Jérémie Galarneau [Wed, 21 Jun 2017 21:54:24 +0000 (17:54 -0400)]
Fix: ctf_fs_metadata is assumed to be non-null
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 21 Jun 2017 18:26:20 +0000 (14:26 -0400)]
Improve logging in the LTTng index parsing code
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 16 Jun 2017 20:12:35 +0000 (16:12 -0400)]
copytrace: check clock_class_count before getting the clock by index
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 16 Jun 2017 20:21:53 +0000 (16:21 -0400)]
Fix: wrong output name in --help for implicit CTF source
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 16 Jun 2017 19:21:47 +0000 (15:21 -0400)]
Test the trimmer component
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 16 Jun 2017 19:10:27 +0000 (15:10 -0400)]
pass --clock-gmt option to the trimmer if enabled
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 19 Jun 2017 15:53:36 +0000 (11:53 -0400)]
Tests: don't rely on the test script's location
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 16 Jun 2017 18:51:41 +0000 (14:51 -0400)]
test_trace_copy: compare the two traces with diff rather than wc -l
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 16 Jun 2017 18:30:30 +0000 (14:30 -0400)]
Test debug-info
Add a test program source code, instructions to compile it with the
already compiled libhello library, and provide a lttng-ust trace of this
program running.
The test runs babeltrace on this trace with the
--debug-info-target-prefix option set to the tree of babeltrace which
allows it to find the debug symbols, and we verify that the text output
contains the complete debug_info context structure.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 19 Jun 2017 15:29:37 +0000 (11:29 -0400)]
Clean-up: coding style adjustments
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 16 Jun 2017 23:06:16 +0000 (19:06 -0400)]
Cleanup: typo fix in comment
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Thu, 15 Jun 2017 19:29:34 +0000 (15:29 -0400)]
fs-sink: add the single-trace option
This option allows the user to control the output directory of the trace
written (instead of the path being derived from the source trace path
and hostname). This implies that the component only receives one trace.
If the output directory exists and is not empty, an error is returned.
To use it:
--component sink.ctf.fs --path /tmp/out -p single-trace=true
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Thu, 15 Jun 2017 17:48:48 +0000 (13:48 -0400)]
Test: validate that the output of ctf-fs sink is identical to source
Process all the "succeed" CTF traces with the CTF-FS sink, check the
return code of babeltrace and compare the number of events in the new
trace from the number of events in the source trace.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Thu, 15 Jun 2017 17:48:02 +0000 (13:48 -0400)]
Fix: event_header is optional
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Wed, 14 Jun 2017 21:38:24 +0000 (17:38 -0400)]
Copy the packet_header in the plugins
The trimmer, debug-info and fs-sink plugins now copy the packet_header
from the trace received as parameter.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Wed, 14 Jun 2017 19:05:20 +0000 (15:05 -0400)]
Fix: wrong return code check on try_set
try_set_structure_field_integer can return 0 if it did not perform any
action, it is not an error.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Wed, 14 Jun 2017 18:45:46 +0000 (14:45 -0400)]
Fix copytrace: copy packet_context
Since we are not filtering the fields in the packet_context anymore, we
can now use the deep copy function bt_ctf_field_copy for the
packet_context.
Refactor the code to work with that and create a function to set the
packet_context on a packet or on a stream.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Wed, 14 Jun 2017 18:23:43 +0000 (14:23 -0400)]
copytrace: packet_context is optional
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Wed, 14 Jun 2017 18:21:59 +0000 (14:21 -0400)]
copytrace: create empty stream classes by default
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Wed, 14 Jun 2017 18:21:27 +0000 (14:21 -0400)]
fs-sink: packet_context is optional
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Tue, 13 Jun 2017 18:01:25 +0000 (14:01 -0400)]
Fix: the fields to reset are optional
This is a not a fatal error, just ignore if the fields are not there.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 16 Jun 2017 01:36:57 +0000 (21:36 -0400)]
ctf: remove strict metadata mode, update automatic CC mapping behaviour
Update the automatic mapping of specific fields to clock classes when
no explicit mapping is found in the integer field type to follow this:
| 0 CCs | 1 CC | 2+ CCs
--------------------------------------------------------------------
FT is mapped | Unreachable | OK as is | OK as is
FT is not mapped | Map to auto 1 GHz | Map to single CC | Error
The (FT is not mapped, 0 CC) cell follows Babeltrace 1's behaviour: when
a special timestamp field is found in the packet context
(`timestamp_begin` or `timestamp_end`) or event header (`timestamp`)
scope, when this timestamp field is not explicitly mapped to a clock
class, and when there's no clock class at this point in the trace, then
create an automatic, 1 GHz clock class and, in this case, name it
`default`. Map the field to this implicit 1 GHz clock class. See
section 8 ("Clocks") of CTF v1.8.2.
