On Tue, 2015-06-16 at 14:06 +0100, Jonathan Wakely wrote: > On 16 June 2015 at 13:43, Nick wrote: > > By "disappears" I assume you mean the contents of the space are gone, > > not that the actual address space is removed, correct? > > I don't understand the distinction. What does "the actual address > space is removed" mean? By "removed" I get the impression that it's somehow literally gone. For example, a 4GB address space is now 3GB. The address space is still there, it just contains undefined content which naturally will lead to undefined behavior if accessed. > > So in the case > > of the OP, that would amount to allocated objects (ex. function > > template) > > Template instantiations are not allocated. They are code, not data. Right, in the case of a basic C++ function template. From the OP I was thinking about something like std::function which I'd expect (though I didn't look at the implementation) has some data along with it. > > as well as code from the shared object are potentially wiped > > (but certainly invalid regardless) and therefore any access to them is > > undefined. > > > > If that's all correct, and if the shared object and the main app are > > using the same memory manager, then in the OP, isn't the function > > template still around in memory (assuming it was allocated by the app's > > mem mgr)? > > I'm not sure what you mean by "function template". Do you mean the > variable of type "std::function<void()>"? That is not a template, it's > an object. It's still around. Right. > > And so the problem is that the shared object's *code* is gone > > and so the std::function dtor segfaults because it's trying to execute > > code that's no longer in the address space? > > Correct. > > It's trying to execute an instantiation of a function template (in the > basic C++ sense of the term, not the "std::function" sense, which > coincidentally has the same name). That instantiation is a piece of > code defined in the shared library. When the shared library is > unloaded the code cannot be called. It's address no longer points to > anything in the process' address space. Very helpful, thank you.
