httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <>
Subject Re: I/O filter API
Date Tue, 26 Aug 1997 08:10:59 GMT
The only problem with layering, regardless of what we use ... is keeping
the memory to memory copies to a minimum.  It doesn't bother me a whole
lot if the dynamic/filtered paths involve copies (this is almost by
definition) ... but I want to be sure the static path remains copyless. 
This means we will probably want a writev() equivalent in the API since
writev() is instrumental in pushing copies into the kernel (where they
typically have to happen anyhow). 

I'm still unconvinced of the unportability of SFIO... I think it degrades
-- doesn't implement as many features if the necessary underlying stuff
isn't there.  But I don't know for sure. 


On Tue, 26 Aug 1997, Alexei Kosut wrote:

> On Tue, 26 Aug 1997, Ben Laurie wrote:
> > I don't see why sfio should be a problem - we had it running all over
> > the place with RST's experimental version.
> That's not my recollection. Only about five people ever ran apache-XX,
> and I defenitely recall the porting issues involved with the sfio package
> included being non-trivial. I have a memory of a long configuration file
> that needed to be set up for each OS, although there were defaults for a
> couple. Does anyone have a link where I can download a recent copy of
> sfio and look at it?
> I do know that RST switched to bstdio (a BSD-based stdio replacement that
> Chris Provenzano used for his pthreads implementation, I believe), for
> whatever reason, in later versions of apache-XX. I don't remember why
> exactly he ceased using sfio, but there very well may have been a reason
> for it.
> While I agree that APIs are useful, this is really less of an interface
> issue, and more of a "which code should we use" issue, because the API Ed
> and I have been talking about can be implemented just as well on top of
> sfio or any other full-featured stdio replacement as it can be written
> into the BUFF code.
> -- Alexei Kosut <>

View raw message