It is definitely handy to be able to disabdle certain tests (e.g. tests that are
known to hang or crash the test machine on certain kernels), so re-introducing
the capability of avoiding certain tests just by placing them in a file is
useful.
Introduce a command line option to specify the expunged file name. The file will
exist in each tests/* sub-directory so that only the testname is required, and
can be managed independently. The use of a command line parameter allows
multiple expunge files to exist simultaneously in the one xfstests tree which
simplifies management of a source repository used for multiple versions of a
distro.
Typical usage:
$ cat tests/generic/3.0-stable-avoid
280
$ sudo ./check -X 3.0-stable-avoid generic/280
FSTYP -- xfs (debug)
PLATFORM -- Linux/x86_64 test-1 3.0.39-dgc+
MKFS_OPTIONS -- -f -bsize=4096 /dev/vdb
MOUNT_OPTIONS -- /dev/vdb /mnt/scratch
generic/280 [expunged]
Passed all 0 tests
$
Eventually we should be able to automate setting up expunged files based on
distro release or kernel version through this infrastructure.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Introduce a top level common directory and move all the common.*
files into it. Because there is now a directory named common, the
prefix can be dropped from all the files. Convert all the tests to
use this new directory for including common files.
for f in common.*; do \
git mv `echo -n "$f " ; echo $f | sed -e 's;n\.;n/;'` \
done
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
[rjohnston@sgi.com reworked for TOT changes]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Currently the callers pass $seq, rather than the full path to the
test. As a result, it creates the link in the top level directory
and it points nowhere. Fix it to create the link in the correct
place.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Essentially the change is simply this. Converting:
... >> $seq.????
to:
.... >> $RESULT_DIR/$seq.????
so that output files are directed to the defined output directory.
sed to the rescue:
$ sed -i -e '/^seq=.*$/a seqres=$RESULT_DIR/$seq' -e 's/seq.full/seqres.full/' tests/*/*
will do most of the work automatically.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
[rjohnston@sgi.com reworked for TOT changes]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Currently each test gets it's sequence number from it's name. It
separates this from the path via basename, and uses it for
outputting full, notrun and other status/log files. Hence these end
up in the top level directory.
All these output files need to go somewhere other than the top level
directory. Right now the check script is looking for them in the new
test directories (e.g. for the notrun files), but it would be good
to be able to separate the test source form the test output.
Hence create an output directory which has a similar heirarchy to
the test source directory. Create it on demand when we build the
list of tests to run if it doesn't already exist.
Change the high level check script to set up this variable
appropriately for each test that is being run, and to use this new
output directory for it's result files as well. The next commit will
change all the tests themselves to use the new output directory.
This is the first (small) step in being able to store test results
in an external location for archival/data mining purposes.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
[rjohnston@sgi.com rm whitespace and fix typos in commit message]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Now that there are no tests in the top level directory, the exlude
group functionality no longer works (-x group) as it cannot find
tests to build an initial list of tests. Rework the the exclude
group implementation to operate correctly on the new test locations.
Also, switch from sed to grep for exclusion because sed has problems
distnguishing the '/' in path names from regex control...
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Now that tests are in separate subdirectories, they cannot be selected purely by
test number on the command line. We need to specify the test by subdir and test
name, so effectively we move to specifying them by regexes on the command line
rather than by expanding a range internally. This is needed to support
non-numeric test names as well, so the change may as well be made here.
This means the command line parsing needs to change from trying to detect tests
by a regex match to a processing loop that simply parses the tests and checks
for there existence. Hence the moment we find an argument that is not a switch
(i.e no preceeding "-"), we move from the arg processing loop to the test name
processing loop. IOWs, tests must be specified last on the command line.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
These are tests that are shared between multiple filesystems (moved
to shared), and udf/btrfs/ext4 specific tests, moved to appropriate
directories.
