The motivation is to eventually be able to get an effective/compatible
message interchange protocol version number from such a set.
As of this patch, the API is available, but it's not used anywhere.
A component descriptor set is a shared object.
You add a component descriptor to a component descriptor set with
bt_component_descriptor_set_add_descriptor(). Internally, the descriptor
gets appended to one of the source, filter, or sink descriptor array to
eventually have bt_component_descriptor_get_*_descriptor_count() and
bt_component_descriptor_set_borrow_*_descriptor_by_index_const()
functions if needed.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I337d4c596bd988b20849700daf9747a3f188450d
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1868 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
Philippe Proulx [Mon, 12 Aug 2019 19:46:05 +0000 (15:46 -0400)]
bt2: py_plugin.py: cache `_PluginInfo` objects
This patch makes _try_load_plugin_module() cache the `_PluginInfo`
objects it creates. The cache, `_plugin_infos`, maps a Python plugin
absolute path to a `_PluginInfo` object.
This avoids loading the Python module twice, which can create problems,
for example destroying component classes earlier than expected when the
new module replaces the old one in `sys.modules`.
Simon Marchi [Mon, 12 Aug 2019 18:37:58 +0000 (14:37 -0400)]
bt2: raise an exception if an AutoSourceComponentSpec produces no component
We don't do anything special at the moment if an AutoSourceComponentSpec
passed to TraceCollectionMessageIterator produces no component. The
created graph will have no source component and the message iterator
will produce no message:
In [5]: list(bt2.TraceCollectionMessageIterator('/yomadame'))
08-12 14:43:03.938 20284 20284 W CLI-CFG-SRC-AUTO-DISC auto_discover_source_components@autodisc.c:755 No trace was found based on input `/yomadame`.
Out[5]: []
This patch changes the behavior to be a bit more strict. If any
AutoSourceComponentSpec passed to TraceCollectionMessageIterator ends up
producing no component, an exception is raised. In the following example, the
directory `/home/smarchi/lttng-traces` does contain some valid traces, and
therefore produces some components, whereas the string `/yomadame` does not
produce any.
In [2]: bt2.TraceCollectionMessageIterator(['/home/smarchi/lttng-traces', '/yomadame'])
08-12 14:45:06.433 24428 24428 W CLI-CFG-SRC-AUTO-DISC auto_discover_source_components@autodisc.c:755 No trace was found based on input `/yomadame`.
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
<ipython-input-2-107461e1b36b> in <module>
----> 1 bt2.TraceCollectionMessageIterator(['/home/smarchi/lttng-traces', '/yomadame'])
~/build/babeltrace/src/bindings/python/bt2/build/build_lib/bt2/trace_collection_message_iterator.py in _auto_discover_source_component_specs(auto_source_comp_specs, plugin_set)
194
195 msg = 'Some auto source component specs did not produce any component: ' + ', '.join(unused_inputs)
--> 196 raise RuntimeError(msg)
197
198 return comp_specs
RuntimeError: Some auto source component specs did not produce any component: /yomadame
Change-Id: I6c6df9fad18915e294559948812a8213596affdf Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1885 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Mon, 12 Aug 2019 16:25:59 +0000 (12:25 -0400)]
tests: use os.pathsep when appending to BABELTRACE_PLUGIN_PATH
On Windows, we expect BABELTRACE_PLUGIN_PATH entries to be separated
with ';', not ':' (like PATH). Python conveniently provides an
OS-dependent definition for this.
Change-Id: Ie53af04e933258963b8495c72e7607a623e83ddb Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1883 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Philippe Proulx [Sun, 11 Aug 2019 15:33:40 +0000 (11:33 -0400)]
test_plugin.py: do not search in user/system directories for tests
In `test_plugin.py`, make bt2.find_plugin() not search in user and
system directories so that it needs to find the specific plugin in the
test's search path (given by `BABELTRACE_PLUGIN_PATH`) or fail.
Simon Marchi [Sun, 11 Aug 2019 02:31:41 +0000 (22:31 -0400)]
Fix: bt2: replace bt2.CreationError instances with bt2._MemoryError
Commit
4acc866e8298 ("bt2: rename CreationError to MemoryError, handle it in and out of Python bindings")
removed bt2.CreationError in favor of bt2.MemoryError (now
bt2._MemoryError). However, some usages of bt2.CreationError were
introduced afterwards, probably by patches that were in the pipeline.
Replace them with bt2._MemoryError.
Change-Id: Id3cb6542e507b1e41c165ca61df21069d72fb44f Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1867 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Mon, 5 Aug 2019 18:45:38 +0000 (14:45 -0400)]
bt2: add auto source discovery support to TraceCollectionMessageIterator
This patch makes TraceCollectionMessageIterator capable of automatically
detecting source components from given input strings, similar to how the
command line interface does.
A new type AutoSourceComponentSpec is introduced. The user can
instantiate this type, passing a mandatory input string (equivalent to a
non-option argument on the CLI). The user can then pass instances of
AutoSourceComponentSpec in the source_component_specs list when creating
a TraceCollectionMessageIterator. In
TraceCollectionMessageIterator.__init_, we run the automatic source
algorithm on all inputs passed through AutoSourceComponentSpec
objects. We create ComponentSpec from the results of the auto discovery
and append them to the source component spec list. The rest of the
TraceCollectionMessageIterator then works as before.
The user can attach params and log levels to AutoSourceComponentSpec
inputs. The semantic for those is the same as for the CLI. There is
also an `obj` parameter, allowing to pass an arbitrary Python object.
The semantic for this one is the same as the log level: if an
auto-discovered component comes from multiple inputs, it will receive
the obj of the last input that contributed to getting it instantiated
that has a non-None `obj`.
For convenience, some shorthands when creating a
TraceCollectionMessageIterator are equivalent to using an
AutoSourceComponentSpec. For example, these three forms are equivalent:
New test cases are added to test the new feature. They are based on
the corresponding CLI tests, in that they test equivalent scenarios
(same inputs, expecting the same results).
In order for TraceCollectionMessageIterator to find the test plugin
(both bt_plugin_test.py files), the only way I found was to amend
BABELTRACE_PLUGIN_PATH for the duration of the test.
Change-Id: I0e1100d850c920723861609cfa14707f8c669892 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1826 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Simon Marchi [Mon, 5 Aug 2019 15:05:08 +0000 (11:05 -0400)]
Move autodisc to its own convenience library
Since we'll want to make TraceCollectionMessageIterator have an
automatic source discovery feature, like the CLI has, move the auto
discovery code to its own library, which doesn't have any dependency on
the CLI.
To avoid dependencies on the CLI, here are the changes done to the auto
discovery code:
- To avoid depending on cli_query, use a simplified version,
simple_query, included in the auto discovery code. It doesn't handle
retrying if the status is TRY_AGAIN, as it is not very likely that a
babeltrace2.support-info query implementation will need to do that, as
we want them to be very fast (we can always revisit it in the future).
