RFR JDK-8209173: javac fails with completion exception while reporting an error

[Maurizio Cimadamore]

Ok - thanks for the explanation. In general, making rich diagnostic 
formatter more robust to completion errors seems the right way to go 
about those things, so I'm ok with the fix.

Maurizio


On 24/08/18 21:42, Vicente Romero wrote:
>
>
> On 08/24/2018 04:31 PM, Maurizio Cimadamore wrote:
>> Hi,
>> the fix looks good. If I understand correctly, the issue was that, 
>> since the diagnostic generation ended up with an exception, the 
>> diagnostic never really arrived at the speculative attribution 
>> handler, which meant it was not reported. Correct?
>
> correct, although at the end it was not the speculative attribution 
> handler the one that was involved here. I mistakingly thought that, it 
> was just the default handler, but your still your analysis is correct
>
>>
>> Maurizio
>
> thanks for the review
> Vicente
>>
>>
>> On 24/08/18 20:52, Vicente Romero wrote:
>>> Hi,
>>>
>>> Please review patch for [1] at [2]. The original issue was reported 
>>> for a combination of a Scala class and a client class written in 
>>> Java. When the Java class was compiled with javac, it couldn't find 
>>> in the class path a class imported by the Scala class, the missing 
>>> class belongs to the Scala standard library. The root issue was that 
>>> the symbol corresponding to the missing Scala class could not be 
>>> completed and a completion exception was thrown. The exception was 
>>> fired while javac was producing the corresponding diagnostic error 
>>> message. Later on the compiler was swallowing the exception, but the 
>>> AST ended up having null fields. This provoked a NPE during a later 
>>> compilation phase. The provided test reproduces the issue using Java 
>>> classes only.
>>>
>>> Thanks
>>>
>>> [1] https://bugs.openjdk.java.net/browse/JDK-8209173
>>> [2] http://cr.openjdk.java.net/~vromero/8209173/webrev.00/
>>
>