Optimal pattern for parsing delimited (e.g. new line) network stream

Question

I am looking for optimal for maintenance/easy of programming and elegant patterns to parse delimited network streams, e.g. protocols like SMTP. Some of the concerns I've always had are:

• how to react when a recv(2) call returns 0 - the need to track state to determine if no data has arrived or no more data will arrive
• what if in one packet, we have multiple delimited fields arriving (e.g. HTTP request/response headers)

I've usually ended up building a buffer and parsing buffer until delimiter, handling that then separately, but it's always inelegant. Is there a Knuth-ian or GOF solution that is elegant?

By optimal/elegant I mean maintenance/programmability rather than hard-core use.

Answer 1

Asking for "optimal" is, in a word, nonsensical. As @MetaFight asked, what are you optimizing and what is elegant?

Optimal

• Readability
• Speed
• Robustness
• Lines of code

and a number of other things.

Design Patterns

• Finite State Machine - implement as a set of hierarchical state machines
• Lexer/Parser - this is a specific type of hierarchical state machines which is often supported by code generation libraries, such as yacc/lex, bison, etc.
• State Monad - Another implementation of state machine patterns.

Elegance

An open-ended question with lots of personal taste thrown in. Personally, I would start with an implementation in Scala, using a State Monad approach, with Either values that allow capturing error information to help reporting and recovery. Your mileage may vary (YMMV).