- Avoid using BT_CLI_LOGE_APPEND_CAUSE, replace with the new
BT_AUTODISC_LOGE_AND_APPEND.
- Don't use require_loaded_plugins, get_loaded_plugins_count and
borrow_loaded_plugin, as those are CLI-specific. Instead, the caller of
auto_discover_source_components must provide an array of plugins.
- plugin_restrict, used to filter down to a single plugin by name, is
removed. Instead, it is the caller's responsibility to pass a single
plugin to `plugins` if it wants to restrict the search to a single
plugin.
A side effect of moving the code is that a warning is now printed by
default when no input is found from non-option argument. I think this
is desirable to have this warning printed, so I have opted to update the
test case to expect it.
Change-Id: Id4009976430126d926eac46ff044365a6b2bc006 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1825 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Fri, 9 Aug 2019 21:17:35 +0000 (17:17 -0400)]
tests: make auto-source-discovery-{log_level,params} tests use sink.text.details
For the same rationale as the previous patch, change how the test
component classes for test_auto_source_discovery_params and
test_auto_source_discovery_log_level report what they receive from a
print to stdout to stream name.
Change-Id: I62405ccd1fbdbe708fbec5dacc52894d4060dfb0 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1864 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Simon Marchi [Fri, 9 Aug 2019 20:08:37 +0000 (16:08 -0400)]
tests: make auto-source-discovery-grouping test use sink.text.details
This patch changes the way that test component classes for the auto
source discovery grouping CLI test reports the inputs passed to each
instantiated component.
Currently, they print to stdout the input names
in the components' __init__. That is then check by the test script
(test_auto_source_discovery_grouping). The
TraceCollectionMessageIterator class in the Python bindings will soon
have the capability to automatically discover source components.
However, it's not practical to use the same test method in Python.
To facilitate this, make the test component classes report the inputs
they receive by sending a stream beginning message, where the stream
name contains the input names. The eventual Python test will be able to
inspect the stream name to validate that the right inputs were passed to
each component.
Since the information is now passed in messages, the
test_auto_source_discovery_grouping test now uses the details sink.
Change-Id: Ie5c12e390f5b68fb50a6d18c53463860fb7a1c99 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1863 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Simon Marchi [Thu, 8 Aug 2019 17:58:47 +0000 (13:58 -0400)]
tests: print executed CLI command line in bt_cli
When debugging a test, it is often useful to have the complete
CLI command line executed, so we can copy paste it and reproduce
the failure by hand.
This patch makes bt_cli echo the executed CLI command line to
stderr, such that it will show up when running one test by hand and in
the test logs when running "make check".
This allows simplifying bt_diff: we no longer have to pass the
CLI arguments to have them included in the error message, since
the full CLI command line will typically be printed just above.
Change-Id: I64f8c80f648d06319efe4e7544447183ba8069cc Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1839 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Thu, 8 Aug 2019 15:09:11 +0000 (11:09 -0400)]
tests: add bt_cli util function
Add a function that simply runs the CLI, redirecting stdout and
stderr to the specified files. This will allow more flexible tests than
the existing "diff" function.
Change-Id: I8319c4a74a0baa3b40afb871bfb0605ac81c3270 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1837 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Fri, 9 Aug 2019 20:30:09 +0000 (16:30 -0400)]
Fix: muxer: check for MUXER_MSG_ITER_CLOCK_CLASS_EXPECTATION_NONE in validate_new_stream_clock_class
There is a problem currently when trying to send multiple streams that
don't have a clock class (which is a supported use case) to a muxer
instance.
The first time we enter validate_new_stream_clock_class without a clock
class, we set the expectation to
MUXER_MSG_ITER_CLOCK_CLASS_EXPECTATION_NONE. The second time we enter
without a clock class, we wrongfully go to the error case. We are
missing a check for MUXER_MSG_ITER_CLOCK_CLASS_EXPECTATION_NONE, which
is the expected and valid case. This patch adds it.
Also, the error message is erroneous. If we are in that situation, we
expect a stream class _without_ a clock class, but are met with a stream
class _with_ a clock class. Fix that at the same time.
Change-Id: Iea692578498ea87a33ecba5d2fd54b2980056cd0 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1862
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Philippe Proulx [Fri, 9 Aug 2019 18:57:00 +0000 (14:57 -0400)]
Fix: translate-trace-ir-to-ctf-ir.c: dereference after `NULL` check
We only need to call must_protect_variant_option_name() when the tag
field class is not created by the component (`fc->tag_is_before`).
Furthermore, the local `must_protect` variable is not used, so the
corresponding must_protect_variant_option_name() parameter is not
either. They are both removed. Because the
must_protect_variant_option_name() function actually sets the name of
the option to use (protecting it or not), it's renamed to
maybe_protect_variant_option_name().
Found by Coverity Scan.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Ieccc0bcddc5570f473d4303f7cc470a56af65a9e
Fix: src.ctf.lttng-live: omitting stream end on stream hang up
Issue
=====
When receiving a LTTNG_VIEWER_INDEX_HUP request status, a
`src.ctf.lttng-live` component quickly removes the associated stream
iterator and move on to the next request without sending the stream end
message to properly close the stream. This breaks the guarantee that
stream end messages are always present.
Solution
========
Whenever we receive an HUP status from the server, we trigger the
`bt_msg_iter` code to hit an BT_MSG_ITER_MEDIUM_STATUS_EOF so it
generates the stream end message that closes the stream.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Change-Id: I89924ee95b35ca62c103651ad44c9eaf71f6df53
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1840 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Philippe Proulx [Fri, 9 Aug 2019 13:44:51 +0000 (09:44 -0400)]
Fix: muxer.c: compare_events(): initialize `ret` to 0
GCC complains when this `ret` is not initialized:
muxer.c: In function ‘compare_events’:
muxer.c:993:9: error: ‘ret’ may be used uninitialized in this
function [-Werror=maybe-uninitialized]
return ret;
It reaches `return ret` with an uninitialized `ret` when all the tests
fail to find a difference (which can happen).
src.ctf.lttng-live fix: remove overly strict check for session name
The live component checks that the path of a stream contains a
session's name. While this check works when interacting with pre-2.11
peers, there is no reason for a session's name to necessarily be in a
stream's path.
The LTTng relay daemon, as of 2.11, now expresses its path relative to
a session's output path, the session's name no longer appears in a
stream's path.
There are no comments explaining the reason for this check and it is
not present in babeltrace 1.5.x. Hence, it appears safe to remove it.
flt.utils.muxer: deterministically order messages of same timestamps
Issue
=====
The ordering of messages with the exact same timestamp is runtime
dependent. It is causing flakiness in tests using a `sink.text.details`
sink component where such differences in ordering is checked.
Note: It's common to have messages with the same timestamps because that
messages with no clock snapshot are assumed to happen at the same time
that of the previous message.
