Sorry for being rather late with this, I've been swamped with other things.

Thanks for this, Binxinag, it's always good to sort out these small issues.

However, there is a bit of a complication, at this stage the code copies a pointer and also an indicator of whether the object pointed to belongs to a sampler. If you copy a sampler which owns pAdapt and delete both of them then they will both delete the same object.

I suspect that the cleanest option would be to do a deep copy of the object pointed to and then to set adaptBelong to TRUE regardless of what the original configuration was, but I haven't thought very much about this.

Any other thoughts on this?

Got it. Perhaps we could do it like

 *pAdapt = *sFrom.pAdapt;
 adaptBelong = sFrom.adaptBelong;

Then even if adaptBelong == true, we do not delete the same object.

Then one new problem may come out. When pAdapt == null at first, *pAdapt may bring segment fault. :(

Quite so. This is where pAdapt would be initialised so to go down that route you'd need to allocate memory before assigning to it -- but that wouldn't need much more than an additional
pAdapt = new...
assuming that the deep copy model is the approach we want to go with here.

Also, if we do do this then it should probably be
adaptBelong = true
rather than
adaptBelong = sFrom.adaptBelong
because even if the source object doesn't own the object pointed to by its pAdapt the copied object does.

Thanks Adam for picking up on this. I'd forgotten about this issue of dealing with the pointer in the copy. I think I put this off because I didn't have a great solution at the time but I like the sound of your solution.

There's been a history of omitting the copy constructor because it was fiddly for various reasons dating right back to the first implementation. I think we're closing in on something that will at least work -- it will be great to have something in place, but we should remember to be clear in any documentation exactly what the behaviour on copying is as it might not be quite what is expected.

In particular, when the object pointed to by pAdapt is a derived class we'll call the base class constructor / assignment operator if we go down this route and the copied sampler won't behave in quite the same way as the original. The user can address this by explicitly setting the value of this pointer, as long as they're aware that they need to.

(The only obvious alternative that I can see would be to always copy the pointer and then set adaptBelong = false but that runs into problems if the original sampler owned the object and is destroyed before the copy... but if anyone has other thoughts they'd be very welcome.)