Cell (FT is not mapped, 0 CC) is a fix for old versions of LTTng, and
cell (FT is not mapped, 1 CC) is a fix for old versions of CTF writer.
Cell (FT is not mapped, 2+ CCs) is an error because we wouldn't know
which clock class to automatically map the field to.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 15 Jun 2017 22:38:45 +0000 (18:38 -0400)]
Remove the concept of event class attributes
The concept of event class attributes could be confusing with the
eventual concept of user attributes to be introduced in CTF 2. This is
not strictly required for Babeltrace 2.0: we can create dedicated
accessors for CTF 1.8 attributes (ID, name, log level, EMF URI). This is
what this patch does, removing the attributes API and adding:
* bt_ctf_event_class_get_log_level()
* bt_ctf_event_class_set_log_level()
* bt_ctf_event_class_get_emf_uri()
* bt_ctf_event_class_set_emf_uri()
There are already accessors for the name and ID attributes. As to the
stream class ID (`stream_id`) attribute, it should never have existed in
the first place because you can always use the parent stream class's ID.
Thus this patch also fixes a bug where there was a synchronization
problem between a stream class's ID and the `stream_id` attribute of its
contained event classes.
The bt_ctf_event_class_get_log_level() and
bt_ctf_event_class_set_log_level() functions use the new
`enum bt_ctf_event_class_log_level` enumeration which contains all the
log levels of LTTng-UST, which happen to be CTF 1.8's log levels too
because Babeltrace 1 prints those names for those values whatever the
trace's producer. A translation is performed from the numeric values to
this enumeration in plugins/ctf/common/metadata/visitor-generate-ir.c,
and the reverse is done in plugins/text/pretty/print.c. Knowing this,
the `loglevel_string` attribute that visitor-generate-ir.c used to add
to event classes (and which plugins/text/pretty/print.c used to
pretty-print the names) is removed. This was also an issue because
Babeltrace 1 does not know `loglevel_string` and sink.ctf.fs, through
the CTF writer implementation, was serializing this attribute in the
generated metadata.
tests/lib/test_ctf_writer.c is updated to test the individual accessors
instead of the generic attribute API.
plugins/libctfcopytrace/ctfcopytrace.c is updated to copy the known
attributes intead of using the generic attribute API.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 15 Jun 2017 20:36:47 +0000 (16:36 -0400)]
BT_CTF_BYTE_ORDER_NONE -> BT_CTF_BYTE_ORDER_UNSPECIFIED to match base
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 15 Jun 2017 20:32:38 +0000 (16:32 -0400)]
test_ctf_writer.c: do not check for invalid event class name
All event class names are valid. This is an old limitation.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 15 Jun 2017 20:32:16 +0000 (16:32 -0400)]
plugins/lttng-utils/Makefile.am: add all source files to _SOURCES var.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 15 Jun 2017 20:10:23 +0000 (16:10 -0400)]
Add BT_CTF_INTEGER_BASE_UNSPECIFIED
This is in preparation for CTF 2 in which a default base is not
specified for integer-like field types.
sink.text.pretty defaults to BT_CTF_INTEGER_BASE_DECIMAL when it gets
BT_CTF_INTEGER_BASE_UNSPECIFIED.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 16 Jun 2017 19:16:45 +0000 (15:16 -0400)]
Fix: Stream classes' id field is always serialized
A stream class' "id" must not be expressed in the trace metadata
if the trace's packet header does not contain a stream_id field.
Otherwise, the reader will try to ensure the packets' stream_id
corresponds to the stream class' id and will fail to do so.
The fix is to omit the serialization of the stream class' id
field entirely if the stream_id field is not found within the
trace's packet header.
It remains unclear whether or not a stream class has an actual
"id" in this case, or if some "unspecified" id state should
be implemented.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Tue, 13 Jun 2017 18:03:50 +0000 (14:03 -0400)]
Fix: flush stream without packet_context
Ensure we can flush a stream with only one packet if it does not have a
packet_context or the packet_size field.
