Subject: Re: [boost] [log] Segfault in "rotating_file" example
From: Jason Roehm (jasonr_at_[hidden])
Date: 2013-04-16 16:48:51
On 04/16/2013 04:38 PM, Andrey Semashev wrote:
> Well, from the backtrace it seems like the library tries to perform the last
> file rotation and calls filesystem::path::parent_path at some point, which
> tries to construct a path and access a locale facet for character code
> conversion. The problem is that the locale is stored as a function-local
> static in Boost.Filesystem and is destroyed at this point.
>
> I think this should be reported to Boost.Filesystem maintainer as this
> limitation with a global locale is quite unexpected from a user's standpoint.
> I think there is also a thread safety issue related to it that I reported
> quite some time ago, and it is still not fixed.
>
> As for Boost.Log, I'm not sure I can work it around in my code, except to
> avoid using Boost.Filesystem, which would be unfortunate. You can try to work
> around it on yor side though. If you access filesystem::path in the way so
> that the locale gets initialized _before_ you make any use of Boost.Log, then
> the locale should still be valid when Boost.Log is deinitialized.
>
Thanks for the recommendation. I wasn't able to figure out how to access
filesystem::path appropriately in order to keep the locale alive, but I
did create a workaround. Just before application shutdown, I call a
function that explicitly removes the sinks from the logging core, which
then causes them to be destroyed. The application then shuts down
successfully without any problems.
Jason