Solution
========
We need to order messages deterministically even when timestamps are the
same. This commit adds functions to order, in a predictable manner,
messages that have the same timestamp (or no timestamp at all) so that
they are sorted in the same way every time a trace is read.
With this commit:
* We first try to sort the messages by their trace name, stream class
id, or stream id,
* If not possible, we check if the messages have different types and if
they do, we sort them using the following scheme:
SB < PB < EV < DE < MI < PE < DP < SE
* If they are of the same type, we sort them comparing their properties.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Change-Id: Id6fc1fe8b19d02ec1804e54255a73c64820d1e13
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1719 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Fix: extra '%' causes babeltrace2 to abort() in logging code
An extra '%' in a TRACE logging statement causes
bt_common_custom_vsnprintf() to interpret '!' as a conversion
specifier. This causes handle_conversion_specifier_bt() to invoke
abort().
Simon Marchi [Wed, 7 Aug 2019 03:48:44 +0000 (23:48 -0400)]
Fix: bt2: use unsigned iteration variable
This fixes the following compiler diagnostic (gcc 9.1.0):
bt2/native_bt.c: In function ‘_wrap_field_class_enumeration_unsigned_get_mapping_labels_for_value’:
bt2/native_bt.c:9387:25: error: comparison of integer expressions of different signedness: ‘int’ and ‘uint64_t’ {aka ‘long unsigned int’} [-Werror=sign-compare]
9387 | for (int i = 0; i < *arg4; i++) {
| ^
Change-Id: I543c61d64c0cf7b42cb242bdbb7b71cbf67dbe48 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1834 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Philippe Proulx [Tue, 6 Aug 2019 16:49:54 +0000 (12:49 -0400)]
Fix: `ctf` plugin: msg-iter.c: do not switch packet twice
Issue
=====
In read_packet_header_begin_state(), bt_msg_iter_switch_packet() is
called systematically at the beginning of the state. However,
buf_ensure_available_bits() can fail below (the medium can return
`BT_MSG_ITER_MEDIUM_STATUS_AGAIN`), and read_packet_header_begin_state()
will be called again next time, calling bt_msg_iter_switch_packet()
twice for the same packet.
One of the problems of calling bt_msg_iter_switch_packet() twice is
that, the first time, the current snapshots are copied to the previous
snapshots and the current snapshots are reset; the second time, the
current (reset) snapshots are copied to the previous snapshots. This
shatters the whole snapshot states. Because the medium installed by
`src.ctf.lttng-live` can return `BT_MSG_ITER_MEDIUM_STATUS_AGAIN`, all
the messages that depend on the previous snapshots, like the discarded
events/packets messages, are wrong.
Solution
========
In `msg-iter.c`, add a new `STATE_SWITCH_PACKET` state which is
dedicated to doing what bt_msg_iter_switch_packet() used to do. Once it
completes, the next state is `STATE_DSCOPE_TRACE_PACKET_HEADER_BEGIN`.
read_packet_header_begin_state() does not switch packets itself now (the
`STATE_SWITCH_PACKET` state already occured at this point), so it can
safely occur twice.
Known drawbacks
===============
There's one more state to handle per packet, but this have a significant
performance hit.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com> Reported-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Change-Id: I24dd2482762c61e339ab0310fddee6c9aa69d438
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1833 Tested-by: jenkins <jenkins@lttng.org>
Philippe Proulx [Tue, 6 Aug 2019 16:47:26 +0000 (12:47 -0400)]
src.ctf.lttng-live: `live_viewer_connection`: make `obj` first member
This fixes a leak. The `bt_object` member must always be the first
because the internal API casts any `void *` parameter to `bt_object *`.
The object API was not getting the correct reference count here.
Philippe Proulx [Tue, 6 Aug 2019 00:31:35 +0000 (20:31 -0400)]
lib, bt2: make query method receive custom data
Just like a component class initialization method can receive custom
user data (`init_method_data`) along with generic parameters, this patch
makes this possible for a query operation.
Library changes
===============
On the library side, this patch introduces the new
bt_query_executor_create_with_method_data() function which is like
bt_query_executor_create(), but accepts a `void *` user data parameter.
Internally, bt_query_executor_create() calls
bt_query_executor_create_with_method_data() with `NULL` as the
`method_data` parameter.
The component class query method type is changed to:
The component class is guaranteed, for a given query executor, to always
being passed the same `method_data` value to its query method.
Python bindings changes
=======================
QueryExecutor.__init__() accepts a new, optional `method_obj` parameter
which defaults to `None`.
`None` gets translated to `NULL` as the `method_data` parameter, and
vice versa when a Python query method is called from the native part.
QueryExecutor.__init__() validates that you can only pass a Python
method object when the component class you query is a Python component
class.
Because you build the query executor and then make a query operation in
two steps, a Python reference to the method object must exist as long as
the query executor exists. This patch makes it reside within the Python
query executor wrapper (`_method_obj` attribute) as there's no way
currently to lose the Python wrapper and retrieve it through a library
call. If this becomes possible in the future, then we can introduce
query executor destruction listeners and add one from the Python
bindings to put the `PyObject *` reference when the query executor is
destroyed. A new test checks this
(test_query_with_method_obj_del_ref()), deleting the test's reference
between the query executor's construction and the actual query
operation. I tried without making the query executor keep a strong
reference and the test causes a segmentation fault, as expected, which
shows that the test is valuable.
Philippe Proulx [Mon, 5 Aug 2019 22:01:09 +0000 (18:01 -0400)]
Add bt_graph_add_simple_sink_component() tests
As of this patch, the tests are basic and only check that the statuses
returned by the user functions are correctly translated and returned by
bt_graph_run_once(). The user initialization function status is not
returned by bt_graph_add_simple_sink_component() because the
initialization function is called within the sink component's "graph is
configured" method.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Id8ccfd7df4e4a8f0a39af386ed4ad13fe657f86a
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1830 Reviewed-by: Simon Marchi <simon.marchi@efficios.com> Reviewed-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Philippe Proulx [Mon, 5 Aug 2019 21:57:27 +0000 (17:57 -0400)]
Fix: src/lib/graph/graph.h: do not override `comp_status` value
This made, for example, bt_graph_run() return
`BT_GRAPH_RUN_STATUS_ERROR` when the sink component's "graph is
configured" method returned
`BT_COMPONENT_CLASS_SINK_GRAPH_IS_CONFIGURED_METHOD_STATUS_MEMORY_ERROR`,
therefore not percolating the correct status.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Ic89d9315f8bc1c3226a14955c9e6419462d5e588
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1829 Reviewed-by: Simon Marchi <simon.marchi@efficios.com> Reviewed-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Philippe Proulx [Mon, 5 Aug 2019 20:15:32 +0000 (16:15 -0400)]
lib: add bt_graph_add_simple_sink_component()
This patch adds bt_graph_add_simple_sink_component(), an easy way to add
a simple sink component which has a single message iterator iterating a
single input port named `in` and which calls user-provided functions at
specific locations.
