Subject: Re: [boost] [msm] Version 2.0 available
From: Christophe Henry (christophe.j.henry_at_[hidden])
Date: 2009-11-12 10:18:50
Hi Andreas,
>:-) You won't get any flames from me. Yes, I don't agree with some of the
>points in MsmSession_Handout.pdf (e.g. 2.4, a user in his right mind would
>put Stopped, Open, Paused and Playing into an outer CDLoaded state and then
>only say if (state_downcast<const CDLoaded *> != 0)).
True, it is a possible solution. But 1) you don't always want to
refactor your state machine and 2) sometimes you can't.
For example, to take the example in the Statechart
documentation(http://www.boost.org/doc/libs/1_40_0/libs/statechart/doc/tutorial.html#StateQueries),
if ( ( state_downcast< const NumLockOff * >() != 0 ) &&
( state_downcast< const CapsLockOff * >() != 0 ) &&
( state_downcast< const ScrollLockOff * >() != 0 ) )
could have been replaced by
if(fsm.is_flag_active<IsOffState,Active::Flag_AND>()). And in this
case I don't see a working refactoring solution. In my experience, the
flag system has proven invaluable (to define, for example, several
success states in a test case defined as a state machine).
> However, although I haven't yet reviewed anything in-depth so far, I'm definitely impressed by the progress the library has made. I hope I'll find the time to contribute
> an in-depth review.
This would be great and your experience could prove invaluable. I'd be
very thankful for it.
> Out of curiosity, have you made any tests how many states compilers can manage without running out of memory? IIRC, a Boost.Statechart FSM with 128 states
> and several hundred transitions was the most I was able to fit into one TU.
Good point indeed. There are 2 limitations. First, Msm is based on
boost::mpl::vector for the transition table and on boost::fusion::set
to contain the states. Both have a default limit of 20, which can be
extended to 50. This is the theoretical limit. Of course, you can add
some vectorxxx.hpp headers (which I did). I tried up to 80 transitions
and 80 states in one machine.
The second limitation is the compiler. The test compiles on VC and
g++, eventhough you might need lots of RAM (1.4GB with VC9, 400MB with
g++) and 40-60s. The execution speed stays the same, thus proving the
O(1) double-dispatch.
I suppose that, while it must be possible to push this further,
Boost.Statechart will likely break the compiler later than Msm.
Regards,
Christophe Henry