I created the "shared" directory to indicate tests that are not
truly generic, but also not filesystem specific. They might rely on
a feature that is only implmented in a few filesystems and so can't
be truly generic.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
[rjohnston@sgi.com reworked for TOT changes]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
These files are generated by the tests, they don't belong in
the repository.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Create a specific tests/ subdirectory to hold all the tests so they
are no longer need to be in the top level directory. This patch does
not move any tests there, however, and that will be done in
subsequent commits.
The tests/ subdir will have it's own subdirectories for different
classes of tests. Initially, there will be a per-FSTYP subdirectory
for filesytsem specific tests, and a generic directory for tests
that span multiple filesystems.
Each class will have it's own group file to indicate what groups the
tests belong to, and these will be parsed appropriately by the high
level check script to build the test list.
The change in parsing results in the test output also emitting the
path to the test as well as the name of the test, instead of just a
raw number. This allows duplicate test names in the sudirecotries to
be unambiguous when the summary is written out.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Check supports an expunge file for tests - a way of marking tests
not to run even if you specify it to run. Use is like this:
$ echo 002 > expunged
test-2:~/src/xfstests-dev$ sudo ./check 001-003
FSTYP -- xfs (debug)
PLATFORM -- Linux/x86_64 test-2 3.5.0-rc1-dgc+
MKFS_OPTIONS -- -f -bsize=4096 /dev/vdb
MOUNT_OPTIONS -- /dev/vdb /mnt/scratch
001 4s ... 4s
002 - expunged
003 0s ... 0s
Ran: 001 002 003
Not run:1
Passed all 2 tests
This is arguably useful(*), but gets in the way of splitting up the
tests into multiple directories. Remove it for now, but such
functionality should be considered for re-implementation at a later
date.
(*) e.g. having a different "expunged" file for each distro release
you have to test to avoid the tests known to fail due to fixes or
features that will never be back ported to older releases....
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
This is an old script from the auto-qa days back at SGI. It no
longer is in use or, AFAIK, ever been used for xfstests. If anyone
needs it, they can pull it back out of git, so lets remove it to
simplify check.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The options are parsed in a messy loop of option parsing and
actions on secondary arguments. Turn it into something much neater
and easy to understand rather than a mess of temporary variables
and tortured logic...
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
[rjohnston@sgi.com fixed typos in commit message]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
There is only one user of the common file now - check. Fold the two
into one file as the split of functionality is not necessary
anymore.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Left over from the days of ASG, all stale. Remove them, leaving just
an simple example that defines all the variables that can be
configured. Also, add a localhost defined config set up for simple
KVM based guest test configs using virtio for their block devices to
simplify test setup in such scenarios..
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Back many years ago, the owner field was used to email the test
owner when auto-qa failed that test. It is not needed anymore - if
you want to know who wrote the test, look it up in git....
Script used was:
$ sed -i -e "/^# creator/d" -e "/^owner/d" [0-3][0-9][0-9]
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
[rjohnston@sgi.com added TOT changes]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The benchmark framework inside xfstests is basically unused,
bitrotted and not very useful. If we need benchmarks, lets use a
real benchmark framework, not xfstests. Kill it to remove
dependencies on common and common.rc.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
We don't ever do wholesale rebuilds of the output files anymore, so
kill the script that does it to reduce the dependency on common and
common.rc.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
_require_nobigloopfs has been removed.
We should use _require_no_large_scratch_dev in the test script.
Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Test 289 ignored the fact that historically journal is not accounted as
fs overhead in ext3. For larger filesystems it is hidden in 1% tolerance
but for filesystems smaller than 12G the test fails. So make the
counting precise to work everywhere.
CC: Eric Sandeen <sandeen@sandeen.net>
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
[rjohnston@sgi.com: add lower case units to filter]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The backref resolver test for btrfs was run on a static file system so far.
Resolving backrefs on a busy file system is what happens in reality and that
is what should be checked by this test.
I added a parameter such that the script can easily be changed to the former
behavior for development purposes.
Furthermore, this increases two constants which make the test simply cycle a
few seconds longer, increasing the chance to hit on something suspicious in
case we broke something.
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Dave Chinner
Zhi Yong Wu
Wang Sheng-Hui
Jan Kara
Jan Schmidt