timeout instead of just the wrapping shell script.
* Fix workaround for LP #1333215: Don't run with "su -l" as that causes
extra stdout spew from su for nonexisting home directories; instead just
source /etc/profile and ~/.profile as these are the parts that we need.
* Run aa-clickhook after every click install, otherwise it doesn't apply to
newly installed clicks. Works around LP #1337253.
* Open files which can potentially contain non-ASCII data with explicit
UTF-8 encoding, to avoid UnicodeDecodeErrors under C locale.
(Closes: #753582)
* Fix metavariable names in adt --help.
* ChrootRunner tests: Retry removal of temporary chroot when busy.
* Don't let a first failed apt-get update run break the test due to stderr;
it gets retried 15 seconds later.
autopkgtest (3.0.2) unstable; urgency=low
* Fix -s/--shell-fail to actually work again (regression in 2.19).
* Quiesce the sourcing of /etc/profile and ~/.profile to avoid "stdin: is
not a tty" clutter.
autopkgtest (3.0.1) unstable; urgency=low
* Install ssh setup scripts executable.
* doc/README.running-tests.rst: Add ssh runner example.
autopkgtest (3.0) unstable; urgency=low
Improvements:
* Add support for click packages (https://click.readthedocs.org):
- Add two new actions --click and --click-source to run tests of a click
package. For now, --click-source is mandatory as parsing/getting the
source from a click's x-source manifest entry is not yet implemented.
- Add doc/README.click-tests.rst to document test definition for click
packages.
- Add setup-commands/ubuntu-touch-session: --setup-commands script for
installing click, the Ubuntu SDK, some extra packages for autopilot, and
launching Xvfb and an user upstart session. This approximates the
environment of an Ubuntu touch device. You can also run
setup-commands/ro-apt afterwards.
* If the testbed does not have root (null runner or schroots without
root-user/group) or a writable dpkg database (as on Ubuntu system
images), fall back to unpacking test dependencies into a temporary
directory and set *_PATH environment variables accordingly. This will
make a limited number of test dependencies work in these test beds.
* Add setup-commands/ro-apt for making dpkg/apt readonly, to
approximate what happens with Ubuntu system images.
* Add adt-virt-ssh runner: Generic runner for an externally set up testbed
which assumes nothing else than a working ssh connection. This can call a
"setup script" to create/configure a testbed (such as spinning up a cloud
VM with nova or setting up SSH on a phone through ADB). See the manpage
for details. Thanks to Jean-Baptiste Lallement!
* Add ssh-setup/SKELETON for a basis to write ssh setup scripts.
* Add ssh-setup/adb: Set up an ssh connection to an adb host. If that adb
host is an Ubuntu Touch system, it also does some extra configuration like
disabling the screen timeout and allowing Autopilot to introspect running
apps. See header commends for details. Thanks to Jean-Baptiste Lallement!
* Add ssh-setup/nova: Sets up a nova instance to use as an autopkgtest
testbed. It assumes that the host system is already prepared to run nova
commands. This is not very robust at the moment and more like a proof of
concept. Thanks to Jean-Baptiste Lallement!
* Convert doc/README.* to reStructuredText and rename to *.rst. Build HTML
files from them using rst2html and install them also.
* adt-run: Never run testbed commands with newlines in them. Those make it
ridiculously hard to implement runners which can only accept shell
strings.
Bug fixes:
* lxc: Don't leak the host environment into the container (lxc-attach
currently defaults to that), but build a clean and minimal environment
from /etc/environment and /etc/default/locale.
* Fix calling a virt server with full path.
* lib/VirtSubProc.py, downtmp_remove(): Only remove the downtmp if it is
actually set. This avoids confusing error messages if something fails
during early test bed setup.
* adt-run: set APT_LISTBUGS_FRONTEND=none when installing packages. This
will avoid the apt-listbugs prompt when apt-listbugs is installed.
Thanks to Antonio Terceiro!
* schroot: Ending the schroot session sometimes fails on "Device or resource
busy". Retry 10 times before giving up instead of failing immediately.
