gen_fsm to parse binary protocol frames

Andre Nathan <>
Wed Aug 4 00:09:41 CEST 2010


I'm dealing with a simple protocol whose frames consist of a fixed-size
header, which contains a length field which specifies the size of the
message following it.

I implemented a gen_fsm to handle this, using two states,
assemble_header and assemble_message. Each state responds to a 'data'
event whenever new data arrives from a socket (this data is sent from a
separate gen_server process). The assemble_header state responds to the
header_complete event switching to assemble_message, and
assemble_message responds to a message_complete event by starting
message processing.

My problem is the following. Since data comes from a socket, I don't
have control of how much data is read. It's possible, then, that a whole
frame (or more) has been already buffered when it's sent to the FSM.
This would cause the FSM to parse the header and switch to the
assemble_message state, but there will be no more messages to trigger
the 'data' event, since the whole frame has already been sent.

To solve this, I'm doing calling assemble_message({data, Rest}, State)
directly from the assemble_header state when this situation is detected.
That is, I'm forcing a state switch and triggering an event so that the
data I've already received can be processed. This feels kinda ugly...

So my question is, is the gen_fsm the preferred way to do this kind of
thing? If so, is there a cleaner solution to the problem which wouldn't
require the gen_server which feeds the data to the gen_fsm to know about
the protocol message format?

>From a little discussion on IRC, it seems that doing this processing in
the gen_server process would be a simpler solution. Is this what people
generally do?

Thanks in advance,

More information about the erlang-questions mailing list