[boost] Re: [prereview request][fsm] - No code generator required => Manual coding?

From: Andreas Huber (ah2003_at_[hidden])
Date: 2004-05-26 08:16:18

Kwee Heong Tan <tan.k.h <at> juno.com> writes:

> I would like to get some clarification, so please correct me if I'm wrong.
> 
> The first feature listed in the boost::fsm documentation is
> "straightforward transformation from UML state chart to executable C++
> code".
> 
> Since straightforward is in the eye of the beholder, I think the following
> would convey more information :
> "Direct mapping of UML state elements to boost::fsm elements for simple
> but manual transformation from UML state chart to executable C++ code."

Agreed, that is probably a better description. I'll change that sentence.

> I am confused why having no code generator is considered a feature.

Code generation is useful, especially when both of the following conditions 
hold:
- Humans don't need to read and/or modify the generated code.
- There is no library that allows you solve the problem with similar ease as 
the code generator approach.

If, for a given problem, both of the above conditions do not hold then I 
believe you're almost always better off with a library. We could now engage in 
a discussion about the pros and cons, but I believe this is rather off-topic 
for the boost list (recently there was a discussion about code generation on 
c.l.cpp.m and I have expressed my opinion there).

The most important argument is that with boost::fsm you have a choice:
- You can easily manually implement FSMs
- You can write a code generator that generates boost::fsm client code

For FSM frameworks that are geared towards a generator front-end you almost 
never have that choice as the generated code is much harder to read and 
understand (tiny bits of user code are embedded in large chuncks of generated 
code).

> It
> would seem the reverse is a feature, because a code generator would
> save coding time and eliminate human translation errors.

Sorry, but that sounds like the usual marketing stuff you hear from companies 
selling code generators. I did work with FSM code generators and initially I 
thought that they are a good idea. However, with time and experience I got 
more and more convinced that in the long run they typically introduce more 
complexity than they purport to remove, especially if there is a library that 
allows you to very directly and easily convert UML to code.

Regards,

Andreas