On Fri, Apr 17, 2020 at 9:59 PM Ruediger Pluem
>
> On 4/17/20 6:47 PM, ylavic
> >
> > - if (r->status != HTTP_OK) {
> > + if (!ap_check_request_header(r)) {
> > access_status = r->status;
> > goto die_early;
>
> This means we now die_early where we did not before and may not reply with
> custom error pages configured for the vhost.
See my reply for r1876664, die_early will also call ap_die() finally
so custom error pages are still in the place.
There are two semantic changes compared to before r1876664 (the first
commit) though, first there are cases where we previously called
ap_send_error_response() only while now we ap_die() always (almost).
It did not make sense to me to call ap_die() on read_request_line()
failure but not later, if ErrorDocument is relevant with request line
errors I think it is for later ones.
Second, this commit (r1876674) checks the Expect header in
ap_check_request_header(), that is before post_read_request hooks,
while we cheched (and failed) after that before. I think this change
makes sense too.
Regards,
Yann.
