Alexandre Montplaisir [Tue, 15 May 2012 14:20:57 +0000 (10:20 -0400)]
tmf: Cleanup generated files when a request is cancelled
This adds methods to state storage backends to define how to
delete the files it generated.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 18 May 2012 18:20:45 +0000 (14:20 -0400)]
lttng: Handle the case where the trace is empty
Alexandre Montplaisir [Fri, 18 May 2012 17:53:55 +0000 (13:53 -0400)]
tmf: Use the correct data type in coalesced requests
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Thu, 17 May 2012 19:05:16 +0000 (15:05 -0400)]
tmf: Avoid a try at every loop in range queries
It's more simple to define the last valid end time first.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Thu, 17 May 2012 18:54:00 +0000 (14:54 -0400)]
tmf: Correctly store null state values in the backend
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Thu, 17 May 2012 15:30:58 +0000 (11:30 -0400)]
tmf: Make sure the resolution in range queries is valid
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Thu, 17 May 2012 15:25:15 +0000 (11:25 -0400)]
tmf: Start resolution range queries at the t1 parameter
...not at the start time of the first interval. The
"resolution" should be aligned to t1, the interval's
start time is irrelevant here.
Also updated one of the tests, this changes the expected results.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Bernd Hufmann [Wed, 16 May 2012 19:58:55 +0000 (15:58 -0400)]
Replace printStackTrace() with proper logging in TMF and LTTng
Matthew Khouzam [Tue, 15 May 2012 19:00:04 +0000 (15:00 -0400)]
Fix some findbugs
Matthew Khouzam [Tue, 15 May 2012 18:58:16 +0000 (14:58 -0400)]
Fix seek of empty stream
Matthew Khouzam [Tue, 15 May 2012 18:35:38 +0000 (14:35 -0400)]
Check if a trace is empty before going to the last event.
Matthew Khouzam [Tue, 15 May 2012 18:26:54 +0000 (14:26 -0400)]
Fix go to last event.
Alexandre Montplaisir [Tue, 15 May 2012 18:12:59 +0000 (14:12 -0400)]
tmf: API fix for the state system queries
Wanted to get this in before the API freezes...
Rename loadStateAtTime() -> queryFullState(). Originally this
would simply seek an inner vector to the given location, now
it actually returns that vector. It's more in line with the
other query*() methods.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 15 May 2012 16:00:37 +0000 (12:00 -0400)]
lttng: Don't update the status of dead processes
There is always a sched_switch that will follow a process_exit,
to take away the now-dead process from the CPU. In that particular
case, we don't want to set its status to Waiting, since it's
not waiting anymore, it's dead (Jim).
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 15 May 2012 14:43:15 +0000 (10:43 -0400)]
tmf: Also pick the interval at the end time of range queries
When doing a range query with resolution, include the end time
of the range (t2) as the last "resolution point". This means
that the interval crossing t2 will be returned no matter what,
which is consistent with range queries with no resolution.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Bernd Hufmann [Mon, 14 May 2012 16:07:20 +0000 (12:07 -0400)]
Use supplementary directory for state history tree
Bernd Hufmann [Tue, 15 May 2012 11:18:43 +0000 (07:18 -0400)]
Add missing TMF constants file
Alexandre Montplaisir [Mon, 14 May 2012 20:36:48 +0000 (16:36 -0400)]
Revert "Use supplementary directory for state history tree"
This reverts commit
9716e61fa1550422463f938cf0767a0130187ef5.
A new TmfCommonConstants file was missing from it. Unfortunately
the author is gone for today, so we'll temporarily revert this
commit so the thing keeps building.
Alexandre Montplaisir [Mon, 14 May 2012 19:55:57 +0000 (15:55 -0400)]
tmf: Work around a simple CtfTmfEventType warning
The automatic assign-on-construction isn't ideal, but at least
we expect this behavior.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Mon, 14 May 2012 19:19:45 +0000 (15:19 -0400)]
tmf: Also assign the end time when opening a CTF trace
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Mathieu Denis [Mon, 14 May 2012 20:03:28 +0000 (16:03 -0400)]
Fixed lost event bug.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Bernd Hufmann [Mon, 14 May 2012 16:07:20 +0000 (12:07 -0400)]
Use supplementary directory for state history tree
Bernd Hufmann [Thu, 10 May 2012 18:05:08 +0000 (14:05 -0400)]
Add handling of supplementary files of traces
Alexandre Montplaisir [Mon, 14 May 2012 14:33:05 +0000 (10:33 -0400)]
lttng: Use a different CtfTmfTrace object for every test
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Matthew Khouzam [Fri, 11 May 2012 21:49:09 +0000 (17:49 -0400)]
Clean up code.
