I am chasing an issue with a proprietary module on an operating system
with some special function pointer handling (AIX)

It occurred to me just now that a module should not register a pconf
cleanup that uses its own functions, because:

* dso_cleanup is registered against pconf
* cleanups are LIFO

This is similar to an issue raised by Garrett Rooney on dev@apr
"Problems with DSOs and Pools".

Does this limitation make sense?  I could only find one obvious
(variable named pconf) offender, mod_ssl, so I am questioning if this
can really be so unsafe.

The obvious solutions, once you know there's a problem with your
casually allocated cleanup, are:

* create a subpool for the cleanup, so you can be sure you won't be unloaded.
* use a "pre_cleanup" (I had no idea this was a thing)

Does this make sense as a limitation, or is it a red herring? Are
there any mystery mod_ssl exit reload issues floating around?

-- 
Eric Covener
[email protected]

Reply via email to