All,

The attached zip file contains a Win32 implementation for Java's
wait()/notify()/notifyAll() functionality for D.

If you examine the source file waitnotify.d, it contains the following declaration in the WaitNotifyObject class:

public bool dummy = false;

It is an unused variable.  With it in there, the library works fine.  When I take out the dummy variable, Access Violations occur.  (Try compiling by issuing "make" and run "main.exe".)

Why is this happening?  Is it a problem with the compiler?  With the public.d / privateImpl.d paradigm in use?  Is it an alignment issue?

--Dan

Thank you for your waitnotify library. I was looking for something like this. I have implemented something myself but your implementation is better. I believe these wait(), notify() functionalities should be part of the Object class, which means to recompile phobos. The problem is that it is not possible to add members to the Object class. I've tried that and the compiler fails on an assertion on the Object size.

For your SIGSEGV problem, I believe that it is due to the public/private implementation. In your waitnotifyImpl.d you added an int member to the WaitNotifyObject. It means that the size of a WaitNotifyObject will be the size of Object + the size of an int. In your public waitnotify.d, if you don't add a member the WaitNotifyObject will appear to have the same size of a basic Object but it is bigger in fact. I believe the hidden int member gets altered by other objects in memory, the place where it lies looks available. I had similar problems when I was playing with phobos (adding members to classes and recompiling).

Another thing that is important when using public/private implementation is to
always keep the methods in the same order. This is because the D compiler stores
the methods in the virtual function table in the order they appear in the source
file. For example if you have a public file like this :
class A {
   void fun();
   void afun();
}
and a private file like this
class A {
   void afun() { ... }
   void fun() { ... }
}
then when you will call A.fun() from a program importing your public d file and
linked with your already compiled private implementation you will have A.afun()
called instead. The same happens when you try to call A.afun(), you call A.fun()
instead.
I had this problem too.

I hope you could read my poor english and that I could help you.

Vincent

== Quote from catrino (catrino@cbs.cnrs.fr)'s article
> Thank you for your waitnotify library. I was looking for something like this. I
> have implemented something myself but your implementation is better. I believe
> these wait(), notify() functionalities should be part of the Object class, which
> means to recompile phobos. The problem is that it is not possible to add members
> to the Object class. I've tried that and the compiler fails on an assertion on the
> Object size.
> For your SIGSEGV problem, I believe that it is due to the public/private
> implementation. In your waitnotifyImpl.d you added an int member to the
> WaitNotifyObject. It means that the size of a WaitNotifyObject will be the size of
> Object + the size of an int. In your public waitnotify.d, if you don't add a
> member the WaitNotifyObject will appear to have the same size of a basic Object
> but it is bigger in fact. I believe the hidden int member gets altered by other
> objects in memory, the place where it lies looks available. I had similar problems
> when I was playing with phobos (adding members to classes and recompiling).
> Another thing that is important when using public/private implementation is to
> always keep the methods in the same order. This is because the D compiler stores
> the methods in the virtual function table in the order they appear in the source
> file. For example if you have a public file like this :
> class A {
>    void fun();
>    void afun();
> }
> and a private file like this
> class A {
>    void afun() { ... }
>    void fun() { ... }
> }
> then when you will call A.fun() from a program importing your public d file and
> linked with your already compiled private implementation you will have A.afun()
> called instead. The same happens when you try to call A.afun(), you call A.fun()
> instead.
> I had this problem too.
> I hope you could read my poor english and that I could help you.
> Vincent

Vincent,

I appreciate the advice.  I will try a few things, like experimenting with volatile, making sure methods are in the same order, etc.

Thanks.
Dan