Stas Bekman wrote:
> Philippe M. Chiasson wrote:
>
>> Stas Bekman wrote:
>>
>>> Philippe M. Chiasson wrote:
>>> [...]
>>>
>>>> After a long promised investigation session, I believe I have pinned
>>>> it down. It seems to have been introduced by change r160562 when the
>>>> anon_cnt initialization code was moved around as a result. Didn't have
>>>> enough time tonight to dig down as to _why_ this is breaking though...
>>>>
>>>> Not entirely sure why this breaks specifically on Darwin and all, but
>>>> the following patch seems to make the problem dissapear. Does it fix
>>>> the problem you are seeing too ?
>>>>
>>>> Index: src/modules/perl/modperl_handler.c
>>>> ===================================================================
>>>> --- src/modules/perl/modperl_handler.c (revision 170139)
>>>> +++ src/modules/perl/modperl_handler.c (working copy)
>>>> @@ -89,9 +89,6 @@
>>>> modperl_modglobal_lookup(aTHX_ "ANONSUB");
>>>> MP_TRACE_h(MP_FUNC, "init $PL_modglobal{ANONSUB} = []");
>>>> MP_MODGLOBAL_STORE_HV(gkey);
>>>> -
>>>> - /* init the counter to 0 */
>>>> - modperl_global_anon_cnt_init(p);
>>>> }
>>>>
>>>> /* allocate and populate the anon handler sub-struct */
>>>> Index: src/modules/perl/mod_perl.c
>>>> ===================================================================
>>>> --- src/modules/perl/mod_perl.c (revision 170139)
>>>> +++ src/modules/perl/mod_perl.c (working copy)
>>>> @@ -532,6 +532,9 @@
>>>> modperl_global_init_server_rec(pconf, s);
>>>>
>>>> modperl_tls_create_request_rec(pconf);
>>>> +
>>>> + /* init the counter to 0 */
>>>> + modperl_global_anon_cnt_init(pconf);
>>>> }
>>>>
>>>> /*
>>>
>>> Philippe is talking about
>>> http://svn.apache.org/viewcvs.cgi?rev=160562&view=rev
>>>
>>> http://svn.apache.org/viewcvs.cgi/perl/modperl/trunk/src/modules/perl/mod_perl.c?rev=160562&view=diff&r1=160562&r2=160561&p1=perl/modperl/trunk/src/modules/perl/mod_perl.c&p2=/perl/modperl/trunk/src/modules/perl/mod_perl.c
>>>
>>
>> Yup, that one.
>>
>>> But it can't be reverted , since it was a fix for a bug in
>>> vhosts+anonsub+ithreads. I couldn't add a test reproducing the problem,
>>> as it wasn't suitable for the test suite (would have affected other
>>> tests). So it needs to stay that way and figured out what needs to be
>>> fixed elsewhere.
>>>
>>> Philippe, your patch can't be applied either, since it has exactly the
>>> same effect as reverting r160562, since you moved
>>> modperl_global_anon_cnt_init() to modperl_init_globals() which is called
>>> only for the top-level server
>>
>> Yes, but I am not reverting that change, keeping it as is, hash
>> initialization
>> and all, and only moving the modperl_global_anon_cnt_init() up so it's
>> called
>> only once. I don't think initializating that anon_cnt global more than
>> once is
>> a feature (and it registers it's destruction twice, causing the
>> segfault on OSX).
>
> No, you are reverting it. Follow the calls trace, you've just moved the
> original case (before my change) to a different place.
Unless I am missing something terribly obvious, I believe you are talking
about this chunk of code:
modperl_modglobal_key_t *gkey =
modperl_modglobal_lookup(aTHX_ "ANONSUB");
MP_TRACE_h(MP_FUNC, "init $PL_modglobal{ANONSUB} = []");
MP_MODGLOBAL_STORE_HV(gkey);
And that's what needs to be initialized in each server/vhost, correct?
And that still is. The only code I am suggesting to move around is the
modperl_global_anon_cnt_init() call
>> Are you saying that you need more than one distinct global anon_cnt in
>> some
>> vhost+anonsub+ithreads setup?
>
> Yes. Otherwise an anon-sub registered at the server startup in a vhost
> +Parent doesn't work.
But what needs to be initialized for each vhost isn't the anon-sub counter
integer, but rather $PL_modglobal{ANONSUB}, right ?
> First it was moved from running from each vhost to just the base server:
> http://marc.theaimsgroup.com/?l=apache-modperl-dev&m=110177550408046&w=2
> The whole thread is here (i think it was OSX too):
> http://marc.theaimsgroup.com/?l=apache-modperl-dev&m=110168848916838&w=2
>
> but then Dominique Quatravaux has encountered another bug, which you can
> see in this thread:
> http://marc.theaimsgroup.com/?t=111227919600001&r=1&w=2
--------------------------------------------------------------------------------
Philippe M. Chiasson m/gozer\@(apache|cpan|ectoplasm)\.org/ GPG KeyID : 88C3A5A5
http://gozer.ectoplasm.org/ F9BF E0C2 480E 7680 1AE5 3631 CB32 A107 88C3A5A5
signature.asc
Description: OpenPGP digital signature
