> On Thu, Nov 08, 2001 at 10:32:45AM -0500, Bill Stoddard wrote:
> > This patch is a first rough implementation of a socket iol in APR (from 
> > discussions
with
> > Sander Striker a few months back).
> >
> > Public API Changes:
> > No new APIs were created.
> > apr_socket_create() has an extra parameter, a pointer to an 
> > apr_socket_iol_t.
> >
> > New Public Structures:
> > apr_socket_iol_t - contains pointers to callback functions.
> >
> > Usage:
> > If you do not want to redirect socket calls and just use existing APR 
> > calls, use
> > apr_socket_create() thusly:
> > apr_socket_create(&new_sock, APR_INET, SOCKSTREAM, NULL, r->pool);
>
> I thought we were trying to move away from function pointers because
> people thought they were slow.  Could/should we create a parallel API
> for this instead?  -- justin
>

If we created a parallel API, the all of the HTTP Server (for example) 
apr_socket calls
would have to be migrated to this new API; seems rather pointless. Adding a lot 
of
complexity to the API just to avoid one function call.

Yep, adding another function call/level of indirection chews a bit of CPU but 
it is not
that big a deal overall. And in the common case, we will -not- be making socket 
calls via
function pointers. In the common case, we just make a direct function call. 
Only when we
want to redirect the function calls will we use the function pointers.

Bill


Reply via email to