This makes it possible to create a sink component without having to:
* Create a sink component class with initialization, "graph is
configured", "consume", and finalization methods.
* Create an input port in the component class's initialization method.
* Create an input port message iterator in the component class's "graph
is configured" method.
* Put the input port message iterator's reference in the component
class's finalization method.
The goal of this new function is to make it easy to get messages at the
sink endpoint of a graph, just like we used to do with the output port
message iterator concept (but the graph model is honored now). The user
must still call bt_graph_run() or bt_graph_run_once() to make her
consume function (see details below) called: the added simple sink
component is not special in any way.
bt_graph_add_simple_sink_component() receives three function pointers
(and custom user data):
Initialize (optional):
Called after the simple sink component creates the input port
message iterator in the "graph is configured" method.
The user function receives the message iterator to perform any
initial task.
Consume:
Called for each "consume" method call of the simple sink component.
The user function receives the message iterator and can get the next
messages with bt_self_component_port_input_message_iterator_next()
as usual.
Finalize (optional):
Called when the simple sink component is finalized.
The message iterator is not available at this point.
I'm not wrapping this one in Python because it's so easy to replicate
with our bindings:
class _SimpleSink:
def __init__(self, params, consume_cb):
self._consume_cb = consume_cb
self._add_input_port('in')
Philippe Proulx [Sun, 4 Aug 2019 04:24:06 +0000 (00:24 -0400)]
cli, tests: rename "leftover" terms to "non-option"
"Leftover" is a popt term, and we don't use popt anymore. Arguments
which are not options are not leftover anymore: they are treated within
the same loop as option arguments, and their position relative to option
arguments are important, so we simply call them "non-option arguments".
Note that "non-option" is also a term that getopt(1) and getopt(3) use.
Philippe Proulx [Sun, 4 Aug 2019 14:07:18 +0000 (10:07 -0400)]
test_try_again_many_times(): use three times `None`'s ref count
100,000 iterations was arbitrary anyway, so let's just depend on the
initial reference count of `None` and iterate three times that since
this is the object we want to avoid destroy.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I3d587c730b4587d5381d07541d0d86ccff344858
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1822 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
Philippe Proulx [Sun, 4 Aug 2019 14:04:03 +0000 (10:04 -0400)]
lib: remove output port message iterator
The output port message iterator concept was not a good idea in the
beginning: it breaks the graph model, making it possible to iterate a
disconnected output port (which is weird from the component's
perspective), blocking the graph (user cannot call bt_graph_run()), and
forcing the graph not to contain any sink component. This looks and
feels like a hack.
For the C part, a subsequent patch should implement an easy way to add a
simple sink component to a graph based on a simple callback and custom
user data instead of going through the inconvenience of creating a sink
component class, setting the methods manually, creating an input port
message iterator once the graph is configured, etc.
For the Python part, we'll focus on `TraceCollectionMessageIterator` to
replace `_OutputPortMessageIterator`. `TraceCollectionMessageIterator`
should cover most of the use cases and is easier to use: you don't need
to set up your graph, add your own `flt.utils.muxer`, etc.
For more advanced use cases in Python, it's always possible to create a
"proxy sink component", just like `TraceCollectionMessageIterator` does
internally, to get full control on the input port message iterator.
To adapt the current tests, `TestOutputPortMessageIterator` in
`utils.py` can be used like an output port message iterator. Such an
iterator cannot seek however, so `test_message_iterator.py` needed
special treatments to make the eventual input port message iterator
seek.
Philippe Proulx [Sun, 4 Aug 2019 05:13:15 +0000 (01:13 -0400)]
bt2: make `TraceCollectionMessageIterator` not use an output port msg iter
Because there's a plan to drop the output port message iterator concept
altogether, make `TraceCollectionMessageIterator` not use any output
port message iterator.
Instead, a `TraceCollectionMessageIterator` instance adds a proxy sink
component (`_TraceCollectionMessageIteratorProxySink`) to its graph,
connecting it to the last filter component in the chain, and sharing
with it a list having a single item.
When the sink consumes a message from its upstream message iterator, it
places it in the shared list as the first item.
TraceCollectionMessageIterator.__next__() is changed so that it calls
Graph.run_once() to make the proxy sink consume, and then reads the
consumed message from the shared list.
Philippe Proulx [Sun, 4 Aug 2019 05:11:03 +0000 (01:11 -0400)]
bt2: wrap bt_graph_run_once() (Graph.run_once())
As opposed to Graph.run() which simply returns `None` when it's finished
running, Graph.run_once() needs a way to indicate that the whole graph
is finished. In that case, the method raises `bt2.Stop`. Otherwise,
Graph.run_once() returns `None`.
While bt_graph_run() makes the graph run until the end,
bt_graph_run_once() makes it run a single time, that is, a single sink's
(next in the queue) "consume" operation.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I65dc06c93425d5b5aa1f9e31edf3e22bbeaf77f0
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1818 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
Philippe Proulx [Sat, 3 Aug 2019 18:36:42 +0000 (14:36 -0400)]
bt2: pass custom Python object to Python component's __init__()
Just like you can pass custom data (`void *`) to the initialization
function of a component class written in C with the
bt_graph_add_*_component_with_init_method_data() functions, this patch
makes it possible to pass any Python object to the __init__() method of
a component class written in Python with the Graph.add_component()
method.
This patch installs a mechanism to share Python data between a Python
graph user and the methods of a Python component without relying on
`nonlocal`, global variables, or other such hacks. This data can be as
simple as an integer and as complex as a database connection, for
example.
The __init__() method of a Python component used to look like:
def __init__(self, params):
...
It's now:
def __init__(self, params, obj):
...
When you pass any Python object to Graph.add_component() as its `obj`
parameter, the Python component's __init__() method eventually receives
it.
Graph.add_component() ensures that the component class to instantiate is
a Python component class if `obj` is not `None`.
Internally, `None` gets converted to `NULL` as the `init_method_data`
parameter of the bt_graph_add_*_component_with_init_method_data()
functions, and then `NULL` gets converted back to `Py_None` in
component_class_init().
Now there is the risk that you call
bt_graph_add_*_component_with_init_method_data() in C with a Python
component class and pass a non-`NULL`, non-`PyObject *`
`init_method_data` parameter. However I consider an insignificant
drawback as you're not supposed to use the `init_method_data` with a
component class you don't know.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Ib879ece9e423b3495b9449ca73674082020865c5
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1815 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
Philippe Proulx [Sat, 3 Aug 2019 18:29:27 +0000 (14:29 -0400)]
bt2: remove BT CC entry from global HT in _UserComponentType.__del__()
The global `bt_cc_ptr_to_py_cls` hash table maps BT component class
pointers to Python component classes (`PyObject *`) for component
classes created in Python. The key and value are weak references.
