RFR: 8196340: (coll) Examine overriding inherited methods in ArrayList and ArrayList.SubList

[Ivan Gerasimov]

On 5/14/18 5:22 PM, Claes Redestad wrote:
>
>
> I was actually toying with and testing a change to this effect anyway, 
> since it's a nice cleanup and might help the JIT somewhat:
>
> http://cr.openjdk.java.net/~redestad/8196340/open.02/
>
> A note on correctness: none of the code in ArrayList that utilize 
> modCount is "correct" or accurate in a concurrency or thread-safety 
> context where ordering of loads and stores is critical for 
> correctness: lacking volatile or other means to ensure correct 
> ordering of operations (or atomicity of modCount increments) the 
> runtime is given large amounts of freedoms to do as it please to 
> optimize things.  This is intentional as ArrayList isn't supposed to 
> be neither thread-safe nor concurrent on its own.  The main utility is 
> instead about detecting and avoid coding errors, mainly bugs like 
> adding to or removing items from a list you're simultaneously 
> iterating over. So what I'm saying is that the ordering of operations 
> within a method might not be all that important as long as they are 
> sound within the scope of the method execution. Still, making it a bit 
> more like correct concurrent code might improve the best effort 
> behavior in some circumstances, and costs us nothing.
>
I totally agree with this!

For this reason let's move the line
586         final int otherModCount = other.modCount;
to the beginning of equalsArrayList(ArrayList<?> other), so it is 
initialized before other.size is read?

With kind regards,
Ivan

> Thanks - and sorry for dragging you along for a bit longer...
>
> /Claes
>

-- 
With kind regards,
Ivan Gerasimov