Also truncate the stream to the content_size instead of the default 32kB
packet size.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 16 Jun 2017 18:55:52 +0000 (14:55 -0400)]
Add BT_LOG*_ERRNO() logging macros
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Mon, 12 Jun 2017 19:06:27 +0000 (15:06 -0400)]
debug-info fix: packet_context is optional
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Jun 2017 22:13:31 +0000 (18:13 -0400)]
Fix: uninitialized return value in copy.c
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 20:30:31 +0000 (16:30 -0400)]
Remove legacy printf_verbose()/printf_debug() and others
This patch removes all the legacy logging macros found in
babeltrace-internal.h. It also removes the babeltrace_verbose and
babeltrace_debug symbols.
To remain as backward compatible as possible, the CLI checks the
BABELTRACE_VERBOSE and BABELTRACE_DEBUG environment variables and
overrides the default log level (WARN, or the one set with --log-level)
with VERBOSE if BABELTRACE_DEBUG is 1 and INFO if only
BABELTRACE_VERBOSE is 1. This default log level applies to all the known
loggers, except the ones that have their log level explicitly set with a
dedicated environment variable.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 16:34:00 +0000 (12:34 -0400)]
visitor-generate-ir.c: automatically map specific fields to trace's clock class
In non-strict mode (default), automatically map packet context field
type's `timestamp_begin` and `timestamp_end` fields, as well as
event header field type's `timestamp` and `ts` fields, when the
following conditions are all true:
* The decoder is in non-strict mode.
* The field type is an integer field type.
* The integer field type is not mapped to a clock class.
* The trace contains only one clock class at this point.
This attempts to fix semantics errors committed by some known tracers
(very old versions of LTTng and old versions of CTF writer).
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 15:15:00 +0000 (11:15 -0400)]
ctf: add metadata decoder config. and `strict-metadata` param in src.ctf.fs
The common metadata decoder now accepts an optional configuration
structure. The offset (seconds and nanoseconds) to apply to all the
clock classes is found in there, as well as a new option which makes the
decoder strict. In strict mode, the decoder does not allow the decoded
metadata stream to be off semantically. For example, in strict mode,
clock classes named `monotonic` are not forced to be absolute when it is
detected that the stream was produced by an LTTng tracer.
It's not expected that the strict mode be something that any user would
want under normal conditions because the non-strict mode fixes errors
made by known tracers to make things work for the other components of
the graph. As such, there's no quick --strict option in the CLI's
`convert` command: the user can instantiate its own source.ctf.fs
component with --component for this and pass the options with --params:
babeltrace -c src.ctf.fs --path ~/my-traces/trace -p strict-metadata=yes
I'm also fixing the parameter names and types for the clock class
offsets, both in the CLI and in source.ctf.fs, so that all the project's
modules are in sync regarding this. The parameters are now
`clock-class-offset-s` and `clock-class-offset-ns`, and with the
`convert` command, they are set with resp. --clock-offset and
--clock-offset-ns. They used to be passed to the `run` command as string
values (using --key and --value), but now they are part of the
component's --params option so that they are recognized as constant
integer values (even if negative).
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 06:50:29 +0000 (02:50 -0400)]
source.ctf.fs: remove ctf_fs_debug symbol
Not needed anymore because of the standardized logging.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 06:31:12 +0000 (02:31 -0400)]
Fix: cli: help command: accept leftover when --component is not specified
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 05:17:23 +0000 (01:17 -0400)]
cli: add global --log-level option
This new global option (applies before the command name) sets the
default log level of all the known (project's) loggers. This default
log level can still be overridden by a specific environment variable,
for example:
BABELTRACE_PLUGIN_CTF_METADATA_LOG_LEVEL=D babeltrace --log-level=I ...
In this example, all the log levels are set to INFO, but
BABELTRACE_PLUGIN_CTF_METADATA_LOG_LEVEL overrides the `ctf` plugin's
metadata module's log level to DEBUG.
When you don't specify --log-level, it defaults to W (WARN), the current
behaviour.
--verbose is now the equivalent of --log-level=I, and --debug is the
equivalent of --log-level=V. --verbose and --debug do the same thing,
for the `convert` command, whether they are global or specific to the
command. --verbose and --debug can still be specified together, and
whatever the order, the lowest log level always wins: VERBOSE.