When the Python side calls native_bt.bt2_component_class_*_create() with
a Python component class, an entry is added to `bt_cc_ptr_to_py_cls` on
success. This works most of the time because all Python classes are
normally destroyed when the interpreter is finalized, but there could be
issues with specific/unusual import patterns.
To avoid potential issues, remove an entry from `bt_cc_ptr_to_py_cls` in
_UserComponentType.__del__(), where a strong BT component class
reference is finally released.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I450a1c6f179f352e758b9e3e5cac8f4711aa3c88
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1816 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
... all source components discovered from `my-traces-1` will have log
level TRACE and all source components discovered from `my-traces-2` will
have log level DEBUG.
It becomes less obvious when components are given inputs coming from
multiple leftovers (because of the auto-discovery grouping feature):
which log level do they receive? For example, if the following line:
leads to these components getting instantiated, with these inputs:
* Source component X with inputs `my-traces-1/x` and `my-traces-2/x`.
* Source component Y with input `my-traces-1/y`
In this case, each component receives (on the `run` command line) the
log level options of all leftovers that contributed to its inputs, in
the same order as they are provided on the command line. The resulting
`run` command line for the example above could therefore look like:
... all source components discovered from `my-traces-1` will receive
`foo=2` and all source components discovered from `my-traces-2` will
receive `bar=3`.
It becomes less obvious when components are given inputs coming from
multiple leftovers (because of the auto-discovery grouping feature):
which parameters do they receive? For example, if the following line:
leads to these components getting instantiated, with these inputs:
* Source component X with inputs `my-traces-1/x` and `my-traces-2/x`.
* Source component Y with input `my-traces-1/y`
In this case, each component receives the parameters of all leftovers
that contributed to its inputs, in the same order as they are provided
on the command line. The resulting `run` command line for the example
above could therefore look like:
resulting in these parameters being passed to the components:
* Source component X receives parameters `foo=4,bar=3`
* Source component Y receives parameters `foo=2,bar=3`
Implementation details
----------------------
The auto discovery mechanism now returns, for each result, which
input from the passed `inputs` array contributed to that result.
This allows us, for the component that we create from a given result, to
get the parameters from the leftovers that have contributed to it.
Change-Id: Ic048e4e137c2e1f93b6da13a62629343500cb75a Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1809 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Philippe Proulx [Fri, 2 Aug 2019 17:43:59 +0000 (13:43 -0400)]
bt2: move common internal functions to their own files
This patch moves C code from `native_bt_component_class.i` to the new
`native_bt_bt2_objects.h` and `native_bt_log_and_append_error.h` files,
namely:
* bt_bt2_init_from_bt2(), which was bt_bt2_cc_init_from_bt2()
* bt_bt2_exit_handler(), which was bt_bt2_cc_exit_handler()
* The common logging and error appending functions.
This is just cleaner as those functions are not strictly related to the
component class interface.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Ic9d7fed993b94b45f624e28e4c512b5ccab776bd
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1814 Reviewed-by: Simon Marchi <simon.marchi@efficios.com> Tested-by: jenkins <jenkins@lttng.org>
Philippe Proulx [Fri, 2 Aug 2019 17:25:42 +0000 (13:25 -0400)]
bt2: put SWIG interface file C code in separate files
This patch moves any code in `%{ %}` blocks within SWIG interface files
(`.i`) to their own files. For a SWIG interface file named `file.i`, the
code is moved to `file.i.h`. In `file.i`, there is:
%{
#include "file.i.h"
%}
This helps with text editors which do not always know how to highlight
the code of a `.i` file.
In `Makefile.am`, the `.i.h` files are added to the
`SWIG_INTERFACE_FILES` variable as they can be considered part of the
SWIG interface files.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I28975c9b3c827e65ba77af9134521ce1b920d740
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1812 Reviewed-by: Simon Marchi <simon.marchi@efficios.com> Tested-by: jenkins <jenkins@lttng.org>
Simon Marchi [Thu, 1 Aug 2019 14:57:22 +0000 (10:57 -0400)]
tests: reorganize CLI's `convert` tests
Create a `convert` directory under tests/cli, so we can have
multiple separate tests targeting specific features of the convert
command.
Move test_convert_args in there, this is the file containing the things
that are easy to test, comparing the output of --run-args with an
expected output.
Move the auto source discovery test in there too, as it's a feature of
the convert command. Rename it to "auto source discovery grouping",
since that is mostly what it exercises, and we'll want to add tests for
other areas of auto source discovery.
Change-Id: I20e5085d6b41eec322b777f45be1fee87894fe91 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1808 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Simon Marchi [Mon, 29 Jul 2019 18:42:23 +0000 (14:42 -0400)]
cli: handle leftovers in the same loop as components
[this patch is better viewed with `git show -w`]
Handle leftovers (put them in the leftovers array) at the same time as
we process --component options and things that apply to components
(--params and --log-level). When we handle a leftover, assign the
current_item_type variable, to override the previous current item.
This change makes it so that we prevent processing --params or
--log-level after a leftover. Previously, this command line:
Would apply the `a=2` param to the `src.bon.jour` instance, since it was
the last component declared. This is confusing, however, because of the
some-leftover leftover in between.
With this patch, the `some-leftover` leftover becomes the "currently"
processed item when we reach it. And since it's not possible (for the
moment) to apply --params or --log-level to a leftover, the command line
shown above now result in the error:
No current component of which to set parameters:
a=2
I improved test_convert_args to be able to check the error message of a
bt2 run we expect to fail. This lets us avoid cases where babeltrace
fails for an unexpected reason, but the test still passes.
Change-Id: I0e065c1cd5f32f59292c9a40c6a8077a52d35237 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1806 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Simon Marchi [Tue, 30 Jul 2019 04:34:58 +0000 (00:34 -0400)]
cli: remove `run` and `convert` commands' --name option
There are currently two ways of setting a component's name:
- Using the --name option
- Using the NAME: prefix to the --component option
Having two ways of doing the same thing is not really useful here.
Supporting the --name option complicates the code a little bit: we have
to track whether the current component's name has been set, so that when
we start a new --component, we can generate a name if needed.
This patch removes the --name option in favor of always using the NAME:
prefix of --component. This way, as soon as we handle --component, we
know if the user has provided a name or not, and if not, we can generate
one immediatly (applies only in the convert command; in the run command, the
user must provide a name).
Change-Id: Id278e9d110e4ea415f66538ac55e9459e49b1d0e Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1805 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Tue, 30 Jul 2019 01:29:51 +0000 (21:29 -0400)]
cli: remove `convert` command's --path and --url options
It was decided that these two options can be removed. They are
shortcuts for things that can be set using --params.
The --path option, in particular, is not useful as it used to be: the
de-facto standard parameter that sources are going to accept for their
input data is no longer `path` (it is now `inputs`).
