Re: [boost] Re: Re: IOStreams formal review start

From: Carlo Wood (carlo_at_[hidden])
Date: 2004-08-31 13:10:13

On Tue, Aug 31, 2004 at 09:55:19AM -0600, Jonathan Turkanis wrote:
> >    file > bzip2 (binary) > text > (process) > bzip2 (binary) > file
> 
> Line-ending conversions can be done by sticking a newline filter in between the
> binary and text filters. When converting_stream is up and running, code
> conversion will be inserted at the appropriate place in a filter chain
> consisting of mixed narrow- and wide-character components.

Hiya, me again :)

I suppose I should really just download the code and start playing
with it in order to be able to review it - but I have no time for
that.  So, don't take my questions as formal critics please; I am
just being interested as a possible (future) user (and possibly
even contributor).

My concern, when I see those filter chains, remains performance
as a result of unnecessary copying of data.

Can you tell me; would the average filter copy the data
(from one place in memory to another)?  I'd suppose so, because
it is unlikely that an arbitrary filter can be trusted to use
the streambuf as a 'work space'.

And if so, does this mean that if I put ten filters in a chain
that then the data is copied ten times?  Please enlighten me :).

Thanks for your time in advance!

-- 
Carlo Wood <carlo_at_[hidden]>