The --log-level option can override previous --verbose, --debug, or
--log-level options because it didn't exist in Babeltrace 1 anyway so
there's no way to break backward compatibility:
babeltrace --log-level=W --verbose --log-level=E ...
In this case, the final log level is ERROR.
Other example with implicit `convert` command:
babeltrace --log-level=E /path/to/trace -v
In this case, the command's --verbose option sets the log level to
INFO because it's lower than ERROR (set by --log-level=E).
One of the main use cases of the new --log-level option is to completely
disable logging:
babeltrace --log-level=N ...
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 02:42:58 +0000 (22:42 -0400)]
Use BABELTRACE_FORCE_COLORS environment variable to force color support
This allows this use case:
BABELTRACE_FORCE_COLORS=1 babeltrace ... |& less -R
to see the logging's color in `less`.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 02:42:00 +0000 (22:42 -0400)]
Logging: add colors depending on log level
The whole line is colored and this indicates the log level.
VERBOSE: no color
DEBUG: no color
INFO: blue
WARN: yellow
ERROR: red
FATAL: red
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 02:07:10 +0000 (22:07 -0400)]
visitor-generate-ir.c: logging: add stream's line number and other info
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 00:31:08 +0000 (20:31 -0400)]
visitor-semantic-validator.c: standardize logging
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 00:30:55 +0000 (20:30 -0400)]
plugins/ctf/common/metadata/visitor-parent-links.c: standardize logging
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 00:09:47 +0000 (20:09 -0400)]
plugins/ctf/common/metadata/objstack.c: add logging
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 14 Jun 2017 00:07:36 +0000 (20:07 -0400)]
plugins/ctf/common/metadata/decoder.c: standardize logging
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Tue, 13 Jun 2017 22:35:16 +0000 (18:35 -0400)]
plugins/ctf/common/metadata: logging: standardize in parser and lexer
YYDEBUG is set if the VERBOSE mode is enabled for the common metadata
module. YYFPRINTF is defined to BT_LOGV so that everything uses the
standard logging macros.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Tue, 13 Jun 2017 20:20:09 +0000 (16:20 -0400)]
visitor-generate-ir.c: logging: standardize messages and levels
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Mon, 12 Jun 2017 22:03:40 +0000 (18:03 -0400)]
ir: fields.c: fix public function names to get/set struct. field
To remain symmetric, use bt_ctf_field_structure_get_field_by_name() and
bt_ctf_field_structure_set_field_by_name(), as it is possible that
bt_ctf_field_structure_set_field_by_index() exists in the future.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Tue, 25 Oct 2016 21:13:57 +0000 (17:13 -0400)]
Port: Enable libtool to produce DLLs
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Mon, 5 Jun 2017 18:00:40 +0000 (14:00 -0400)]
Port: Remove unused header pwd.h
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Fri, 2 Jun 2017 17:43:18 +0000 (13:43 -0400)]
Add logging to compat
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Mon, 29 May 2017 20:49:17 +0000 (16:49 -0400)]
Add missing files to gitignore
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Fri, 17 Feb 2017 20:11:57 +0000 (15:11 -0500)]
Port: Remove unused header in test_plugin
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Wed, 15 Feb 2017 20:39:34 +0000 (15:39 -0500)]
Don't add module symbols to global namespace
The G_MODULE_BIND_LOCAL flag specifies that symbols in the module should not
be added to the global name space. The default action on most platforms is
to place symbols in the module in the global name space, which may cause
conflicts with existing symbols.
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 9 Jun 2017 23:23:58 +0000 (19:23 -0400)]
Set version extra to pre1
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 23:00:22 +0000 (19:00 -0400)]
Fix: debug info plugin incorrect assumption about header and clock
Fix a few assumptions in the copy helper:
- A stream may not have event headers,
- A trace may not have clocks.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 22:38:36 +0000 (18:38 -0400)]
Fix: various fixes for ctf-traces/succeed test cases
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 21:52:01 +0000 (17:52 -0400)]
configure.ac: make the minimal, build-time log level be VERBOSE
Distributions can override this at build time to increase performance,
e.g.:
BABELTRACE_MINIMAL_LOG_LEVEL=WARN ./configure ...