Change-Id: I023e9cdaf319d7a9d2951315ce595ce099648391 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1804 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Sat, 27 Jul 2019 05:10:16 +0000 (01:10 -0400)]
Remove popt from project's dependencies
All of the argument parsing having been moved to use the internal argpar
library, we can now stop linking with popt. And since it was the last
use of popt in the repo, remove anything related to it.
Change-Id: Id2aee4869780e586833d6b6646e12e19c6d78f6a Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1795 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Mon, 29 Jul 2019 15:42:58 +0000 (11:42 -0400)]
cli: remove babeltrace2-log
babeltrace-log (now renamed to babeltrace2-log) is a legacy tool
provided by Babeltrace 1 used to transform `dmesg` output to CTF. The
version currently provided by Babeltrace 2 simply runs babeltrace2 with
a pre-defined graph topology. It does nothing that a user couldn't do
on the command line.
Simon Marchi [Sat, 27 Jul 2019 03:37:22 +0000 (23:37 -0400)]
cli: use argpar for parsing help command's arguments
All straightforward, except that there is a functional change as part
of this patch. It was possible to pass multiple leftovers, and only the
first one was considered.
babeltrace2 help ctf tintin
did the same as
babeltrace2 help ctf
It now prints a beautiful error:
./src/cli/babeltrace2 help ctf tintin
ERROR: [Babeltrace CLI] (/home/smarchi/src/babeltrace/src/cli/babeltrace2.c:2564)
Command-line error: retcode=1
CAUSED BY [Babeltrace CLI] (/home/smarchi/src/babeltrace/src/cli/babeltrace2-cfg-cli-args.c:1461)
Extraneous command-line argument specified to `help` command: `tintin`.
Change-Id: I2c8f8610004d60a0844d18c9d622037cc7c1c04c Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1792 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Fri, 26 Jul 2019 21:20:08 +0000 (17:20 -0400)]
cli: use argpar for parsing run command's arguments
A non-obvious change is the handling of the --retry-duration value.
Before this patch, it was automatically converted to a long by popt. We
now call g_ascii_strtoll and validate that it's parsable as an integer.
Also, introduce the help_option_is_specified function (which will be
used for other sub-commands as well). The goal is to check for --help
before everything else, such that doing:
$ babeltrace2 run hello --help
prints the help, rather than saying "Unexpected parameter `hello`".
Change-Id: Ib6873a2e2183e73f340cf25026924df298c8c1e9 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1790 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Wed, 24 Jul 2019 15:56:43 +0000 (11:56 -0400)]
cli: use argpar for top-level args
This patch replaces the hand-made parsing of the top-level arguments
with something based on the internal argpar library.
No functional changes expected from the point of view of the user.
One internal difference is that previously, the argc/argv passed to
sub-commands included an argv[0] that was ignored by popt. In the
subsequent patches, we replace popt with argpar, and argpar doesn't want
an argv[0] containing the name of the program. So instead of passing an
unnecessary argv[0], I have made it such that the argc/argv the
top-level passes to subcommands only represent the actual arguments. So
for example, with:
babeltrace2 convert -c src.ctf.fs
the bt_config_convert_from_args function used to receive this as argv:
["convert", "-c", "src.ctf.fs"]
With
babeltrace2 --debug -c src.ctf.fs
it used to receive
["--debug", "-c", "src.ctf.fs"]
With this patch, it will only receive `["-c", "src.ctf.fs"]`. So
functions bt_config_*_from_args are updated to cope with this
change. In particular, I passed the POPT_CONTEXT_KEEP_FIRST flag to
poptGetContext. But in practice, they will all disappear anyway, in
favor of argpar.
Change-Id: I9f1210f1c338c7eb39e228e20218c83e46961ee4 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1789 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Philippe Proulx [Sun, 7 Jul 2019 03:15:58 +0000 (23:15 -0400)]
Add internal command-line argument parser API
This patch adds an internal command-line argument parser API.
The exact API is well documented in `src/argpar/argpar.h`.
The features that I'm looking for are:
* C API.
* Support for ordered arguments in the results, even between options and
non-option arguments (sometimes called positional arguments).
* Support for GNU-style arguments, including "glued" short options and
long options with `=` to set the option's argument.
* Portable.
* Compatible license.
* No global variables (we're trying to avoid that programming style
throughout the project).
* Easy to use.
None of the popular libraries I looked at, including popt of course, met
all those specifications.
The goal of this is:
1. To simplify the parsing of general options (before the command name)
in bt_config_cli_args_create(). It is currently hard-coded.
This is not possible with popt because it would fail with an unknown
option, and you would not know the position of the command name's
argument within the array.
This is possible with g_option_context_parse().
2. To make it possible, in the `convert` command, to assign parameters
and other position-specific options to a non-option argument, for
example:
This is not possible with popt as it collects all the non-option
arguments as an array of leftover arguments.
This is also not possible with g_option_context_parse() for the same
reasons as with popt.
getopt_long() could satisfy both 1. and 2., but it's somewhat a pain to
use and to maintain the available options, as you need to specify the
long options in a table and the equivalent short options as a single
string with a special encoding (the `convert` command's option string
would be particularily ugly). Also: getopt_long() plays a lot with
global variables; it's not thread-safe: the parser's state is global.
Also: the upstream getopt_long()'s (glibc) license is LGPL, which is
more restrictive than our MIT license, so I think we want to avoid that.
I believe having our own (tested) CLI argument parser is beneficial,
especially in the long term: we can drop a direct dependency (and popt
is getting old), not introduce a new one, it's about 450 lines of
documented C code, and if we ever need something very specific in the
argument parsing strategy in the future, we can add it directly. As a
reference, FFmpeg, a project which has complex argument parsing, similar
to Babeltrace's `convert` and `run` commands, has its own argument
parser (see parse_options() in [1]).
Only the CLI will use the bt_argpar_*() API, but to make the unit tests
simpler, I put it in its own convenience library.
Philippe Proulx [Mon, 29 Jul 2019 19:32:08 +0000 (15:32 -0400)]
configure.ac: refine Python 3 interpreter and dev. libraries detection
This patch refines how the `configure` script detects the Python 3
interpreter and development libraries.
The goal is to clean up this part of `configure.ac` and to make the
Python 3 interpreter available as a general tool for testing, even
without building the Python bindings or the Python plugin provider.
Before this patch, `configure` ensures that the Python interpreter and
the development libraries exist and are valid if you use the
`--enable-python-bindings` or `--enable-python-plugins` option.
With this patch, `configure` always tries to find the Python
interpreter. If it finds it, it sets the local `have_python` shell
variable and defines the `HAVE_PYTHON` Automake definition.
If `configure` finds the Python interpreter, it always tries to find
`python-config`. It if finds it, it sets the local `have_python_dev`
shell variable and defines the `HAVE_PYTHON_DEV` Automake definition.
