[Development] Tracing Qt

Philip Ashmore contact at philipashmore.com
Sat Aug 24 02:36:28 CEST 2013


On 24/08/13 00:24, Philip Ashmore wrote:
> On 23/08/13 12:59, Blasche Alexander wrote:
>> Hi Robin,
>>
>> I am not sure whether you are aware of http://qt.gitorious.org/qtplayground/qlogger
>>
>> It is pretty much what you describe. It was even close to a merge into qtcore already (only rejected due to feature freeze). It gives you a complete API including ways to integrate with qDebug/qWarning, runtime activation, little to no overhead when deactivated and there was reasonably broad agreement already. Also see:
>>
>> http://lists.qt-project.org/pipermail/development/2012-December/008886.html
>>
>> If you have spare cycles to actually make this happen I would propose you start with a 95% done system rather than new.
>>
>> --
>> Alex
> For a tracing system to be useful and stand a chance of being adopted by
> the libraries used by Qt as well as Qt itself, it shouldn't be part of Qt.
> 
> My humble attempt is available as part of v3c:
> 
> http://sourceforge.net/projects/v3c/
> 
> It has commits going back to 2011 but I deleted the earlier ones so it's
> even older.
> 
> You can control tracing at compile time, link time and runtime.
> You can cache trace output and only dump it if it's noteworthy, e.g. if
> something went wrong or a corner case is hit.
> 
> This uses a stringstream to cache the trace output.
> 
> This will be available in the next version - the one I'm using with my
> projects now (the version in SourceForge has the TraceCache macro but I
> never got around to using/testing it until recently).
> 
> Some examples:
> void fn(int a, int x)
> {
>     TraceWrap(__PRETTY_FUNCTION__ << " a: " << a);
>     trace << "the value of x is " << x << '.' << endl;
> }
> 
> It has a pluggable architecture, but the main application decides if
> trace output goes to cout, cerr or a file or nowhere.
> 
> It doesn't have module-level trace control but that could be added.
> 
I forgot to mention I've also got v3c-qt:
http://sourceforge.net/projects/v3c-qt/
There's an "examples" tar-ball there that shows how to integrate tracing
into some demo Qt Applications.

My projects use the automake "make && make check && make distcheck &&
make install" commands, as well as "make debian/ make ubuntu".
For iterative development I really like "make -j9".

Oh and you can configure build targets, the defaults are "debug" and
"release".

The point of proposing a tracing system/library that's not part of Qt is
that it can be adopted by non-Qt libraries which Qt might eventually use.

> Regards,
> Philip Ashmore
>>
>> ________________________________________
>> From: development-bounces+alexander.blasche=digia.com at qt-project.org [development-bounces+alexander.blasche=digia.com at qt-project.org] on behalf of Robin Burchell [robin+qt at viroteck.net]
>> Sent: Friday, August 23, 2013 13:20
>> To: development at qt-project.org
>> Cc: Poenitz Andre
>> Subject: [Development] Tracing Qt
>>
>> Hi,
>>
>> (apologies in advance for the wall of text)
>>
>> It's inevitable when working with Qt that sooner or later, you
>> probably have a desire to work on tracing Qt - or at least part of it.
>> Qt Creator's QML profiler is one such very prominent example of an
>> area where this may be useful, but there are many others: usually
>> sprinkled around Qt hidden behind various defines, environment
>> variables, and other hard-to-find places. This makes digging into a
>> problem somewhat hard, unless you are (or can be handheld by) a Qt
>> expert when you start digging.
>>
>> It's also a bit of a problem in that each of these methods uses
>> different ways to communicate information, meaning it is difficult to
>> write tooling to inspect this data and visualise it. We also miss out
>> on tracing in a number of areas that *could* potentially be valuable,
>> such as touch event processing, the event loop, the sky's the limit.
>>
>> Thanks to Carsten and Andre for their input in hashing the details of
>> this proposal out.
>>
>> Proposal
>> ========
>>
>> Let's look at unifying tracing inside Qt. Let's add code to Qt itself
>> to enable easy introspection of what exactly is going on.
>>
>> When looking at tracing, it can be boiled down into two levels - an
>> area (which may be a feature being inspected, or a subsystem - for
>> instance, QML performance, vsync timestamps, ...) containing events,
>> so from this, we can gather that we need a way to categorise logging
>> by area, and the capability to switch logging on by area.
>>
>> An area should be associated with a simple numeric identifier to avoid
>> expensive code when manipulating events, but there will be some form
>> of name => id mapping (for the environment variable below, for
>> instance).
>>
>> Suggested area-based API:
>> enum TraceArea {
>>     TraceAreaDrawing (drawing event, frame event)
>>     ... other areas as required
>>     TraceAreaUser = 1000 // held for possible future expansion into
>> allowing dynamic area registration
>> }
>>
>> QT_ENABLE_TRACE=inputevents,frameevents,magicalmonkeyevents
>> rationale: providing system-wide defaults enables tracing of
>> system-wide behaviour is very useful. Being able to inspect exactly
>> what an entire set of Qt-based applications are doing while a user is
>> interacting with it would be a useful thing.
>>
>> enable_trace(TraceArea area)
>> disable_trace(TraceArea area)
>> rationale: a developer may want to turn tracing of an individual area
>> on/off at runtime, as they start something they are particularly
>> interested in tracing. This has some inspiration in
>> CALLGRIND_START_INSTRUMENTATION/CALLGRIND_STOP_INSTRUMENTATION also.
>>
>> Along with this user-facing API, we also need some mechanism to enable
>> different applications of the tracing data (for instance, the QML
>> profiler from Creator, Android's systrace
>> https://developer.android.com/tools/help/systrace.html, plain console
>> logging, etc). Multiple tracers should be able to handle a single
>> area. Tracers would live in seperate plugins.
>>
>> Example tracer API:
>> class QTracer { virtual void logStart() = 0; virtual void logEnd() = 0; }
>>
>> QTracer subclasses would register their interest in a particular event
>> type on construction:
>>
>> struct MyCoolConsoleTracer : public QTracer {
>>     MyCoolConsoleTracer() : QTracer() {
>>         q_register_tracer_for_area(this, TraceAreaDrawing); // we are
>> interested in this particular event type
>>     }
>>
>>     virtual void logStart() {
>>         qDebug() << "Trace event for drawing, start: " <<
>> QDateTime::currentMSecsSinceEpoch()
>>     }
>>     virtual void logEnd() {
>>         qDebug() << "Trace event for drawing, end: " <<
>> QDateTime::currentMSecsSinceEpoch()
>>     }
>> };
>>
>> The invocation of tracer subclasses is controlled by a simple RAII class:
>>
>> struct QTraceEvent {
>>     QTraceEvent(TraceArea area); // invokes logStart on all interested tracers
>>     ~QTraceEvent(); // invokes logEnd on all interested tracers
>> }
>>
>> Example use being something like:
>>
>> void draw() {
>>     QTraceEvent e(TraceAreaDrawing);
>>     // do painting
>>     paintOnScreen();
>> }
>>
>> Thoughts, flames, cookies?
>>
>> BR,
>> Robin
>> (with jolla hat on)
>> _______________________________________________
>> Development mailing list
>> Development at qt-project.org
>> http://lists.qt-project.org/mailman/listinfo/development
>> _______________________________________________
>> Development mailing list
>> Development at qt-project.org
>> http://lists.qt-project.org/mailman/listinfo/development
>>
> _______________________________________________
> Development mailing list
> Development at qt-project.org
> http://lists.qt-project.org/mailman/listinfo/development
> 



More information about the Development mailing list