Iterator should raise StopIteration, not return None.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Original implementation compared self and then parent node. Parent
node again compared self and children => infinite recursion.

In the end we actually don't want to compare parents, because we care
about what is after this node. This implementation compares only Node
and children.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

 Run avocado when called by 'python -m avocado'

selftests..multiplex_tests: Check that vars changes

Replace test_run_mplex_timeout with test_run_mplex_params test which
executes 'env_variables.sh' and checks that output contain correct
output. This test checks the multiplexer capability to change variable
value.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Add a main entry point when avocado is called
by the command line option 'python -m avocado'.
Signed-off-by: NRudá Moura <rmoura@redhat.com>

Plugin tweaks

Put the sysinfo and test_lister plugins into separate
plugin files. Although not strictly necessary, the
neat 1 plugin - 1 source file relation makes things
organized and easy to understand.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

This way we define more clearly the attributes of the
plugin object (many of them will implement a parser
object).
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Get to a more toned down output for the subcommands
'datadir', 'plugins' and 'list'.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

It is useful information to the user where the config file
is, to reinforce where avocado is picking the default values.
So add it to the output of `datadir` and `list` subcommands.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

multiprocessing.Queue().put() is asynchronous and in case the test died
critically before the put() was executed, TestRunner.run() waited for
infinity for early_state. This patch uses
multiprocessing.queues.SimpleQueue(), where it's guaranteed that after
executing put() the data are already in the SimpleQueue.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

avocado.test: Support params passing to DropinTests [v2]

This patch injects all params as env variables to
DropinTests. This is very useful for multiplexing
these tests.

Additionally the env_variables.sh example test was
adjusted to print CUSTOM_VARIABLE, which can be used
in selftest.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Make use of python-subprocess32 module if available.

Include wrappers script when creating `avocado-examples` package.

python-subprocess32 [https://code.google.com/p/python-subprocess32/]
is a backport of Python 3.2 subprocess module with:

* Better process forking.
* More reliable when running with threads.

Although we're not going to use any feature of subprocess32,
this change turn it possible to use subprocess32 as a drop-in
replacement for the original subprocess module.
Signed-off-by: NRudá Moura <rmoura@redhat.com>

Include all examples/wrappers/*.sh scripts when building
the `avocado-examples` package. The scripts will live under
`/usr/share/avocado/wrappers` when installed.
Signed-off-by: NRudá Moura <rmoura@redhat.com>

Fix rpmbuild

setuptools is not a standard module and makes things
unpredictable and forces us to push MANIFEST.in to
include more files so the binary distributions get
filled appropriately. In a retrospect, it was a bad
idea. Let's revert it.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Fix abort test taking too long to execute

It is likely that a race condition in avocado makes the abort
test to take a long while executing (24 hours), but that
was only identified in the Travis CI environment. Instead
of taking that long, we should end the abort test quickly,
after a test timeout. This will allow us to continue to
see the failure, only that the rest of the tests will be
executed, and no penalties to Travis due to inefficient
builds.

Of course, we still need to find the actual bug and fix it.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>

Fixes a sloppy copy and paste I made in the spec file.
Signed-off-by: NLucas Menghel Rodrigues <lmr@redhat.com>

GDB: use gdbserver --server-sdtderr option if available

GDB remote protocol support [v2]

A new command line option is being added to upstream GDB. This new
option allows gdbserver's own stderr (but not its inferior process
stderr) to be redirected to another file.

If that feature is available on the about to be used gdbserver, then
let's make use of it. This redirects its stderr to /dev/null, because
avocado itself has no real use for gdbserver's own stderr output.
Signed-off-by: NCleber Rosa <crosa@redhat.com>

GDB talks to targets via a so-called remote protocol. These targets
can be things like gdbserver, ROM monitors on specific hardware boards,
or applications that implement the same protocol.

One noteworthy example of application is QEMU, which implements some
basic GDB Remote commands and also a few extenions of its own.

This initial implementation allows connecting to the remote target,
sending commands and receive results. The API can certainly grow later
based on demand.

Changes from v1:
 * Fix a typo in the REMOTE_MAX_PACKET_SIZE comment
Signed-off-by: NCleber Rosa <crosa@redhat.com>

It turns out that, in some conditions, such as when immediatelly trying
to communicate with a gdbserver process after it is executed, results in
failure.

The reason seems to be that gdbserver itself takes "some" time between
being executed and binding to the TCP port.

While prototyping GDB remote client in pure Python, this kind of race
condition was found.

Now, by default, there will be a 2 seconds timeout and a explicit check
for gdbserver's "ready to use" status.

Changes from v1:
 * added documentation for the __init__ method
Signed-off-by: NCleber Rosa <crosa@redhat.com>

multiplex_test uses sleeptest to check it supplies the values correctly.
This patch uses passtest instead.

For now it uses the same multiplex file and in `test_run_mplex_timeout`
still uses sleeptest, but there is FIXME message to use envtest once
it supports params passing (another pull request)
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

CLoader is C (faster) version of the Loader. In case it's unavailable
it fall-backs automatically to the default Loader.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

Instead of creating OrderedDicts and then converting it into
tree structure, this modifies the `mapping` constructor to create
the tree structure directly. This way we only walk through the yaml
file once.

Also remove the `tree_unittest.TestReadYAML` selftest as it lost it's
meaning. Valures are checked in the TestTreeNode selftest.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

When initializing TreeNode we should use the add_child function
in order to set the parent correctly.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

.travis.yml: Run selftests in verbose mode

In case of timeout we might want to get the name of the last running
test, therefor setting verbose mode for them.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

.travis.yml: Return apt-get update

Since we have to install extra packages in the VM, some
of which might have been moved/updated, apt-get update
turns out necessary.
Signed-off-by: NLucas Meneghel Rodrigues <lmr@redhat.com>