* adt-run: Read /etc/profile and ~/.profile when running tests as user. This
emulates the bits of "su -l" that we want, until this gets cleaned up for
using "su -l" completely.
* tests/run-parallel: Get sudo password early, to avoid asking for it in
background jobs.
* Restrict --apt-pocket to *.ubuntu.com and *.debian.org sources, as
third-party repositories such as launchpad.net don't (necessarily) have
pockets. (LP: #1334332)
* Fix high adt-virt-* CPU usage on reading commands on stdin.
autopkgtest (2.20) unstable; urgency=medium
Improvements:
* Add "Test" argument/member for a shell string command, instead of
specifying a path to a test executable. Not currently supported by Debian
parser.
* Convert doc/README.* to Markdown, so that it's easy to format them as PDF
or HTML, e. g. with pandoc.
* If the --user option is given or the testbed specifies a normal user, pass
this to --setup-commands through $ADT_NORMAL_USER.
* Allow --setup-commands which would ordinarily reboot the testbed to
suppress this by creating a stamp file in /run. This is useful for
commands which run processes that aren't started at boot.
* Default to C.UTF-8 locale instead of C. This is supported in Debian >= 7
(wheezy) and Ubuntu >= 12.04 LTS, and much closer to reality than C.
* If the testbed does not have root (null runner or schroots without
root-user/group) or a writable dpkg database (as on Ubuntu system images),
fall back to unpacking test dependencies into a temporary directory and
set *_PATH environment variables accordingly. This will make a limited
number of test dependencies work in these test beds.
Bug fixes:
* Don't include "adt-satdep" dummy package in <test>-packages output files.
* adt-buildvm-ubuntu-cloud: Use host timezone as default for the VM, to
avoid TZ changes during boot when running tests. Thanks Jean-Baptiste
Lallement!
* Run tests with specified locale also when they run as user.
* Fix test failure with Ubuntu's pkgbinarymangler installed.
autopkgtest (2.19) unstable; urgency=medium
Improvements:
* Support new argument type --changes (implied if file name ends with
".changes"). This behaves as if you specified all debs and the dsc from
the given binary changes file and implies --no-built-binaries. I. e. this
is the right thing to test a local package that you just built.
(Closes: #738552)
Bug fixes:
* schroot: Always advertise "revert" capability, as it implements revert
with simply close()/open().
* adt-run: Move warning about missing root on testbed into the correct
place.
* Install build-essential and build dependencies in one apt run, which is
slightly faster.
* Explicity specify UTF-8 encoding for reading/writing Debian control files,
to fix UnicodeDecodeError under C locale.
* Determine testpkg-version also for --built-tree tests.
* schroot: Refuse to work with non-ephemeral schroots, i. e. "directory"
type schroots without union-type. Point this out in the manpage, too.
(Closes: #751574)
* schroot: Verify that /proc is mounted, and abort with a proper error
message if not. (Closes: #751575)
* adt-buildvm-ubuntu-cloud: Don't start cloud-config with a leading newline,
not yet supported in Ubuntu 12.04's cloud-init. Thanks Jean-Baptiste
Lallement! (LP: #1330950)
* adt-buildvm-ubuntu-cloud: Don't use write_files cloud-init command, not
yet supported in Ubuntu 12.04's cloud-init. Replace with runcmd echos.
* adt-buildvm-ubuntu-cloud: Purge cloud-init and shutdown after cloud-init
is done, not while it's still running.
* adt-buildvm-ubuntu-cloud: Create upstart job for ttyS0 console unless it
already exists, for compatibility with Ubuntu 12.04.
* adt-virt-qemu: Use python3 or python for eofcat depending on which one is
available. Fixes operation with older Ubuntu versions which don't yet
install Python 3 by default.
Code cleanup:
* Unify logging functions into lib/adtlog.py, make their API easier and more
consistent, and use them in adt-run and all adt-virt-*.
* Move test control parsing and keeping data into lib/testdesc.py. This
eases factorization, abstraction for future support of other source
package formats such as click, and testing. Add new tests/testdesc
unittests. This also drops the "tree0t-" like prefixes from test names in
the results and the output file names.
