[tracecompass-dev] Towards automatic time synchronization

[Matthew Khouzam <matthew.khouzam@xxxxxxxxxxxx>]

Here is a problem that is pretty interesting:

You have two traces on the same host, NTP happens between the start of
both traces, now, the clock sources are offset by 500ms. Let's say it's
an LTTng Usermode and Kernel Trace. The problem is that you can appear
enter a callstack (user code) in the middle of a syscall or even worse,
an interrupt!!!!

To solve this problem, the host names must be checked and then
synchronized. From a code perspective it's about 20 lines. The more
complicated issue is to do this in a clean way that makes sense for the
user.

I propose using the time synchronization facilities already there.
Quite simply, the synchronize trace handler would not only perform the
current synchronization, but if applicable will also offset the traces
with the same hosts. It would be transparent for the user and make sense
from a trace viewer point of view, as each offsetted trace is stored
next to the main trace so there is a cue to the user that something changed.

Thoughts?