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 21:48:27 +0000 (17:48 -0400)]
cli: use WARN as the default log level when not explicitly set by user
For all known log levels (library, CLI, plugins, etc.), when the
dedicated log level environment variable is unset, make the CLI set it
to WARN. This means that you can always override the default, -v, or -d
log level with a specific environment variable. For example, if you need
VERBOSE logging for all the loggers but the library's:
BABELTRACE_LOGGING_GLOBAL_LEVEL=N babeltrace -d ...
Using WARN as the default log level will help spot bugs quickly without
a specific option to use this log level.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 21:30:07 +0000 (17:30 -0400)]
cli: set all the log level of all known loggers with -v and -d
Set log levels according to --debug or --verbose. For
backward compatibility, --debug is more verbose than
--verbose. So:
--verbose: INFO log level
--debug: VERBOSE log level (includes DEBUG, which is
is less verbose than VERBOSE in the internal
logging framework)
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 21:15:55 +0000 (17:15 -0400)]
cli: convert: allow multiple paths to be passed as leftover arguments
Allow this:
babeltrace /path/to/trace1 /path/to/trace2 /path/to/trace3
This worked with Babeltrace 1 and needs to work with Babeltrace 2.
What's done here is that a _base_ implicit set of arguments for an
eventual source.ctf.fs component is created implicitly or by default
like it used to be. For each leftover argument, we copy this base set of
arguments and append the `path` parameter with the specific leftover
argument. Then we name and connect those source components as usual.
What's left is one source.ctf.fs component for each individual path.
Each individual component can still recurse.
tests/cli/test_convert_args is updated because it used to check that a
command line with multiple leftovers was expected to fail, but it
doesn't with this patch.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 9 Jun 2017 20:24:36 +0000 (16:24 -0400)]
Fix: remove listener check
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 9 Jun 2017 19:22:22 +0000 (15:22 -0400)]
Fix copytrace: check field exists before copy
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 22:05:07 +0000 (18:05 -0400)]
Fix: uninitialized variable
query.c: In function 'trace_info_query':
query.c:246:2: warning: 'stream_class' may be used uninitialized in this function [-Wmaybe-uninitialized]
bt_put(stream_class);
^
query.c:216:30: note: 'stream_class' was declared here
struct bt_ctf_stream_class *stream_class;
^
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 21:41:53 +0000 (17:41 -0400)]
Fix: event, stream names can be strings
The code is too strict and enforces that they need to be CTF
identifiers, but it's valid to have those as simple strings.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 21:28:42 +0000 (17:28 -0400)]
Fix: make check in out-of-tree
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 21:24:27 +0000 (17:24 -0400)]
Fix: out-of-tree build
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Fri, 9 Jun 2017 21:01:23 +0000 (17:01 -0400)]
Fix: add missing files to make dist, disable python bindings
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 9 Jun 2017 18:01:00 +0000 (14:01 -0400)]
Remove former logging header print.h from source.ctf.fs Makefile
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 7 Jun 2017 20:21:18 +0000 (16:21 -0400)]
source.ctf.fs: implement the trace-info query
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 19:03:46 +0000 (15:03 -0400)]
plugins/lttng-utils/copy.c: fix uninitialized use warning
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Fri, 9 Jun 2017 16:05:01 +0000 (12:05 -0400)]
debug-info: sync the copied trace on static change
When the trace becomes static, we need to ensure that we have all the
streams, stream classes and event classes in our own copy before setting
it static as well.
This is also true when we open the trace and it is already static.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Thu, 8 Jun 2017 18:10:29 +0000 (14:10 -0400)]
debug-info: handle the static notifications
And some cleanup, fix the remaining leaks.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Tue, 6 Jun 2017 21:02:10 +0000 (17:02 -0400)]
debug-info refactoring
Scope the objects belonging to a trace (stream, stream_class, packet,
debug info maps) into a debug_info_trace and use that everywhere instead
of having these objects global to the component.
This will allow to make the static notification handling work.
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Julien Desfossez [Tue, 6 Jun 2017 16:01:48 +0000 (12:01 -0400)]
fs-sink: useless active_stream member
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 00:56:50 +0000 (20:56 -0400)]
include/babeltrace/graph/private-connection.h: add missing `extern`
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 23 Feb 2017 22:05:15 +0000 (17:05 -0500)]
doc/api: add missing babeltrace-cli.png image
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 17:38:08 +0000 (13:38 -0400)]
cli: use -o ctf --output /output/path to quickly instantiate sink.ctf.fs
This should be easier for existing users of Babeltrace 1 which are used
to the --output-format (-o) and --output (-w) options. We had people
previously on the mailing list asking why
babeltrace my-trace -o ctf --output /output/path
was not working. The answer is that there was no CTF sink at the time,
but this usage is still correct and straightforward.
