On Sun, Aug 14, 2016 at 9:53 PM, Jim Jagielski <j...@jagunet.com> wrote:
>
>> On Aug 12, 2016, at 9:58 AM, yla...@apache.org wrote:
>>
>> Modified: httpd/httpd/trunk/include/ap_mmn.h
>> URL:
>> http://svn.apache.org/viewvc/httpd/httpd/trunk/include/ap_mmn.h?rev=1756186&r1=1756185&r2=1756186&view=diff
>> ==============================================================================
>> --- httpd/httpd/trunk/include/ap_mmn.h (original)
>> +++ httpd/httpd/trunk/include/ap_mmn.h Fri Aug 12 13:58:10 2016
>> @@ -532,10 +532,11 @@
>> * dav_success_proppatch.
>> * 20160608.4 (2.5.0-dev) Add dav_acl_provider, dav_acl_provider_register
>> * dav_get_acl_providers.
>> - * 20160608.5 (2.5.0-dev) Add ap_proxy_check_backend(), and tmp_bb to
>> struct
>> - * proxy_conn_rec.
>> + * 20160608.5 (2.5.0-dev) Add ap_proxy_check_connection(), and tmp_bb to
>> + * struct proxy_conn_rec.
>> * 20160608.6 (2.5.0-dev) Add ap_scan_http_field_content, ap_scan_http_token
>> * and ap_get_http_token
>> + * 20160608.7 (2.5.0-dev) Add ap_check_pipeline().
>> */
>>
>> #define MODULE_MAGIC_COOKIE 0x41503235UL /* "AP25" */
>> @@ -543,7 +544,7 @@
>> #ifndef MODULE_MAGIC_NUMBER_MAJOR
>> #define MODULE_MAGIC_NUMBER_MAJOR 20160608
>> #endif
>> -#define MODULE_MAGIC_NUMBER_MINOR 6 /* 0...n */
>> +#define MODULE_MAGIC_NUMBER_MINOR 7 /* 0...n */
>>
>
> I know that there are no real ABI guarantees on trunk, but it
> still seems to me that this should have been 2 bumps, one of which
> said "replaced ap_proxy_check_backend with ap_proxy_check_connection"
> for some history.
I rather handled it as a follow up (2 months later..), but the second
bump makes sense, will do.
>
>> else {
>> - ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, s, APLOGNO(03408)
>> + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, server,
>> APLOGNO(03408)
>> "%s: reusable backend connection is not empty: "
>> - "forcibly closed", proxy_function);
>> + "forcibly closed", scheme);
>> }
>>
>
> Why the change in log level?
Reading a response before any request was sent looks quite severe to
me, a warning might better alert the admin about some (backend)
misbehaviour/response splitting, while a debug could stay unnoticed.
