[boost] [MSM] Review

Subject: [boost] [MSM] Review
From: Kenny Riddile (kfriddile_at_[hidden])
Date: 2009-12-01 15:39:22

This is my first time reviewing a Boost submission, so I'll do my best :)

First of all, I'm not going to compare MSM to StateChart for two reasons:

1) I haven't used StateChart
2) I see them both as being individually useful, even if they can solve 
some of the same problems.  Is there not overlapping functionality in 
many STL containers?  Still, they're all well-suited to particular problems.

  - What is your evaluation of the design?

I have been using MSM on and off since January 2009 and have seen the 
design evolve as new features were added.  I first used it in a simple 
iPhone game which required a way to "update" an FSM's current state. 
This feature was generalized into the current ability to "visit" the 
current state.  The more I used the library and communicated with the 
author, it became clear that the design, and any changes to it, were 
thoroughly thought-out.  I especially like the clarity with which one 
defines an FSM (specifically the transition table).  The fact that a 
single instance of each state is default-constructed and bound to the 
FSM's lifetime is something that should be carefully considered when 
deciding if MSM fits with a particular use-case.  It was suggested in 
another review thread that a mechanism to optionally pass prototypical 
states in to a new FSM be added so that states didn't have to be 
default-constructible.  I agree with this suggestion.  Aside from that, 
the design has made it intuitive to accomplish everything I've tried to 
do with MSM.

  - What is your evaluation of the implementation?

I've only looked at the implementation a few times over the last year 
when trying to track down apparent bugs (which were all promptly fixed). 
  Aside from those instances, I try to avoid looking at things that hurt 
my brain.  I've seen the code in "C++ Template Metaprogramming" that MSM 
is based on, and consider it sound.  Is there a test suite for MSM?  Do 
these tests pass on all platforms officially supported by Boost?

  - What is your evaluation of the documentation?

The documentation was invaluable when I was first playing around with 
MSM.  The examples are concise and clearly demonstrate what they are 
meant to.  I would prefer that the documentation be broken into multiple 
pages similar to other released libraries.

  - What is your evaluation of the potential usefulness of the library?

I think MSM is useful because I have found uses for it.  My only 
significant experience with MSM has been in mobile game development, and 
I have used MSM in one shipped iPhone game.  It has worked very well in 
particular situations, such as simple AI for a single game entity.  The 
"staticness" of an MSM FSM is advantageous in this example, since 
constantly creating and destroying states can noticeably affect 
performance when a large number of entities are present in a simulation. 
  Conversely, I wouldn't use MSM to control the high-level application 
flow of a game.  I've found that the persisted states and 
default-constructibility requirement end up being hoops to jump through 
that just aren't worth it in this case where state transition speed 
isn't a concern.  Long compile times in the name of run-time performance 
have never bothered me either.  I can always upgrade my development 
machine, but a customer can't upgrade their iPhone.

  - Did you try to use the library? With what compiler?

Yes.  I've used MSM with MSVC9 and GCC compiling for the iPhone.

  - Did you have any problems?

I've found a bug or two when using MSM, but they were always fixed 
promptly and correctly.  There is of course the transition table size 
limitation imposed by mpl::vector and the compiler.

  - How much effort did you put into your evaluation? A glance? A quick 
reading? In-depth study?

I've used an earlier version of the library in a shipped product, and 
I've played with recent versions prior to the review.

  - Are you knowledgeable about the problem domain?

I understand how FSMs work and what they can be used for.

  - Do you think the library should be accepted as a Boost library?

Yes, I absolutely think MSM should be accepted into Boost, and will 
continue to use it regardless of the review results.  I would like to 
see the documentation restructured and the default-constructibility 
requirement for states removed via the means already described.  My vote 
for acceptance is not contingent on these requests.