The code makes sure that, if something implies `-o text`, for example
--names, --fields, or --no-delta, then you cannot use this form because
--output could also apply to the implicit sink.pretty.text component. We
also make sure that --output exists if `-o ctf` exists (--output without
`-o ctf` applies to the implicit `-o text` as mentioned previously).
The run command equivalent of this command:
babeltrace my-trace -o ctf --output /output/path
as returned by --run-args, is:
babeltrace run --component source.ctf.fs --name source-ctf-fs
--key path --value my-trace --component sink.ctf.fs
--name sink-ctf-fs --key path --value /output/path
--component filter.utils.muxer --name muxer
--component filter.lttng-utils.debug-info
--name debug-info --connect source-ctf-fs:muxer
--connect muxer:debug-info
--connect debug-info:sink-ctf-fs
tests/cli/test_convert_args is updated because the preferred name of the
implicit source.ctf.fs component is changed: it used to be `ctf-fs`,
it's now `source-ctf-fs`, as the preferred implicit sink.ctf.fs
component's name is `sink-ctf-fs`. Also there are a few more tests to
verify the constraint and generated options mentioned in this message.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 00:48:01 +0000 (20:48 -0400)]
bt_ctf_field_type_integer_set_size(): take unsigned int parameter
Take unsigned int parameter like bt_ctf_field_type_integer_create().
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Jun 2017 00:47:25 +0000 (20:47 -0400)]
bt_ctf_stream_get_discarded_events_count(): return int
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Jun 2017 23:00:07 +0000 (19:00 -0400)]
utils.muxer: expect specific clock class properties to mux
This patch makes a filter.utils.muxer component's notification iterator
set what properties it expects from future clock classes from the first
clock classes it checks. The possible expectations are:
* Expect absolute clock classes, any UUID.
* Expect non-absolute clock classes, without a UUID.
* Expect non-absolute clock classes, with a specific UUID.
With this patch you can make the component mux the streams of a single
trace of which the clock is not absolute, or has no UUID.
The `assume-absolute-clock-classes` parameter overrides all those
checks.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Jun 2017 22:58:33 +0000 (18:58 -0400)]
ir: do not automatically generate a UUID in bt_ctf_clock_class_create()
A clock class should be created empty, without a generated UUID which is
different each time. To maintain backward compatibility, a CTF writer
clock automatically generates a UUID on creation to set its private
clock class's UUID.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Thu, 8 Jun 2017 19:38:46 +0000 (15:38 -0400)]
plugin-so: use list of components instead of glib hash table
Use a list of components to track those that have references to their so
handles. This allows removing a dependency on the glib hash table, which
is not very cooperative when initialized from a library constructor in a
static build (mainly on Windows).
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Jun 2017 19:57:08 +0000 (15:57 -0400)]
cli: use --component=NAME:{source,filter,sink}.PLUGIN.COMPCLS
Use a universal -c/--component option for all component class types,
specifying the component class type in the option's argument instead,
for example:
Before: --source=ctf.fs
After: --component=source.ctf.fs
Before: --sink=my-text:text.pretty
After: --component=my-text:sink.text.pretty
This provides a universal way to identify a component instance in:
* Log messages
* Information/error messages
* List of plugins and contained component classes
* Help messages
* Documentation
Then we can always refer to the ctf.fs source component class, for
example, as `source.ctf.fs` instead of all the variations:
* The ctf.fs source
* --source ctf.fs
* ctf.fs (source)
and so on.
Short option -c used to be for --connect: it's -C now.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Thu, 8 Jun 2017 17:35:54 +0000 (13:35 -0400)]
lttng-live: check graph cancel state, check ctf trace pointer
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Jun 2017 17:13:07 +0000 (13:13 -0400)]
Fix build with BUILT_IN_PLUGINS=1
Changes:
* Make library always depend on its convenience libraries. Since the
library is not linked with --whole-archive in the CLI when
BUILT_IN_PLUGINS=1 (like plugins are), it's okay for it to include
convenience symbols.
