[15] RFR: 8248476: No helpful NullPointerException message after calling fillInStackTrace

[Remi Forax]

----- Mail original -----
> De: "David Holmes" <david.holmes at oracle.com>
> À: "Goetz Lindenmaier" <goetz.lindenmaier at sap.com>, "hotspot-runtime-dev" <hotspot-runtime-dev at openjdk.java.net>
> Envoyé: Lundi 29 Juin 2020 15:19:29
> Objet: Re: [15] RFR: 8248476: No helpful NullPointerException message after calling fillInStackTrace

> Hi Goetz,
> 
> On 29/06/2020 9:17 pm, Lindenmaier, Goetz wrote:
>> Hi,
>> 
>> If changing the stack trace by calling fillInStackTrace in user code, the
>> NPE algorithm lacks the proper information to compute the message.
>> Thus, we must omit it after that call.
> 
> For those of us who didn't follow this work in detail could you outline
> exactly when in the lifecycle of a NPE the detailed error message is
> created. I'm having trouble understanding the exact nature of the
> problem, and thus the fix.

I"m not Goetz but let me try,
you want to provide helpful error message for NPE but without any runtime cost for the other exceptions or if the NPE is not printed (or if it is the special shared unmutable NPE used by Hotspot).
the idea is to inspect the backtrace of the NPE only when getMessage is called on a NPE and the detailMessage is null [1].
getExtendedNPEMessage() is a native method that will extract the first stack frame and use the bci to find where the error occurs and try using an abstract interpretation of the bytecode to provide information about the context.

The current bug is that fillInStackTrace can be called on an existing NPE instance thus changing the backtrace to a place that will confuse the algorithm that calculate the context so the helpful message provided in that case is useless. The proposed way to fix that is to detect if fillInStackTrace() is the first stack frame of the backtrace and in that can to not compute the helpful message. This fix works well but for one case if null.fillInStackTrace() is called.

> 
> Thanks,
> David
> -----


[1] https://github.com/openjdk/jdk/blob/master/src/java.base/share/classes/java/lang/NullPointerException.java#L92

> 
>> I implement this by checking for a call to fillInStackTrace at the bci
>> recorded in the exception.
>> http://cr.openjdk.java.net/~goetz/wr20/8248476-NPE_fillInStackTrace-jdk15/01/
>> 
>> Is there any other way of changing the stack trace?  setStackTrace() is
>> no problem as it does not change the internal backtrace datastructure.
>> 
>> I'll push the change to jdk/submit and run it through our CI before pushing.
>> 
>> Best regards,
>>    Goetz.
>>