Re: java.util.Hashtable.size()/isEmpty()

[Artur Biesiadowski]

Eric Blake wrote:
> No - even accessing a simple field must be synchronized, for this
> reason:
>
> Suppose thread A is in the middle of calling clear() when thread B calls
> size().  Notice in clear() that the size is not changed until after all
> entries have been set to null, meaning that B will read a larger size
> than what is actually there.  For thread safety reasons, if you have a
> synchronized object, EVERY public access point to information on that
> object must be synchronized.

No,no,no.

Thread B could as well read the value from before clear happens. If it 
does not synchronize, there is not such thing as 'while A is middle of 
clear'. This two things does not happen in sequence - you have to think 
about them as about quantum physics - all possible cases.

So if we have A thread calling synchronized clear and B thread calling 
unsychronized size, B can return either previous size or 0 or even some 
value from long,long time ago - _regardless_ of sequence of evens that 
happened physically in processor. This is allowed behaviour.

If clear would remove element after element, then indeed, size would 
have to be synchronized - because it could report map having for example 
half elements (whereas it never happened - first it has n elements and 
then 0, because clear is atomic). BUT as clear update size atomically 
(regardless of synchronization), there is not need to also synchronize 
getter.

Artur