* Don't read --setup-commands files multiple times.
autopkgtest (2.18) unstable; urgency=medium
Improvements:
* adt-run: When running from git, show the current SHA and commit summary of
HEAD.
Bug fixes:
* tests: Fix race in cleaning up ChrootRunner mounts.
* adt-run: If apt-get update fails on publishing built binaries, retry after
15 seconds.
* VirtSubproc.cleanup(): Avoid recursion if something in hook_cleanup()
bombs.
* Fix --shell to also work with --output-dir. (LP: #1317078)
Code cleanup:
* VirtSubproc: Document and cleanup execute*() functions.
* Some test robustifications, to avoid making too many assumptions about
debug output.
* adt-run: Eliminate usage of testbed "execute" command, which is too
complicated and limiting. Use "print-execute-command" directly instead.
* Drop "execute" command.
* adt-run: Unify Testbed.command() and commandr() methods and make them
stricter.
* Add prefixes to all temporary directories.
* Remove autopkgtest-xenlvm. It has been orphaned since ~ 2008, few to no
current users, and no maintainer. (Closes: #508660, #680122)
* Simplify build system.
* Drop concept of "shstring". All runners must provide an auxverb now,
possibly with a wrapper and quoting by themselves if their implementation
can only deal with shell commands. This avoids having to tell apart these
two cases in adt-run, clarifies the documentation, and greatly simplifies
the code. (Closes: #750544)
* qemu: Use low-level IO in runcmd. This avoids all the Python overhead and
buffering and is less prone to bytes vs. string issues.
* qemu: Fix text vs. binary confusion in reading and copying /etc/timezone.
* schroot: Call schroot command with -q, to avoid stderr spew if the schroot
has "message-verbosity=verbose".
* Port to Python 3.
autopkgtest (2.17.1) unstable; urgency=medium
* Make $ADTTMP readable and accessible to other users, for tests which use
other users through "su" and similar.
* adt-virt-qemu: Set $HOME for root also after rebooting.
* tests/adt-run: Force stdout to UTF-8 to see proper failures even in
non-UTF-8 environments, instead of crashing with UnicodeEncodeError.
* adt-run: Flush input to the stdout/err cat processes and give them some
time to catch up after the end of the test, to avoid losing the last bits
of output. (Closes: #750343)
* Add missing build-essential and debhelper test dependencies.
* adt-run: Ensure the cleanup handler for the tests tree runs also if
copying up the tests tree fails. (Closes: #749655)
* Put back the LXC runner on Debian, so that folks with manual LXC container
setups can still use it.
* adt-build-lxc: Check for default networking in /etc/lxc/default.conf. If
it's "empty", abort with an error message that container networking needs
to be set up first. Until LXC gets fixed in Debian (#747914) you need to
manually set up a bridge, dnsmasq, and the default config.
* adt-build-lxc: Support specifying $ADT_APT_PROXY. If apt-cacher-ng is
running locally, use that as a default.
autopkgtest (2.17) unstable; urgency=medium
New features:
* adt-buildvm-ubuntu-cloud, adt-setup-vm: If apt-cacher-ng is running
locally, use that as a default for --proxy/ADT_APT_PROXY.
* Add "reboot" testbed capability and implement it for QEMU (most other
runners don't support it conceptually). Reboot the testbed if the setup
commands affected boot relevant directories (/boot, /etc/init.d,
/etc/init/, or /{etc,lib}/systemd/system). (LP: #1324435)
Bug fixes:
* adt-setup-vm: Run the initial apt-get update without fetching
translations.
* Stop shipping the LXC runner and associated tools on Debian. It does not
work out of the box until Debian's LXC containers get proper networking by
default (see Debian #747914).
* adt-buildvm-ubuntu-cloud: Add LSB header to generated SysV init script to
avoid a warning from insserv.
* VirtSubproc: Fix generation of auxverb from shstring.
