Would it make sense to make it so that __gshared implicitly converted to shared? It's my understanding that the main purpose of __gshared is to be able to better interact with C code, but given the issues with shared, lots of people have just used __gshared instead of shared. This causes a problem in some cases. For instance, none of Mutex's functions are currently shared in spite of the fact that it really doesn't make sense to have a Mutex which is thread-local, but it can't have all of its functions be only shared, because then it wouldn't work with __gshared. That means that in order to work with both __gshared and shared, all of its functions must be duplicated, which is obviously less than ideal.

So, given that __gshared is shared across threads like shared is (just with fewer protections), would it make sense to make it so that __gshared implicitly converts to shared? Then a type like Mutex which is intended to be shared, can just make all of its member functions shared, and it'll work with both __gshared and shared.

It may also be necessary to make shared implicitly convert to __gshared for that to work cleanly (particularly when you get stuff like a member function returning a reference variable which then must be shared, even if the original variable were __gshared - because the function itself is shared), and I don't know how big a problem that would be. But I think that the basic idea of allowing implicit conversions at least from __gshared to shared (if not shared to __gshared) is worth exploring. Is there a major reason why it would be a bad idea? My experience with both is limited, and I may just be completely missing something here.

- Jonathan M Davis

On Tuesday, 20 November 2012 at 19:15:01 UTC, Jonathan M Davis wrote:
> Would it make sense to make it so that __gshared implicitly converted to
> shared? It's my understanding that the main purpose of __gshared is to be able
> to better interact with C code, but given the issues with shared, lots of
> people have just used __gshared instead of shared. This causes a problem in
> some cases. For instance, none of Mutex's functions are currently shared in
> spite of the fact that it really doesn't make sense to have a Mutex which is
> thread-local, but it can't have all of its functions be only shared, because
> then it wouldn't work with __gshared. That means that in order to work with
> both __gshared and shared, all of its functions must be duplicated, which is
> obviously less than ideal.
>
> So, given that __gshared is shared across threads like shared is (just with
> fewer protections), would it make sense to make it so that __gshared
> implicitly converts to shared? Then a type like Mutex which is intended to be
> shared, can just make all of its member functions shared, and it'll work with
> both __gshared and shared.
>
> It may also be necessary to make shared implicitly convert to __gshared for
> that to work cleanly (particularly when you get stuff like a member function
> returning a reference variable which then must be shared, even if the original
> variable were __gshared - because the function itself is shared), and I don't
> know how big a problem that would be. But I think that the basic idea of
> allowing implicit conversions at least from __gshared to shared (if not shared
> to __gshared) is worth exploring. Is there a major reason why it would be a
> bad idea? My experience with both is limited, and I may just be completely
> missing something here.
>
> - Jonathan M Davis

I'm short on time, and so can't double-check the online docs or the book, but as I recall (having never actually used __gshared... yet) they differ in that shared is a type constructor and __gshared is a declaration attribute.  So I'm just not sure.  On the one hand, the compiler would always be able to see that either the left or right side of an assignment (or the inside of an argument list) includes a __gshared variable, and could do the "right" thing...  But on the other hand it would be an odd special case of a type qualifier being cast away because of a variable's attribute.  Is this something we want?  Or is it intended as a short-term workaround until shared is more fully/correctly defined semantically?

On 2012-11-20 20:14, Jonathan M Davis wrote:
> Would it make sense to make it so that __gshared implicitly converted to
> shared? It's my understanding that the main purpose of __gshared is to be able
> to better interact with C code, but given the issues with shared, lots of
> people have just used __gshared instead of shared.

I don't think it's just because there are issues with "shared". If "shared" means other semantics than just a global variable the D code will have different semantics when accessing the variable compared to the C code. I don't think that's a good idea. If it's not safe to just access the variable as is, I think it's better to add a safe wrapper then changing the semantics of the bindings.

-- 
/Jacob Carlborg