You can still override the include directories and linker flags with the
`PYTHON_INCLUDE` and `PYTHON_LDFLAGS` environment variables.
Now, if you specify the `--enable-python-bindings` or
`--enable-python-plugins` option, `$have_python_dev` must be `yes`.
What used to be `HAVE_PYTHON` (Automake definition) is renamed to
`ENABLE_PYTHON_COMMON_DEPS`: this is defined if we need to build any
common Python dependency. As of this patch, the only one is the internal
`py-common` convenience library.
In the final `configure` report, I added:
* Whether or not the Python interpreter was found and its path.
* Whether or not the `python-config` program (called "development
libraries") was found and its path.
This patch was only tested manually, locally removing the Python
interpreter, the `python-config` program, and passing the
`--enable-python-bindings` or `--enable-python-plugins` option to test
the different `configure` paths.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Id0ac60ebc97a9ce2951aee713eccdc24c19d3791
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1799 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Michael Jeanson <mjeanson@efficios.com>
The only user of live_viewer_connection_list_sessions() is
lttng_live_query_list_sessions() which calls
live_viewer_connection_create() before. live_viewer_connection_create()
calls lttng_live_connect_viewer() which calls lttng_live_handshake().
There is no need to call lttng_live_handshake() twice. This makes the
relay daemon assign two viewer IDs to the CLI, as seen in Wireshark.
Babeltrace 1.5 does not send the "connect" command twice either.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I640b9397c21ca37ad5c841415b5ed0c12193825a
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1798
Philippe Proulx [Mon, 29 Jul 2019 14:06:58 +0000 (10:06 -0400)]
Fix: live_viewer_connection_create(): `lttng_live_msg_iter` can be `NULL`
Because `lttng_live_msg_iter` can be `NULL` in
live_viewer_connection_create() when it's used from a query operation,
pass the log level to use directly and only read `self_comp` if
`lttng_live_msg_iter` exists.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Ied5ec621bcd066df0bb312a6f81899616f9d1a9f
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1797
Jonathan Rajotte [Tue, 23 Jul 2019 21:58:29 +0000 (17:58 -0400)]
`ctf` plugin: Use CTF_BYTE_ORDER_UNKNOWN in place of -1
Seen on clang-3.9, 4.0 and clang-1001.0.46.4.
Comparison of constant -1 with expression of type 'enum *****'
is always false [-Wtautological-constant-out-of-range-compare]
Note that the enum underlying type is implementation defined and left to
the choice of the compiler by the standard [1] (6.7.2.2 4). Most
compiler default to unsigned int. The use of -1 is not a problem per see
since wrap around of unsigned int behaviour is not undefined. Using -1
is the equivalent of assigning UINT_MAX here. This warning was removed
for later clang for these specific cases since the effect of always being
false is erroneous.
Still, it make more sense to use a specific enum for such case instead
of relying on the compiler type.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Change-Id: I5dcbaa2066cbab8c429ff54e8d8b8b295fc8baf1
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1764
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Jonathan Rajotte [Tue, 23 Jul 2019 19:36:30 +0000 (15:36 -0400)]
`ctf` plugin: Use CTF_SCOPE_PACKET_UNKNOWN in place of -1
Seen on clang-3.9, 4.0 and clang-1001.0.46.4.
Comparison of constant -1 with expression of type 'enum *****'
is always false [-Wtautological-constant-out-of-range-compare]
Note that the enum underlying type is implementation defined and left to
the choice of the compiler by the standard [1] (6.7.2.2 4). Most
compiler default to unsigned int. The use of -1 is not a problem per see
since wrap around of unsigned int behaviour is not undefined. Using -1
is the equivalent of assigning UINT_MAX here. This warning was removed
for later clang for these specific cases since the effect of always being
false is erroneous.
Still, it make more sense to use a specific enum for such case instead
of relying on the compiler type.
get_root_scope_from_absolute_pathstr depends on the fact that
CTF_SCOPE_PACKET_HEADER is 0 to work.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Change-Id: Ie13f6c4dbab5038eb7a7fea6d58ce4451bc551d1
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1762
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Jonathan Rajotte [Wed, 24 Jul 2019 19:47:09 +0000 (15:47 -0400)]
lib: Explicit cast to bt_plugin_find_status
Clang 8.0:
error: implicit conversion from enumeration type 'bt_plugin_find_all_status' (aka 'enum bt_plugin_find_all_status') to different enumeration type 'enum bt_plugin_find_status' [-Werror,-Wenum-conversion
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Change-Id: Ie8304eccaa329b953837eb24e5fe4c2b2b81f320
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1765 Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org>
Jonathan Rajotte [Tue, 23 Jul 2019 19:55:12 +0000 (15:55 -0400)]
src.ctf.lttng-live: Use SESSION_NOT_FOUND_ACTION_UNKNOWN in place of -1
Seen on clang-3.9, 4.0 and clang-1001.0.46.4.
Comparison of constant -1 with expression of type 'enum *****'
is always false [-Wtautological-constant-out-of-range-compare]
Note that the enum underlying type is implementation defined and left to
the choice of the compiler by the standard [1] (6.7.2.2 4). Most
compiler default to unsigned int. The use of -1 is not a problem per see
since wrap around of unsigned int behaviour is not undefined. Using -1
is the equivalent of assigning UINT_MAX here. This warning was removed
for later clang for these specific cases since the effect of always being
false is erroneous.
Still, it make more sense to use a specific enum for such case instead
of relying on the compiler type.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Change-Id: I447d04333ae104ea953437b095053f341f87bbb7
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1763
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Jonathan Rajotte [Tue, 23 Jul 2019 19:23:02 +0000 (15:23 -0400)]
`ctf` plugin: Use is_log_level_set to infer ctf_event_class log_level validity
Seen on clang-3.9, 4.0 and clang-1001.0.46.4.
Comparison of constant -1 with expression of type 'enum *****'
is always false [-Wtautological-constant-out-of-range-compare]
Note that the enum underlying type is implementation defined and left to
the choice of the compiler by the standard [1] (6.7.2.2 4). Most
compiler default to unsigned int. The use of -1 is not a problem per see
since wrap around of unsigned int behaviour is not undefined. Using -1
is the equivalent of assigning UINT_MAX here. This warning was removed
for later clang for these specific cases since the effect of always
being false is erroneous.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Change-Id: I6ce97c2481f36573d1a221b59e2459f1fe35ded9
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1761
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Philippe Proulx [Thu, 25 Jul 2019 22:07:59 +0000 (18:07 -0400)]
lib: prepare the ground for stateful query operations
This patch changes the library (and everything which depends on it) so
that it is possible to make query operations stateful in the future if
needed.
Without this patch, a query operation can return the
`BT_COMPONENT_CLASS_QUERY_METHOD_STATUS_AGAIN` status. The query
executor's user can then retry the operation until it works, or abandon.