Alexandre Montplaisir [Fri, 11 May 2012 22:41:40 +0000 (18:41 -0400)]
lttng: Make sure we assign a value to all created quarks
Rework the fix in
f3013ba. If a quark gets created but is never
ever used over the span of the whole history, it ends up not
having any value whatsover, not even null. At query time, it
would query the empty space in the HT nodes for its value,
returning 0 most of the time.
Now we manually assign a value to all Thread sub-attributes,
depending on the event type.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 11 May 2012 22:51:24 +0000 (18:51 -0400)]
tmf: Use static final's for the StateValue types
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Thu, 10 May 2012 20:54:25 +0000 (16:54 -0400)]
tmf: Make HistoryBuilder use a TMF request for its events
Initially, the IStateChangeInput object would have to open
the trace itself, and merely send state changes. Rework this
interface so that we now have to pass it the events. This
allows HistoryBuilder to hook into TMF's event requests, and
benefit from the coalescing.
This will be tweaked a bit more soon, when we put all indexing
requests together.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Thu, 10 May 2012 19:41:54 +0000 (15:41 -0400)]
tmf: Clarify why CtfTmfEvent's constructor is public
The default constructor for null event should be accessed
via the static singleton constructor in normal cases. I tried
to change it back to private, but it breaks some OSGI stuff
because this is used in extension points.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Matthew Khouzam [Fri, 11 May 2012 21:04:31 +0000 (17:04 -0400)]
Change tabs for commas to separate fields.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Fri, 11 May 2012 20:56:13 +0000 (16:56 -0400)]
Clean up Parser and Lexer. Some fixes are more for findbugs than actual
bugs.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Xavier Raynaud [Thu, 10 May 2012 07:53:07 +0000 (09:53 +0200)]
make Resolution enum public, since it is used in public API
Alexandre Montplaisir [Wed, 9 May 2012 21:48:29 +0000 (17:48 -0400)]
lttng: Split process status in user and kernel modes
The attribute Thread/Status will now also track if the
thread is in kernel mode (in a system call) or in user mode.
This will match more closely what is wanted in the Control
Flow View.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Wed, 9 May 2012 20:15:50 +0000 (16:15 -0400)]
tmf: Ensure we don't return null on GSS queries
This would indicate a problem with the backend anyway, since
by design there should be no "holes" in a state history.
But since this can still happen, if there's a problem with the
backend or if the file is corrupted for example, then we should
catch (and log) the error here, and guarantee what is sent up
is non-null.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 8 May 2012 20:00:32 +0000 (16:00 -0400)]
tmf: Synchronize accesses to the history files
It seems the file channel positions are saved in each thread's
context on Linux, but not on Windows. This makes concurrent
queries to the state history very prone to breakage. This
small change makes sure a complete node is read before a
thread can be preempted.
In pratice the performance impact shouldn't be too big, since
both threads would be blocked by the same disk even if we
allowed concurrent access.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Mon, 7 May 2012 21:58:01 +0000 (17:58 -0400)]
lttng: Change syscalls to a flat attribute instead of a stack
This will not show nested syscalls correctly, but those were
very rare in the first place, and have been completely removed
in recent kernel versions.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Mon, 7 May 2012 18:54:18 +0000 (14:54 -0400)]
tmf: Provide a way to get the number of attributes in a system
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Matthew Khouzam [Wed, 9 May 2012 21:00:48 +0000 (17:00 -0400)]
improve test coverage in CtfLocation
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Patrick Tasse [Wed, 9 May 2012 20:15:31 +0000 (16:15 -0400)]
Fix seek behaviour in CtfAdaptor
Matthew Khouzam [Wed, 9 May 2012 19:02:27 +0000 (15:02 -0400)]
Improve test coverage.