* Drop debhelper dependency, not necessary any more since 2.16.
autopkgtest (2.16.4) unstable; urgency=medium
* adt-build-lxc: Unset $http[s]_proxy for the post-creation apt-get update
if it refers to localhost.
* adt-build-lxc: Fix cache cleaning for upgrades.
autopkgtest (2.16.3) unstable; urgency=medium
* adt-buildvm-ubuntu-cloud: Explicitly use virtio ethernet device, as the
default one is horribly slow with older QEMU versions.
* adt-virt-qemu: Add -c/--cpus option for running VMs with multiple CPUs.
* adt-run: Retry writing to stderr a few times on EAGAIN.
* adt-run: If apt-get update fails, retry after 15 seconds. This should
work around the frequent "hash sum mismatch" race condition.
(LP: #1319416)
* adt-buildvm-ubuntu-cloud: Configure dpkg to use unsafe io, to speed up
tests which run on hard disk backed overlays.
* Simplify the pipelining of test stdout/err, and stop using explicit pipes
and the "downtmp-shared-fifo" capability. This provides live output for
all runners, including qemu. (LP: #1320194)
* Drop obsolete downtmp-shared-fifo capability from all runners.
* Fix "Operation not permitted" errors on cleaning up empty stdout/err
result files with shared downtmp runners.
* adt-buildvm-ubuntu-cloud: Refine init.d script to only launch a shell on
"start".
* adt-build-lxc: Add optional third argument for specifying a non-native
architecture.
autopkgtest (2.16.2) unstable; urgency=medium
* Support ":native" dependencies.
* Add LXC runner test case for running as different user.
* Add QEMU runner test case for performance and integrity of copying (1000
files with 100 kB each).
* copyup/down: Clean up cp process on timeout, to fix eternal hang.
* adt-virt-qemu: Disable shared downtmp. For bigger source trees 9p in
current QEMU is way too slow (in the magnitude of 2 MB/s, and max'ing out
CPU usage), so we can't use it directly as a normal file system.
* Revert "Don't extract/build source packages in shared downtmp", due to
above change. This stops penalizing other runners with extra source tree
copies.
* VirtSubproc: Clean up auxverb processes on timeout.
* adt-run: Stop testbed on TERM and QUIT signals, to avoid leaving stale
schroot/qemu/LXC processes around when e. g. stopping a job from Jenkins.
* adt-virt-qemu: Fix runcmd to make the command's stdin actually look and
behave like a pipe, instead of reading directly from the "stdin" file. The
latter leads to commands exiting prematurely when hitting the file's EOF.
(LP: #1317358)
* Convert one of the QEMU tests from source tree to dsc, to cover that case
as well.
* Add test case for QEMU with --apt-source (using libpng).
* adt-virt-qemu: Create eofcat in /bin instead of the shared /autopkgtest
mount, as running it from the latter again triggers the "invalid numerical
value" bug with 9p from older QEMU versions.
* adt-buildvm-ubuntu-cloud: Purge some unnecessary services to speed up boot
time a bit.
* @builddeps@ now includes build-essential, not just make. (LP: #1317357)
* Make -s/--shell-fail start a shell on failed builds as well.
(LP: #1317055)
autopkgtest (2.16.1) unstable; urgency=medium
* adt-buildvm-ubuntu-cloud: Configure vmalloc=512M on i386 for the udisks2
autopkgtest.
* adt-buildvm-ubuntu-cloud: Disable downloading of package translations.
* Create testpkg-version and testbed-packages before starting the builds, to
also have them when builds fail.
* Fix VirtSubproc.copyup_shareddir() for testbed paths which are not already
in the shared downtmp.
* When cleaning up a temporary testbed path fails with a permission error,
clean it in the testbed (where we usually have root privileges). This
avoids lots of "ignored OSError" spam with the schroot runner.
* Stop assuming that we can chown files in the shared downtmp, it's not true
with QEMU's 9p file system. Fall back to chmod'ing to world-writable files
as with 9p there is no other way to implement this.
* Simplify source tree/package extraction logic.
