* Brosseau, Yannick ([email protected]) wrote: > On Mon, Feb 13, 2012 at 9:26 AM, Mathieu Desnoyers > <[email protected]> wrote: > > * Mathieu Desnoyers ([email protected]) wrote: > >> * Juha Niskanen ([email protected]) wrote: > >> > In absence of clock description in metadata, use a default clock > >> > source which increments once per nanosecond. > >> > > >> > Without this patch, the converter crashes when given old traces > >> > that are lacking clock description block. > >> > >> Merged (with tiny cosmetic fixes). Thanks ! > > > > FYI, with commit: > > > > commit 02dc4610b1dccefaf545ba77378857ee6188ea33 > > Author: Mathieu Desnoyers <[email protected]> > > Date: Mon Feb 13 09:03:55 2012 -0500 > > > > lib: open trace error handling fix > > > > When a trace from a trace collection path fails, consider this as a hard > > error. Print errors to stderr rather than stdout. > > > > The end goal is to never show an output if any of the trace specified > > (or trace within a trace collection) fails to open. > > > > Signed-off-by: Mathieu Desnoyers <[email protected]> > > > > Should you remove the traces that were added to the context before you > found an error?
Error handling of that sort was already missing, so I did not introduce any regression compared to what we had. I agree with you that the proper way to handle this would be to keep an array of opened traces, and free those upon error, to bring the context back to the state where it was prior to the call. Thanks, Mathieu -- Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com _______________________________________________ lttng-dev mailing list [email protected] http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