Fix EventDeclaration.equals()
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Wed, 9 May 2012 19:00:55 +0000 (15:00 -0400)]
Add toString to CtfLocation to allow easier debugging.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Wed, 9 May 2012 15:19:35 +0000 (11:19 -0400)]
Improve API.
Update test cases.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Fri, 4 May 2012 19:48:09 +0000 (15:48 -0400)]
Fix test case. It really is supposed to return true.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Patrick Tasse [Fri, 4 May 2012 19:58:05 +0000 (15:58 -0400)]
Fix for Linux tree item height bug.
Bernd Hufmann [Fri, 4 May 2012 19:46:47 +0000 (15:46 -0400)]
Increase GUI delay value for GUI test case
Bernd Hufmann [Fri, 4 May 2012 18:54:43 +0000 (14:54 -0400)]
Update internal packages export in LTTng 2.0 control + update java doc
Bernd Hufmann [Fri, 4 May 2012 17:33:27 +0000 (13:33 -0400)]
Improve package tangle index for LTTng 2.0 control design
Alexandre Montplaisir [Fri, 4 May 2012 17:58:27 +0000 (13:58 -0400)]
ctf: Allow finalizing CTFTrace's if validation failed
A NPE could happen during finalize() if an exception was thrown
before the constructor could complete. Move the new-ification of
the FileChannel container to the beginning, and use a List instead
of a static array.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Patrick Tasse [Fri, 4 May 2012 17:02:33 +0000 (13:02 -0400)]
Move raw event viewer widget to its own package.
Alexandre Montplaisir [Fri, 4 May 2012 15:55:07 +0000 (11:55 -0400)]
tmf: Don't close the input stream in the SHS constructor
This stream belongs to the History Tree itself, so if we close
it here we could hit ClosedChannelExceptions later on, under
certain conditions.
Instead, don't keep a local reference to this stream, so static
analysis tools don't expect us to close it.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Matthew Khouzam [Thu, 3 May 2012 22:55:57 +0000 (18:55 -0400)]
Improve test cases. Coverage of 85%+ and fix bugs.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Thu, 3 May 2012 18:28:32 +0000 (14:28 -0400)]
Add javadoc and comments
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Bernd Hufmann [Thu, 3 May 2012 20:07:59 +0000 (16:07 -0400)]
More fixes of static analysis warnings for UML2SD
Patrick Tasse [Thu, 3 May 2012 20:14:05 +0000 (16:14 -0400)]
Fix for Linux display bugs in TimeGraphCombo.
Patrick Tasse [Thu, 3 May 2012 18:42:19 +0000 (14:42 -0400)]
Bug 378402: Implementation of ControlFlow view and Resources view for
LTTng 2.0.
Patrick Tasse [Thu, 3 May 2012 18:36:28 +0000 (14:36 -0400)]
Bug 378401: Implementation of time graph widget.
Patrick Tasse [Thu, 3 May 2012 18:07:22 +0000 (14:07 -0400)]
Remove bookmarks file on drag&drop trace copy.
Patrick Tasse [Wed, 11 Apr 2012 22:17:53 +0000 (18:17 -0400)]
Move legacy time analysis widget to lttng.ui and create initial copy in
tmf.ui.
Matthew Khouzam [Thu, 3 May 2012 17:42:20 +0000 (13:42 -0400)]
Remove warnings and fix a windows bug with CtfTmfTrace.getName()
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Thu, 3 May 2012 17:03:31 +0000 (13:03 -0400)]
Improve CTF timestamp output, will now output in y/m/d, h:m:s.ns or ns or s (double)
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Bernd Hufmann [Thu, 3 May 2012 14:19:40 +0000 (10:19 -0400)]
Fix static analysis warnings for UML2SD
Bernd Hufmann [Wed, 2 May 2012 15:01:46 +0000 (11:01 -0400)]
Java Doc and API clean up of TMF UML Sequence diagram framework
Matthew Khouzam [Wed, 2 May 2012 19:58:20 +0000 (15:58 -0400)]
Improve test cases, speed and accuracy.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Alexandre Montplaisir [Tue, 1 May 2012 20:13:46 +0000 (16:13 -0400)]
tmf: Provide a special interval end time return value for views
The complete state model in TMF is based on the fact that states
end *before* the next state is started. This ensures we cannot
have 2 active states for a given attribute at a given timestamp.
