RFR 8231314: java.time serialization warning cleanup

[Peter Levart]

Once more, for the list (sorry)...

Hi,

On 9/21/19 12:31 PM, Chris Hegarty wrote:
> Roger,
>
>> On 20 Sep 2019, at 19:51, Roger Riggs <Roger.Riggs at oracle.com> wrote:
>>
>> Please review code cleanup that will remove the need to suppress soon 
>> to be introduced
>> warnings [1] about static typing of serialization related fields.
>> A few of the implementation Ser classes that serialize java.time 
>> types are affected.
>>
>> Webrev:
>> http://cr.openjdk.java.net/~rriggs/webrev-warn-serializable-8231314/
> I think the change is fine, just a few comments..
>
> - AbstractChronology is not necessarily is not always Serializable, 
> but writeReplace expects that it is.
> Since this is a package-private method, then I assume that it will 
> only ever be called by Serializable subtypes or the serialization 
> framework itself.
>
> - time/chrono/Ser.java : it’s kinda awkward to have to cast the return 
> of readExternal, especially since the set of types is locked down, but 
> without these readExternal methods returning xxxImpl types there is 
> little choice.

readExternal package-private methods could return Serializable, but then 
those methods would have to cast. Perhaps this would be better as it is 
"closer" to the real impl.

Another option (as I mentioned in previous message) would be to simply 
mark the Ser field as transient. The fact that currently it always holds 
an instance that *is* Serializable is just a coincidence, because it 
acts as a serialization proxy for Serializable implementations. Nothing 
in the logic of Ser class "requires" the instance assigned  to the field 
to be in fact Serializable.

Regards, Peter