vopl commented on PR #271: URL: https://github.com/apache/logging-log4cxx/pull/271#issuecomment-1759330480
> > This looks good to me; if it fixes the "my program crashes because I'm logging after calling exit()" problem with no bad side > > effects it sounds good to me and could be something that we enable by default. Any thoughts? > > @vopl could you tell us what your need is? > > "logging after calling exit()" will currently produce output if the user function is added to the on-exit list after Log4cxx statics (so it is called before Log4cxx destructors are called). AFAIK the on-exit list order for module-level statics is a determined by the module placement order a link time (so it is not easily controlled). No output is produced after the APRInitialiser destructor is called. I do not think subsequent logging attempts will cause a crash as the Logger objects no longer belong to a LoggerRepository and Logger methods take an early exit. > I have an application whose functionality is placed in a set of .so . This functionality uses static/global objects for storing state for its internal needs, and also uses threads for code execution. I cannot ensure that the functionality stops during the program shutdown strictly before log4cxx collapses, since this functionality is extremely massive. In such conditions, I need to provide the mentioned functionals with the ability to use the logger during the destruction of the functionals themselves, which happens after exiting 'main' or with the explicit activation of 'exit', at the stage of destruction of objects with a static lifetime (while the state of the logger has already been destroyed by the same mechanism). Attempts to use the logger (ahh.. By 'logger' I don't mean the Logger object, but the entire log4cxx functionality) in such conditions lead to a crash, for example, according to this scenario: Logger::getLogger(name) -> ... -> APRInitializer::getInstance -> using the destroyed object. > Failing to flush buffers is the one problem with this PR that comes to mind, and this is appender/configuration dependent. Yes. That's why I don't suggest enabling the INFINITE LIFE TIME option by default. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: notifications-unsubscr...@logging.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
