From: Paul Lutus <nospam@...>
Date: 2006-11-03T07:05:15+09:00
Subject: Re: net/http and rexml

Louis J Scoras wrote:

> On 11/2/06, Paul Lutus <nospam@nosite.zzz> wrote:
> 
>> You mean, line by line? The socket class you are describing doesn't know
>> about lines, it knows about blocks.
> 
> Nope.  Not line by line.  All the parser should need is a token, and I
> should only have to read as much data as I need to complete one, so
> blocks would be fine.

And you could set things up to read more data when your block-oriented input
stream is depleted, easy to arrange. This will provide the appearance of a
local stream, a common arrangement in socket reading algorithms.

>> So try this: read a block, split it into lines, do your processing. If
>> you do this, you will discover some blocks end in the middle of a line.
>> Then you will say, "Gee, maybe I should read the whole thing, then do the
>> processing."
> 
> No, I wouldn't say that ;)  I'd just read enough segments into a
> buffer until I could complete the next token.

s/segments/blocks/

> 
>>
>> Yep. Read the entire thing. Then process the result.
>>
> 
> Why?  I should be able to start processing simultaneously.

Block by block, yes. The block reading back end can be made to appear to be
a stream locally, but there are excellent reasons to read blocks at the
network-protocol level, and sometimes the bigger the better.

> That's 
> what the stream paradigm was developed for.

Yes. You can always turn a block into a stream locally. And no, you don't
have to read the entire thing, I just prefer it that way. A personal
preference, nothing more, doubtless springing from my unreliable Internet
access.

-- 
Paul Lutus
http://www.arachnoid.com