However, this is not ideal for the viewers, because it leaves
gaps between intervals if we use those timestamps as-is. After
much head-banging, the simplest solution that was found was to
add a method in the ITmfStateInterval interface that viewers can
use to get their own version of the interval's end time.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Wed, 2 May 2012 03:30:28 +0000 (23:30 -0400)]
tmf: Correctly save the history backend's end time
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 22:05:50 +0000 (18:05 -0400)]
lttng: Rework the IRQ event handler
Implement the softIrq handlers. Drop the IRQ_stack sub-tree,
put the IRQ stuff under Resources/IRQs and Resources/Soft_IRQs.
This matches more closely what is needed by the Resource view.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 21:34:32 +0000 (17:34 -0400)]
lttng: Use scope-limiting braces in the event handler
This will prevent event-specific variables from leaking
from one case to another.
I used to get weird compiler errors with OpenJDK 7 about one
year ago when trying to do this, but I just retested with a more
recent version and it seems to work fine now.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 21:13:41 +0000 (17:13 -0400)]
lttng: Cache all top-level attribute nodes
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 19:29:32 +0000 (15:29 -0400)]
lttng: Always create the sub-attributes of TID nodes
This makes the history slightly bigger, but it avoids having
to catch all sort of exceptions in the views, for the cases
where the expected sub-attributes never get populated. Now
they will simply contain a "null" value for the whole duration.
Also relaxed the test cases a bit, changed it so we won't have
to update them every time we make a single change to the state
provider.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 18:52:58 +0000 (14:52 -0400)]
lttng: Don't update previous data wrt sched_switch trace event
Some event types in the event handler update past data
(to correctly identify "unknown" threads for example),
but this should not be done in the case of sched_switch.
The problem is that sched_switch is always called after
a sched_exit, and since the sched_exit's handler deletes
all information related to this process, we do not want
the following sched_switch to restore the Exec_name for
example.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 18:48:05 +0000 (14:48 -0400)]
tmf: Add icons for CTF and LTTng trace types
Matthew Khouzam [Tue, 1 May 2012 20:35:56 +0000 (16:35 -0400)]
fix empty trace bug
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Alexandre Montplaisir [Tue, 1 May 2012 05:12:02 +0000 (01:12 -0400)]
tmf: Internalize all the core state system implementation
Also renamed the old .helpers package to simply tmf.core.statesystem,
which only consists of the exported API now.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 05:06:12 +0000 (01:06 -0400)]
tmf: Move all GSS exceptions to tmf.core.exceptions
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 04:59:19 +0000 (00:59 -0400)]
tmf: Rework unneeded throw in the GSS backend
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Mon, 30 Apr 2012 20:04:28 +0000 (16:04 -0400)]
tmf: Split StateHistorySystem into two interfaces
This will offer a query-only interface, which has no method
to write to or modify the state history in any way. This is
much safer, now that the views are starting to use it.
As part of the process, the history-less StateSystem is no
longer easily available. It was never much used in practice,
but if it's ever needed in the future it should be relatively
easy to bring it back.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Matthew Khouzam [Tue, 1 May 2012 14:19:25 +0000 (10:19 -0400)]
Fix findbugs
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Tue, 1 May 2012 13:43:39 +0000 (09:43 -0400)]
Add test cases for float definitions.
Fix build error caused by rebase.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Thu, 26 Apr 2012 17:36:44 +0000 (13:36 -0400)]
Add support for lost events
Fix display of sequences and arrays
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Wed, 25 Apr 2012 20:58:15 +0000 (16:58 -0400)]
Fix problem with multiple channels including empty channels and ones
created after start.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Tue, 24 Apr 2012 21:39:00 +0000 (17:39 -0400)]
fix empty stream bug.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Fri, 20 Apr 2012 20:40:17 +0000 (16:40 -0400)]
Add support for CTF in Sequence Diagram view
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Mon, 23 Apr 2012 21:55:39 +0000 (17:55 -0400)]
Fix alignment issue. Now traces using more than one alignment can be
read.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Thu, 19 Apr 2012 22:17:45 +0000 (18:17 -0400)]
Fix bug when tracefile is not aligned. Now supports exotic architectures.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Fri, 20 Apr 2012 17:20:05 +0000 (13:20 -0400)]
Add support for float fields in a string output.
