httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Greg Stein <>
Date Thu, 24 Aug 2000 17:17:14 GMT
On Thu, Aug 24, 2000 at 07:26:22AM -0700, wrote:
> > > Grrrr....  No this doesn't make sense.  IOLs are just watered down
> > > buckets.
> > 
> > IOLs have an interface symantically similar to sockets with the ability
> > to read from or write to any data sink/store. The support setting the
> > read/write interface blocking or non-blocking.  Very simple
> > conceptually. Now, we have to create a 'bucket' with a funky looking
> > interface for each data source and data sink. You pass in a variable to
> > tell the i/o hidden in the bucket whether it needs to block or not.
> > It's crufty at best.  Buckets and filters makese sense to me for
> > filtering content, but they make no sense as a replacement for IOLs. How
> > are you going to implement something like ap_send_fb() using buckets and
> > filters that is easy to see and understand?  We are trying to use
> > filters to fix all of Apache's problems.
> You were at the meeting when this was discussed.  There is no need for
> IOLs.  They don't provide anything to us.  As far as implementing
> ap_send_fb, that is why we have the block argument to read.  This was all
> discussed and designed properly.

Let's not rewrite history.

We designed filtering at that meeting. You removed the IOLs.

I do not recall any discussion of the IOLs other than "this will let us
remove IOLs" "we'll cross that bridge later".

Well... it seems that it was crossed, but caught Bill by surprise. He has
vetoed it which means some meeting of the minds must occur, or the patch
will need to be backed out.

Personally, I'm not familiar enough with the CGI stuff (or the other items
modified by this patch) to know why we use a BUFF in there. Passing a pipe
bucket into the content chain does seem like the right thing to do.


Greg Stein,

View raw message