* Do not build tests/lib/test_plugin when BUILT_IN_PLUGINS=1
* Do not build tests/plugins when BUILT_IN_PLUGINS=1
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Jun 2017 02:25:48 +0000 (22:25 -0400)]
Add bt_graph_add_component(), make bt_component_create() internal
With this patch, you can only instantiate a component class through an
existing graph. This can guarantee to the user (component class
developer) that, when the component class's initialization method is
called, the component always belongs to a graph. This simplifies things,
now and in the future. For example, it is possible to cancel the graph
during the execution of bt_graph_add_component() so that the user method
knows if it should retry now or return an error when the graph is
canceled.
bt_graph_add_component() returns BT_GRAPH_STATUS_CANCELED if the graph
is canceled. If the user's initialization method fails, its return
component status is converted to the appropriate graph status.
Another advantage of adding a component to a graph is that, if you add a
"port added" graph listener, it is guaranteed that the listener is
called even for the initial ports of a component. This was not possible
before.
This patch also makes a component's name mandatory, and
bt_graph_add_component() ensures that there are components sharing the
same name part of a given graph. This is a constraint similar to the
fact that all the ports of a component must have a unique name. It
should help to the implementation of a function to retrieve a component
by name from a graph in the future.
We do not check that a graph has or not "loose", disconnected components
in bt_graph_consume(). This is not a change because any component can
remove all its ports once connected once and become loose. A graph with
floating/loose components is said to be disconnected in classical graph
theory. This is not a problem for the moment, as the connected
components can still interact. The disconnected components should
probably be marked and then destroyed in bt_graph_consume() and/or in a
dedicated garbage collecting function if needed
(bt_graph_remove_disconnected_components() for example).
This patch also makes successfully user-initialized components marked as
such so that the user's finalization method is not called if its
initialization method failed. In other words, the initialization method
is responsible for any cleanup if it's about to fail.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 7 Jun 2017 23:23:58 +0000 (19:23 -0400)]
lttng-live: correctly handle ctrl-c and fix leaks
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 22:31:41 +0000 (18:31 -0400)]
End connection on destruction
End connection on destruction so that each port is disconnected once the
connection is destroyed. Otherwise each port keeps its connection
pointer as is which point to freed memory.
bt_connection_end() already makes sure to finalize each created
notification iterator, so all is left in the destructor is to free the
array of iterators and the connection itself.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 22:12:03 +0000 (18:12 -0400)]
Cancel the graph on destruction
Consider the graph as canceled at the beginning of its destruction. This
prohibits the use of some operations that are restricted to be used when
the graph is active and running.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 22:03:54 +0000 (18:03 -0400)]
connection.c: refuse to create a notif. iter. when the graph is canceled
It is useless to create a notification iterator from a connection which
belongs to a canceled graph.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 22:01:16 +0000 (18:01 -0400)]
lib/graph/{filter.c,sink.c,source.c}: fix log messages
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 7 Jun 2017 20:17:35 +0000 (16:17 -0400)]
Remove dependancies on ctf.fs source component caused by former logging API
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 5 Jun 2017 18:30:16 +0000 (14:30 -0400)]
Move ctf-fs source query implementations to their own file
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 21:51:46 +0000 (17:51 -0400)]
Fix: do not use BT_MOVE() when the destination memory could be uninitialized
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 21:45:59 +0000 (17:45 -0400)]
Make bt_private_connection_create_notification_iterator() return a status code
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 20:24:13 +0000 (16:24 -0400)]
Refuse to add port to component when parent graph is canceled
The objects owned by a graph should not be able to perform any useless
action. Having those constraints promise that certain contained objects
remain within certain states when the graph is canceled.
In this case, it is useless for a component to add a port, since this
port cannot be connected to another one within the same graph because
bt_graph_connect_ports() is not permitted when the graph is canceled.
We still allow ports to be removed when the graph is canceled for
symetric teardowns, for example.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 20:12:56 +0000 (16:12 -0400)]
graph.c: do not warn when the graph is canceled during bt_graph_run()
If the graph is canceled during bt_graph_run()'s loop, do not warn: this
was intentionally done during the execution by another thread or by a
signal handler. Log a debug message and quit with the usual
BT_GRAPH_STATUS_CANCELED status.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 19:27:09 +0000 (15:27 -0400)]
Add internal bt_component_borrow_graph()
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Wed, 7 Jun 2017 19:26:34 +0000 (15:26 -0400)]
Make bt_private_component_*_add_*_port() return a status code
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
This page took 0.043328 seconds and 4 git commands to generate.