Add new hist command

[namhyung]

I'm thinking of histogram functionality in the hist sub-command or maybe in a (Python) script. It can show various histogram based on the trace data. This is just to think out loud and get some feedback.

For example, it would show a histogram of execution time of every function by default.

$ uftrace record ...

$ uftrace hist
# execution time  :      count | graph
    0 ns ~  10 ns :          2 |
   10 ns ~ 100 ns :        312 | ******
  100 ns ~   1 us :        249 | *****
    1 us ~  10 us :        158 | ***
   10 us ~ 100 us :        116 | **
  100 us ~   1 ms :          0 |
    1 ms ~  10 ms :         53 | *
   ...

It could show a histogram of execution time of single function if it's given (as an argument)

$ uftrace hist  myfunc
# execution time  :      count | graph
    0 ns ~  10 ns :          0 |
   10 ns ~ 100 ns :         13 | *******
  100 ns ~   1 us :          8 | ****
    1 us ~  10 us :          3 | *
   10 us ~ 100 us :         12 | ******
  100 us ~   1 ms :          0 |
    1 ms ~  10 ms :          5 | **
   ...

And it should be able to show histogram of value from argument or return value (if recorded) like

$ uftrace hist --val arg1 myfunc2
# value range    :        count | graph
     0   ~   1   :            0 |
     1   ~  10   :           13 | *******
    10   ~ 100   :            8 | ****
   100   ~   1 K :            3 | *
     1 K ~  10 K :           12 | ******
    10 K ~ 100 K :            0 |
   100 K ~   1 M :            5 | **
   ...

I think we need an option to specify the range of the values, step or type. It would have a histogram of string values for top-5 most frequent values. For example, it could trace open() function and show which file it accesses mostly (recorded in arg1).

$ uftrace hist --value arg1 --hist-policy range=no,type=string,entries=5,graph-first  open
#       graph |      count :   values
         **** |         21 :   "/dev/null"
          *** |         18 :   "/tmp/qeWor8"
           ** |         10 :   "/proc/self/maps"
            * |          7 :   "/home/user/myfile"
            * |          6 :   "/usr/bin/uftrace"

Thoughts?

[paranlee]

@namhyung Is there an option to adjust the execution time or value range of the histogram to be constant
or exponentially increasing?

$ uftrace record ...

$ uftrace hist
# execution time  :      count | graph
    0 ns ~  10 ns :          2 |
   10 ns ~ 100 ns :        312 | ******
  100 ns ~   1 us :        249 | *****
    1 us ~  10 us :        158 | ***
   10 us ~ 100 us :        116 | **
  100 us ~   1 ms :          0 |
    1 ms ~  10 ms :         53 | *
   ...

$ uftrace hist --val arg1 myfunc2
# value range    :        count | graph
     0   ~   1   :            0 |
     1   ~  10   :           13 | *******
    10   ~ 100   :            8 | ****
   100   ~   1 K :            3 | *
     1 K ~  10 K :           12 | ******
    10 K ~ 100 K :            0 |
   100 K ~   1 M :            5 | **
   ...

[namhyung]

Yes, it should have the fine controls.