Signed-off-by: Matthew Khouzam <matthew.khouzam@ericsson.com>
Matthew Khouzam [Tue, 17 Apr 2012 19:33:21 +0000 (15:33 -0400)]
Add floating point support and Loglevel support
Alexandre Montplaisir [Mon, 30 Apr 2012 18:42:22 +0000 (14:42 -0400)]
tmf: Refresh the project once we're done loading a trace
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Tue, 1 May 2012 01:17:30 +0000 (21:17 -0400)]
tmf: Add the start time to the history file header
The previous method for getting the start time was bugged
anyway, it's much simpler to just save it in the header.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Mon, 30 Apr 2012 20:19:32 +0000 (16:19 -0400)]
lttng: Check for correct start/end times in the tests
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Sat, 28 Apr 2012 09:41:12 +0000 (05:41 -0400)]
lttng: Update queries' tests
Since we dropped statistics (in
2967f37), the quark numbers
changed, so the expected results of the getQuarks() tests
had to be updated.
While at it, updated the file to use the references in
Attributes.java, which was recently introduced.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Sat, 28 Apr 2012 09:11:57 +0000 (05:11 -0400)]
tmf: Fix wrong interval returns in the history backend
Since intervals are sorted (by end time) in their nodes,
the algorithm to pick the correct interval during a query
would first find the "starting point", where the intervals
can begin intersecting the target time.
Then, it would iterate over the remainder of the list, but only
comparing the *start* times, since we know that at this point,
the end times should all fit in.
Turns out this is not the case! It's possible for a node's end
time to be greater than its last intervals' end times. So if a
query was done in this "empty space", the node could wrongly
return an interval whose start time fits, but not its end time!
If every other condition is respected, now we additionally check
the end time too. This avoids a potential infinite loop during
range queries.
Also updated the tests accordingly.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Sat, 28 Apr 2012 08:15:01 +0000 (04:15 -0400)]
tmf: Be more lenient with GSS range queries
In the case where t2 (the end time of the range) is greater
than the end time of the trace/history, simply stop doing
queries once we reach the end, instead of throwing errors left
and right.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Sat, 28 Apr 2012 07:45:46 +0000 (03:45 -0400)]
lttng: Update history construction unit tests
Now that builder.run() returns immediately, we have to call
.close() to actually wait until the complete history is built
before we start the tests.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Sat, 28 Apr 2012 07:21:52 +0000 (03:21 -0400)]
lttng: Centralize all event and attribute names in dedicated classes
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Sat, 28 Apr 2012 06:15:29 +0000 (02:15 -0400)]
lttng: Move CTF dummy state provider to the test package
It's still kept around, since it's very useful for benchmarks.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 27 Apr 2012 21:48:15 +0000 (17:48 -0400)]
tmf: Make HistoryBuilder build its index asynchronously
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 27 Apr 2012 21:20:15 +0000 (17:20 -0400)]
tmf: Expose a method to retrieve an attribute's name
In the past, the only way to retrieve quarks was to pass its
exact name, so it implied we knew the name in the first place...
But with wildcard methods like getQuarks(), we won't necessarily
know the name of the quarks we receive. This new method will
allow the user to do a reverse lookup on the name.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 27 Apr 2012 18:53:51 +0000 (14:53 -0400)]
lttng: Explicitely return a StateHistorySystem in kernel traces
We build it ourselves, so we'll know what the type is.
Also fixed a missing assignement to the class field.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 27 Apr 2012 17:54:00 +0000 (13:54 -0400)]
lttng: Handle existing but invalid state index files
If a ".ht" file already exists we try to re-use it first. But if
this step fails, now we'll just scrap the existing file and build
a new one.
Other than the many possible error causes, this will be useful
for dealing with incompatible changes to the file format, for example.
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Alexandre Montplaisir [Fri, 27 Apr 2012 17:22:48 +0000 (13:22 -0400)]
lttng: Replace CTF- class prefix by Ctf-
Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
This page took 0.04101 seconds and 5 git commands to generate.