mido backend is slower than old python-midi backend

[craffel]

For reference, reading in example.mid takes on average 153 ms with mido backend and 131 ms with python-midi backend. Here are the prun outputs:

mido

         357073 function calls in 0.272 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.037    0.037    0.181    0.181 midifiles.py:299(_read_track)
        1    0.029    0.029    0.040    0.040 pretty_midi.py:218(_load_instruments)
    12613    0.028    0.000    0.078    0.000 midifiles.py:269(_read_message)
    25229    0.028    0.000    0.059    0.000 messages.py:371(__setattr__)
    12613    0.023    0.000    0.036    0.000 messages.py:433(build_message)
    26098    0.014    0.000    0.024    0.000 messages.py:135(check_time)
        1    0.011    0.011    0.269    0.269 pretty_midi.py:43(__init__)
    25226    0.011    0.000    0.011    0.000 {method 'update' of 'dict' objects}
    26975    0.011    0.000    0.011    0.000 midifiles.py:56(read_byte)
    13056    0.010    0.000    0.012    0.000 midifiles.py:88(read_list)
    53142    0.010    0.000    0.010    0.000 {isinstance}
    13485    0.009    0.000    0.015    0.000 midifiles.py:260(_read_variable_int)
    25228    0.007    0.000    0.007    0.000 {method 'format' of 'str' objects}
     7645    0.005    0.000    0.005    0.000 pretty_midi.py:241(__get_instrument)
        1    0.005    0.005    0.005    0.005 pretty_midi.py:149(_load_metadata)
        1    0.003    0.003    0.003    0.003 pretty_midi.py:114(_load_tempo_changes)
    13049    0.003    0.000    0.003    0.000 midifiles.py:65(peek_byte)
        1    0.003    0.003    0.272    0.272 <string>:1(<module>)
    12613    0.003    0.000    0.003    0.000 messages.py:124(get_spec)

python-midi

         314691 function calls in 0.220 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    13050    0.053    0.000    0.124    0.000 fileio.py:57(parse_midi_event)
        1    0.036    0.036    0.054    0.054 pretty_midi.py:213(_load_instruments)
    13049    0.025    0.000    0.031    0.000 events.py:34(__init__)
    12614    0.016    0.000    0.046    0.000 events.py:67(__init__)
    52223    0.010    0.000    0.010    0.000 {isinstance}
    13485    0.010    0.000    0.011    0.000 util.py:2(read_varlen)
        1    0.010    0.010    0.135    0.135 fileio.py:46(parse_track)
        1    0.009    0.009    0.016    0.016 pretty_midi.py:144(_load_metadata)
    39147    0.006    0.000    0.006    0.000 {setattr}
     7645    0.005    0.000    0.005    0.000 pretty_midi.py:236(__get_instrument)
    13049    0.004    0.000    0.004    0.000 {range}
    53592    0.004    0.000    0.004    0.000 {ord}
    24780    0.004    0.000    0.004    0.000 events.py:117(get_pitch)
        1    0.003    0.003    0.007    0.007 pretty_midi.py:109(_load_tempo_changes)
        1    0.003    0.003    0.220    0.220 <string>:1(<module>)
    25952    0.003    0.000    0.003    0.000 {method 'append' of 'list' objects}
     4956    0.003    0.000    0.003    0.000 containers.py:24(__init__)
        1    0.003    0.003    0.003    0.003 containers.py:42(make_ticks_abs)
    12614    0.002    0.000    0.002    0.000 events.py:215(is_event)
    13049    0.002    0.000    0.002    0.000 events.py:102(is_event)
     9912    0.002    0.000    0.002    0.000 events.py:123(get_velocity)

[craffel]

This likely could be helped by some optimizations to mido; beyond that, it would be great to switch to a native C++ backend.