On 06 Oct 2015, at 11:13 AM, Yann Ylavic <[email protected]> wrote:

> By defining this function in util_filter
> (ap_filter_reinstate_conn()?), we also possibly could get rid of
> c->empty (which looks quite hacky), using/hidding an empty brigade per
> filter (in opaque data).

That starts multiplying the RAM really quickly, and starts getting expensive.

Ideally c->empty should be hidden inside an MPM-local context. Each MPM that 
then assign or not assign whatever it needs and this is hidden. If conn_rec had 
an opaque context for an MPM a lot of public stuff in the struct could probably 
be removed.

Regards,
Graham
—

Reply via email to