Hi Josef,
thanks for your explanation.
On Thu, 21 Jul 2016 12:16:28 +0200 Josef Söntgen <josef.soentgen@...1...> wrote:
Hello Johannes,
- Johannes Schlatow <schlatow@...238...> [2016-07-20 16:49:12 +0200]:
The documentation suggests that the trace buffer is assigned to a thread. On the other hand, the trace subject is associated with a session label. What I conclude from this is that a thread be identified by multiple trace subjects but can only be assigned a single trace policy and trace buffer, which makes the interface quite confusing.
A Trace::Subject *may* contain a attached Trace::Buffer and at some point *did* contain a Trace::Source. To enable tracing the Subject must have a valid Trace::Policy that in return is used by the Source to fill the Trace::Buffer. A Subject can only be part of one Session. Subjects that belong to another Session have the state FOREIGN (see [1]).
[1] repos/base/include/base/trace/types.h
I also had a glance at the implementation and noticed that the Trace::Subject has its own buffer, which is forwarded to the Trace::Source. If I understand this correctly, the Trace::Source is coupled to a thread. I therefore assume that I can only trace one trace subject (associated with the same thread) at a time. Is this correct?
You can trace as many subjects as your resources permit within a single Trace::Session. The Subject outlives the Source (that is why it forwards the Buffer to the Source) so one can process the content even after the Source has vanished. As you have noticed a Trace::Source is coupled to Thread and so is a Subject to its Source. That is a Subject does not get reused. After tracing has been enabled for a given Subject it will “live” and consum memory as long as it is not freed by the Session, even when its Source has vanished. New Subjects will be created in a lazy fashion whenever Trace::Session::subjects() is called (see [2]) and there are new Sources. Untraced Subjects will vanishes when its Source vanishes.
[2] repos/base/src/core/trace_session_component.cc
So from a user perspective or rather from the point of view of a Trace-monitor (the component managing the Trace::Session) all you care about is a Trace::Subject. To fill the Trace::Buffer, i.e. create entries, you care about the Source which is a Thread. That is why the methods for creating entries is part of the Thread API.
Okay. I think I got confused by the session label. For a moment I assumed these corresponded to the service provided by the traced component but it's actually the label of the component's CPU session, which makes much more sense now ;-)
Anyway, I have a much better understanding of the implementation now.
Cheers Johannes