[Log4cpp-devel] shutting down and reinit without leaking

[<log...@li...>]

Hello,

I would like to be able to initialise log4cpp from a properties file, to 
shut it down, to reinitialise it from another properties file etc.

In my opinion, there are two problems with this approach:

1. Leaking Appenders: The Appenders are not owned by the Categories when 
the log system is initialised from a properties file. Therefore 
Category::shutdown does neither close nor destroy the Appenders. I could 
close them with Appender::closeAll but not destroy them.

I could work around this problem by calling 
Category::getCurrentCategories() and then for each Category 
getAllAppenders() and then I destroy every Appender that is not owned by 
the Category. This works if the system is initialised exclusively from 
properties file as no Appender is owned by any Category. If, however, 
one Appender was owned by one Category, but not by another, then I would 
run the risk to destroy an Appender twice, once by my code and once by 
the destructor of the Category that owns it.

2. Leaking Categories: I could destroy all Categories by calling 
HierarchyMaintainer::deleteAllCategories(). One problem is that 
HierarchyMaintainer is an internal class, as the comments say, so I 
should not use it. Another problem is that deleteAllCategories() 
destroys the Categories in _categoryMap but does not clear the map. 
Thus, when the HierarchyMaintainer is destroyed automatically upon 
program exit the HierarchyMaintainer destructor attempts to destroy the 
already destroyed Categories and I get a segfault.

I could work around this problem by implementing a 
log4cpp::Log4cppCleanup class (exploiting the "friend class 
Log4cppCleanup" feature in HierarchyMaintainer) and setting a shutdown 
handler (HierarchyMaintainer::register_shutdown_handler) that executes 
HierarchyMaintainer::_categoryMap.clear().

So all in all I get the behaviour I want, but I have the feeling of 
forcing the lib.

I would have preferred

a) that deleteAllCategories cleared the _categoryMap. I think this is a bug.

b) that deleteAllCategories was exported via a call to Category.

c) that there was a straightforward way to destroy all Appenders that 
are not owned by the Categories.

Best regards,
Sorin