* adt-run: Don't extract/build source packages in shared downtmp, to avoid
permission/ownership problems with limited implementations like 9p.
* adt-virt-qemu: Re-enable shared downtmp, now that operation with different
UIDs is fixed.
* Don't create $ADTTMP in shared downtmp to avoid permission/ownership
problems with limited implementations like 9p.
* adt-virt-qemu: Fix exception if workdir is not currently set.
* adt-virt-qemu: Increase timeout for login prompt from 2 to 5 minutes, for
slow VMs or machines under heavy load.
* Improve error handling on copy timeouts.
* adt-virt-qemu: Remove overlay file as early as possible, to avoid leaking
large files in error conditions. Let QEMU run with the deleted inode.
* Don't run apt-get and setup commands with sending output to /dev/stdout.
When redirecting adt-run's output directly into a file this causes a
seek to start and thus partially overwriting the output file. This also
puts the apt-get output back to stdout as it used to be before 2.15.
(Closes: #747041)
autopkgtest (2.16) unstable; urgency=medium
* adt-buildvm-ubuntu-cloud: Use "-monitor null" instead of "-monitor none",
as the latter does not work with older Qemu versions.
* adt-virt-qemu: Don't call qemu-img with -q option, it doesn't exist in
older Qemu versions yet. Redirect stdout instead.
* tools/adt-setup-vm: Disable downloading of package translations to save
some bandwidth. Thanks Jean-Baptiste Lallement.
* adt-virt-qemu: Fix race condition that caused a ValueError when the exit
code file was not written yet.
* adt-virt-qemu: Automatically detect a normal user to run tests as, unless
-u is given. This works similarly to -lxc.
* Add tests for adt-virt-user with non-default users.
* adt-virt-qemu: Disable shared downtmp for now, it breaks on older Qemu
versions and the current setup does not work with different UIDs in the
testbed.
* adt-virt-qemu: Use file locks for host-side ssh port forwards to avoid
race condition.
* adt-run: Drop usage of dh_listpackages, it needs debhelper which is an
undeclared (and also unwanted) dependency. Parse debian/control directly
and make use of our existing arch tag resolution. (LP: #1312276)
* adt-virt-qemu: Copy host's timezone into VM to avoid time zone skew when
copying files.
* adt-virt-qemu: Show determined normal user instead of <user> in ssh
message, to get a fully copy&pasteable command.
* adt-virt-qemu: Ensure tests which run as root have $HOME set.
* Fix --proposed option to also apply to sources.list.d/* lists and to
multiple sources. (LP: #1313278)
* adt-buildvm-ubuntu-cloud: Install haveged, to avoid failing tests due to
running out of entropy.
* adt-virt-qemu: Add --qemu-options for passing additional options to QEMU.
Thanks Diogo Matsubara!
autopkgtest (2.15) unstable; urgency=medium
* Fix --apt-source mode to download the source with apt-get's
"--only-source" option, to actually get what we asked for.
* Add missing python3 build dependency for tests/adt-run. Fixes FTBFS of
backports.
* adt-buildvm-ubuntu-cloud: Add options to pass custom metadata/userdata.
Thanks Diogo Matsubara.
* Replace pbuilder-satisfydepends-classic with parsing and satisfying build
and test dependencies with libdpkg-perl and "apt-get -f install". This
gets rid of the pbuilder dependency and some nasty hacks like the
--internal-chrootexec script, and allows us to be more flexible such as
also supporting Recommends: installation. (LP: #1205110)
* Fix "specifiy" typo in error message. Thanks Jelmer Vernooij.
(Closes: #745413)
* adt-build-lxc: Stop using the -F template option to flush the cache as it
is not available in the debian template. Remove caches manually instead.
(Closes: #745415)
* adt-buildvm-ubuntu-cloud: Fall back to latest stable if there is no
development series.
* Add "needs-recommends" restriction for installation of recommended
packages for the test dependencies. This does not affect build
dependencies. (Closes: #708963)
-- Martin Pitt <email address hidden> Tue, 24 Nov 2015 10:39:08 +0100