However, there's no way for this query function to keep state between
calls, so the process starts over every time whereas it could have
progressed otherwise (for example, a temporary unresponsive network).
Library changes
===============
This patch does two things to address the limitation above:
* It makes bt_query_executor_create() accept the component class, query
object, and query parameters. It also removes those parameters, as
well as the logging level, from bt_query_executor_query().
This ensures that a given query executor is dedicated to a specific
query operation. Once you create a query executor, you cannot change
the target component class, query object, and query parameters
afterwards.
The initial query executor's logging level is `BT_LOGGING_LEVEL_NONE`.
You can set it before you call bt_query_executor_query() with
bt_query_executor_set_logging_level().
Internally, the query executor keeps a reference on the component
class, a copy of the query object, and a reference on the query
parameters.
* It makes a query method receive a private query executor
(`bt_private_query_executor *`) object, and not receive the logging
level.
A private query executor is a private view of the query executor. This
is where we can eventually add bt_private_query_executor_set_data()
and bt_private_query_executor_get_data() to make the query operation
stateful. From the query method's point of view, the private query
executor is borrowed: there are no bt_private_query_executor_get_ref()
and bt_private_query_executor_put_ref() functions.
You can go from a private query executor to a constant query executor
with bt_private_query_executor_as_query_executor_const(). This is how
you can get the query executor's current logging level with
bt_query_executor_get_logging_level().
To keep the changes minimal, query methods still receive the object
and parameters every time. They are guaranteed, however, that for the
same query executor, they remain unchanged. This guarantee is not
relevant now anyway as there's no way to set and get private data.
Python bindings changes
=======================
* The new `_QueryExecutorCommon` class contains the common properties of
a query executor and a private query executor (`is_interrupted` and
`logging_level`). It requires its subclasses to implement the
_as_query_executor_ptr() method to get the query executor pointer.
You now call the query() method without parameters:
query_exec.query()
You can set the logging level with:
query_exec.logging_level = bt2.LoggingLevel.TRACE
* The new `_PrivateQueryExecutor` class inherits `_QueryExecutorCommon`.
An instance of `_PrivateQueryExecutor` is passed to the query method
during a query operation.
Once the user's query method returns, _bt_query_from_native()
invalidates the private query executor in case the user kept a
reference somewhere. Calling any method of an invalid private query
executor raises `RuntimeError`.
CLI changes
===========
To centralize how query operations are performed in the CLI, all the
sites which need to query use the new cli_query(), implemented in
`src/cli/babeltrace2-query.c`.
cli_query() is similar to the static query() function in
`babeltrace2.c`, which already existed, but it accepts the logging level
directly and returns `bt_query_executor_query_status`.
The static query() function still exists, but calls cli_query() now.
The automatic source discovery feature now uses cli_query() instead of
creating its own query executor.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I3c254325817c2c0091b6c4f0030bc7020443b8e0
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1787 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
Philippe Proulx [Fri, 26 Jul 2019 18:56:31 +0000 (14:56 -0400)]
bt2: honor self component or query log level when logging
This patch makes the Python bindings native code honor the self
component or query log level in log statements when possible.
The locations where this is not possible are:
* Registration of a BT component class pointer to a Python component
class in global hash table.
* Lookup of a BT component class in global hash table.
* Module's destructor.
* Creation of a BT component class from the `_UserComponentType`
metaclass.
* Trace and trace class destruction listeners.
* Graph's port added and ports connected listeners.
All those sites still use the module's
`bt_python_bindings_bt2_log_level` hidden symbol which is initialized
from the `BABELTRACE_PYTHON_BT2_LOG_LEVEL` environment variable at
construction time.
Move this file so that other parts of the project than plugins, namely
Python bindings, can use it.
Also: in `comp-logging.h`, the requirement that `BT_LOG_OUTPUT_LEVEL`
and `BT_COMP_LOG_SELF_COMP` are set is removed because they are not
needed to use the generic BT_COMP_LOG_CUR_LVL() macro.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: Ic961c3a97863dcef63fb7ee673841650cf9ca9e9
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1788 Reviewed-by: Simon Marchi <simon.marchi@efficios.com> Tested-by: jenkins <jenkins@lttng.org>
Philippe Proulx [Thu, 25 Jul 2019 16:47:00 +0000 (12:47 -0400)]
Add `babeltrace.` prefix to `trace-info` and `support-info` query objects
The `trace-info` and `support-info` query objects are supported by the
project's plugins, the project's Python bindings, and the project's CLI.
To avoid name clashes with third-party query objects, prepend
`babeltrace.` to the object names as a namespace.
The dot notation is not standardized in any way here: it is an arbitrary
format that the project chooses. We don't necessarily expect third-party
developers to follow this scheme.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Change-Id: I6aec9bf24f7a8a41796c3a1e84d982c99be6aee5
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1783 Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Simon Marchi <simon.marchi@efficios.com>
Simon Marchi [Wed, 24 Jul 2019 16:11:01 +0000 (12:11 -0400)]
lib: rename INVALID_OBJECT status to UNKNOWN_OBJECT
Rename INVALID_OBJECT to UNKNOWN_OBJECT, because it represents better
the situation: the passed object is not necessarily invalid (any string
is a valid object), it's just that the component class doesn't know that
object.
Change its numerical value to something positive, because it should not
be considered as an error (which negative numerical values represent).
A function returning UNKNOWN_OBJECT should not append an error cause.
Also, in the Python bindings, change the default implementation of
_query to raise bt2.UnknownObject instead of NotImplementedError.
NotImplementedError would result in an ERROR status code, whereas
bt2.UnknownObject becomes UNKNOWN_OBJECT. This matches the behavior of
a component class implemented in C, where if it doesn't provide a query
method, it will automatically return UNKNOWN_OBJECT.
Change-Id: Ica1418272b5bf2bbe8e96d639c16f56191151d79 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1760
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Simon Marchi [Wed, 24 Jul 2019 18:54:15 +0000 (14:54 -0400)]
bt2: move __version__ to version.py.in
The only bit of __init__.py that is set at configure time is the package
version. Move it to its own file, version.py.in, and make __init__.py
not generated. This will make life a bit easier when we need to modify
__init__.py.
Change-Id: Ibc9fb67a897b226c46b4fddb3af7c2219d1f94e9 Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Reviewed-on: https://review.lttng.org/c/babeltrace/+/1759
CI-Build: Philippe Proulx <eeppeliteloop@gmail.com> Tested-by: jenkins <jenkins@lttng.org> Reviewed-by: Philippe Proulx <eeppeliteloop@gmail.com>
Philippe Proulx [Wed, 24 Jul 2019 21:57:01 +0000 (17:57 -0400)]
*_string() enumerator to string functions: remove common prefix
Those functions are used for logging and error cause messages. Removing
this prefix makes the output less noisy. The logging statement typically
indicates what field is logged anyway, for example:
projects / babeltrace.git / log
