schedule event needs to be fixed

[honggyukim]

I think current linux:schedule event is not mature and it has the following problems as of now.

• It doesn't work well with record time time-filter
  • So when distributing the recorded data, always have to mention to use time-filter for analysis.
  • otherwise, the output screen is covered by too many noisy scheduling events as mentioned in "inverted time" #699 (comment).
• --no-event simply hides this scheduling events, but it also disables task-name event as well.
  • The only way to work around this is to use --no-event -E linux:task-name, but it doesn't look nice.

In most cases, I'm only interested in task-name and not interested in scheduling event.
I would like to suggest the following choices.

1. --no-sched option to disable only schedule event.
2. disable schedule event by default and add --sched option to turn it on.
3. schedule event doesn't work with time-filter #658 (comment) can be a solution

[honggyukim]

As we discussed today, I think we better provide a way to apply default options from a pre-defined opt file that is inside of uftrace.data.

We can generate the opt file inside uftrace.data to apply the time-filter used by record command. This can be the simplest fix as wrote in "3" above.

[honggyukim]

Otherwise, It would be also good to always use -E linux:task-name even if --no-event is used.

[honggyukim]

Otherwise, It would be also good to always use -E linux:task-name even if --no-event is used.

@namhyung What about applying this as well? Is there any benefit of not recording linux:task-name?

[namhyung]

Well, it doesn't honor the meaning of NO events and maybe sometimes users want to remove overhead of perf events at all.