See issue #18.

It's important to know if the backend server doesn't respond according to
the pattern "HTTP/1.1 ###*" (eg. "HTTP/1.0 503 ..." if there is a problem
with the Exchange server name in the request). So raised the log level for
the corresponding message from info to error.

Mixing request_rec pool and backend connection bucket allocator in
apr_brigade_create() leads to a segmentation fault if the brigade is
accessed after the request record was destroyed.

To re-generate the configure script, run autoreconf.

APR's implementation of format strings does not support %llu on 64 bit systems.
But apparently using %lu works equally well without any need for type casts.

Also APR's implementation does not support %p properly. Up to now we resorted
to using %x and casting the pointer to an unsigned int. On 64 bit systems would
only output 32 bit of the pointer address. This is fixed by using %pp
(apparently the APR replacement for %p).

Fixes #2

Regarding the 64 bit limitations see also ticket #8.

The previous commit fixed the issue only for RPC_OUT_DATA requests.

Fixes #6

The added code is based on some almost identical code in mod_proxy_wstunnel.c
from Apache httpd upstream SVN branch 'trunk'.

Fixes #6

This reverts commit 158280ab.
Apparently this commit causes segfaults on systems that do not support the used
format string characters for size_t et al.

Fixes #2

Fixes #1

Based on Sophos commit bd655315.