Re: [cpp-netlib-devel] Pion's HTTPParser

[Michael D. <mi...@mi...>]

On Dec 11, 2007, at 12:26 AM, Dean Michael Berris wrote:

> Hi Mike!
>
> On Dec 11, 2007 10:40 AM, Michael Dickey <mi...@mi...> wrote:
>> Dean and I have been talking on the boost development list about ways
>> to integrate Pion code into the current cpp-netlib structure.  Seemed
>> like a good idea to move that discussion over to this list instead.
>
> I agree, though I'd also like to have other people in the Boost
> mailing list involved somehow. I like the input that we get there, and
> I'm thankful that the discussion has brought out interesting insights
> from people in the list. :)

I'm new to the Boost list, so I just wasn't sure about overwhelming it  
messages on this topic -- in particular as we get more into  
implementation details.  I can re-post my message there though if you  
think others would be interested in chiming in.

...
>> In any case, if we could figure out how to merge this functionality
>> over I'd be happy to switch over to using cpp-netlib in Pion instead
>> of our own message-like classes.
>>
>
> Sweet! That sounds like good motivation for me to actually write more
> documentation, try porting your better parser, and see how else we can
> improve on decoupling the logging/networking/etc. from the actual
> parsing.

Great, feel free to use as much or little of it as you like!  =)

And of course, let me know if you have any questions..

> Just a question: why shouldn't we be using Boost.Spirit for this  
> sort of thing?

The main reason I didn't use it is that I'm not familiar with  
Boost.Spirit, and the parsing code was original based on Chris'  
http_server examples in ASIO (which do not use Spirit).

I got the impression from reading through the Spirit pages that it  
does incur a performance penalty, and I wanted to make everything in  
Pion as fast as possible.  Also, I wasn't clear if Spirit could handle  
the incremental processing needs.  Namely, the parser needs to be able  
to consume as few as 1 byte at a time to make it asynchronous, low- 
memory and HTTP/1.1 compatible.

Take care,
-Mike