is this possible?

iceman jegan... at
Fri Aug 31 05:38:57 UTC 2007

sorry about the vague description..

basically i am working with click modular routers.i am trying to
separate the data plane and the control plane..

i will try to represent the control plane with a state machine..the
the packets[like those of sk_buffer]will have to be to parsed and and
the the control action[like binding update and router solicitation
etc]should be converted to tokens and fed into the state
the state machine depening upon the state an action should take place
with state transition..

what i am worried about is that will it be possible for me to use the
parser associated with  that of ragel to parse the incoming packet[bit
patterns] at real time?

On Aug 31, 6:58 am, Steve Horne <stephenhorne... at> wrote:
> That's a very vague requirement!
> Let's see...
> > i would like to parse packets[bit patterns] and get out token which
> > represent the function..
> Ragel generates a state machine, which it uses to extract tokens from
> input. In principle, the state machine takes as input a stream of
> characters, though in practice those 'characters' can be anything you
> can represent using integers (the precise type can be specified to get
> more bits per character).
> It isn't clear from your description whether the state machine you are
> talking about would be the one Ragel generates for you, or some other
> one, but either case is fine. You basically supply your own 'template'
> code for the Ragel state machine, so you can get your input character
> stream any way you want and use your output tokens in any way you
> want.
> You could even use the output from one Ragel machine as the input for
> another one.
> The only doubt I have is your mention of 'bit patterns'. As you can
> handle input any way you want, there is in principle no difficulty,
> but to the best of my knowledge you cannot use the current state to
> determine how to decode the next input. You could not take variable-
> bit-width characters from the input, for instance.
> You certainly can handle something like UTF-8 decoding, though - you
> just have a machine that takes 8-bit 'characters' as input and outputs
> unicode codepoints as its result (possibly sending them to another
> Ragel machine that tokenises the unicode character sequence).
> Actually, there's no reason why you can't convert UTF-8 to unicode
> codepoints in a single Ragel machine, but it's probably easier to
> cascade two machines to do the job.
> Equally, you could process a Huffman-encoded input if you really want
> to. You couldn't handle the input in multi-bit chunks, but you could
> handle it one bit at a time. The only condition is that the particular
> encoding would have to be fixed. This would work because Huffman
> encoding is a variable-width encoding, designed in such a way that you
> can find the end of each character using a finite state machine.
> The Ragel manual is very good, by the way. Just read the section on
> the 'interface to the host program' and I think all your questions
> will be answered.
> However, just on the off chance, I'll also recommend that you take a
> look at the SMC project...
> I've not actually used it myself, whereas I use Ragel a lot, but
> depending on your state machine requirements it could be more
> appropriate.
> Ragel has a far more powerful model for specifying finite state
> machines in that it supports regular grammar handling features (like
> regular expressions, but better) as well as explicit specification of
> transitions, and the two forms can be mixed as needed. It also allows
> backtracking scanners (like flex - the scanner generator, not the
> Adobe web framework thing).
> Where SMC appears to win out is if you already have a pure state-
> transition model (e.g. based on a UML state diagram) and you want to
> ha
ndle events that are like class methods, with parameters.

More information about the ragel-users mailing list