httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeff Trawick <traw...@gmail.com>
Subject Re: Cannot start httpd v2.4.1 with mpm_build on AIX
Date Thu, 15 Mar 2012 21:32:04 GMT
On Thu, Mar 15, 2012 at 2:47 PM, Guenter Knauf <fuankg@apache.org> wrote:
> Hi Jeff,
> Am 07.03.2012 13:27, schrieb Jeff Trawick:
>
>> On Tue, Mar 6, 2012 at 8:00 PM, Guenter Knauf<fuankg@apache.org>  wrote:
>>>
>>> Hi Jeff,
>>> Am 26.02.2012 22:44, schrieb Jeff Trawick:
>>>
>>>> I guess the problem is that those MPM APIs are not decorated with
>>>> AP_DECLARE() so they don't get added to httpd.exp so they don't get
>>>> exported from httpd on AIX.
>>>>
>>>> Can you try the attached patch to mpm_common.h?
>>>
>>>
>>> it seems that this breaks the NetWare build now; I've not looked into it
>>> myself since I'm swamped currently with other stuff, but Norm just sent
>>> me
>>> this mail:
>>>
>>> Hi,
>>> Tried to send this to devAThttpd-apache but it got bounced; how peasant
>>> compared to other lists I've sent issues to. I'll watch for fixes.
>>>
>>> r1294936 causes extra export symbols.
>>> Hi,
>>> Re httpd-trunk:
>>> r1294936 added AP_DECLARE_xxx() to a number of func prototypes/vars in
>>> mpm_common.h. These are now found by the 'build/nw_exp_var.awk' script,
>>> which causes the NetWare apache2 link to fail for missing exports. As
>>> these functions are NOT common to NetWare or likely other non-Unix OS
>>> (all seem to be in mpm_unix or related mpm's such as event and worker),
>>> should these prototypes/vsrs now be in ap_mpm.h instead?
>>>
>>> Norm
>>>
>>
>> Thanks; I'll look into that.
>
> any news on that? Sorry, but unless you give a clear statement on how to
> proceed with these I'm unable to fix it self;

The short form is that I'm going to revert the trunk commit(s) today,
and when there's a good solution I'll put them back with the fix and
propose them for backport to 2.4.x.

I acknowledge wrowe's comment later in the thread, but wonder first
how practical it is to ensure that the declarations are visible only
if there is an implementation without going back to an equivalent of
the old AP_MPM_WANT_thisfunction, AP_MPM_WANT_thatfunction.

> if you say that these
> functions need to be extported by every platform then fine - we must then
> either provide dummies for those platforms which dont have them, or ifdef;
> or we need to move these into a unix-only header;
> its IMO not ok to suddenly decide to export these functions in order to fix
> one platform and break another one for that; and also I'm curious how
> Windows builds can cope with that, and wonder that nothing yet came from
> this end ...
> here are the sysmbols in question:
> LINK obj_release/Apache2.nlm
> ### mwldnlm Linker Error:
> #   Undefined symbol: ] in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_fatal_signal_child_setup in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_fatal_signal_setup in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_mpm_safe_kill in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_process_child_status in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_reclaim_child_processes in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_register_extra_mpm_process in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_relieve_child_processes in
> #   Export list
> ### mwldnlm Linker Error:
> #   Undefined symbol: ap_unregister_extra_mpm_process in
> #   Export list
>
> where the 1st one is a quirk from the awk script ...
>
> Gün.
>
>
>



-- 
Born in Roswell... married an alien...

Mime
View raw message