Hi Sergey,

 

I understand the behavior change of raising an exception vs quietly returning 
null. As per my understanding, this happens only when someone passes a null for 
the locale key right? If so, how about explicitly checking for null, and 
throwing exception?

 

@Semyon,

 

I think we should address this change in a separate bug, since we want behavior 
question answered.

 

Thanks,

Krishna

 

From: Semyon Sadetsky 
Sent: Thursday, March 8, 2018 1:05 AM
To: Sergey Bylokhov <sergey.bylok...@oracle.com>; Krishna Addepalli 
<krishna.addepa...@oracle.com>; Philip Race <philip.r...@oracle.com>
Cc: swing-dev@openjdk.java.net
Subject: Re: <Swing Dev> 
[11][JDK-8197785]javax.accessibility.AccessibilityBundle will reload the 
ResourceBundle for every call to toDisplayString

 

I just noticed that null returning scenario only possible when  locale 
independent state name is null but the latter should never happen. So we simply 
may ignore this null/NPE worry.

--Semyon

 

On 3/7/18 11:17 AM, HYPERLINK 
"mailto:semyon.sadet...@oracle.com"semyon.sadet...@oracle.com wrote:

On 3/7/18 10:40 AM, Sergey Bylokhov wrote:

On 07/03/2018 10:21, HYPERLINK 
"mailto:semyon.sadet...@oracle.com"semyon.sadet...@oracle.com wrote: 



That's not ok, since we change a behavior without discussion that the benefits 
of change outweigh the disadvantages, when we implemented some unrelated fix. 

That's not true. You may find the discussion in the current thread. Fill free 
to provide your arguments to it. 


I found nothing related to the behavior, only some conclusion about possible 
performance difference. 

So you agree?







It is an improvement. NPE is not expected result according to the 
toDisplayString() method spec. 


It also changed result of toDisplayString(), previously it did not returned 
null. I think .01 is better and simpler. 

Returning null is better than throwing NPE especially in the case when NPE 
doesn't correspond to the spec. Krishna kindly agreed to fix both issues at 
once. I didn't get what is the problem? 


Nope, if the method start to return null means that all its usage should be 
checked for null return value. 

You always should check for null return if the spec does not say that null is 
never returned. In this specific case if null check is absent the NPE will be 
thrown in the user code instead of JDK code, so there is no degradation.



As I said it is better to use version .01 which fixed the problem w/o side 
effects. 

Thanks for your advice. Try to provide more arguments next time.

 

Reply via email to