RFR (XS) 8216977: ShowHiddenFrames use in java_lang_StackTraceElement::fill_in appears broken

[David Holmes]

Hi Coleen,

On 29/08/2019 1:19 pm, coleen.phillimore at oracle.com wrote:
> On 8/28/19 9:36 PM, David Holmes wrote:
>> Hi Coleen,
>>
>> On 29/08/2019 6:25 am, coleen.phillimore at oracle.com wrote:
>>> Summary: Return NULL source file and negative line number for hidden 
>>> frames.
>>>
>>> See bug for more details.   Ran hs tier1-3 on linux-x64-debug and 
>>> tier1 on all platforms.
>>
>> This issue made my head spin. I think what you have done has taken 
>> things back to a sane place where ShowHiddenFrames is not being used 
>> to do whacky things unrelated to hidden frames.
>>
>>> open webrev at 
>>> http://cr.openjdk.java.net/~coleenp/2019/8216977.01/webrev
>>> bug link https://bugs.openjdk.java.net/browse/JDK-8216977
>>
>> src/hotspot/share/classfile/javaClasses.inline.hpp
>>
>> Okay - this code was reporting a specially encoded "line number" that 
>> actually contained the BCI, but only if ShowHiddenFrames was true. 
>> Getting rid of this is good because it has nothing to do with hidden 
>> frames and there wasn't even a check that the current frame was in 
>> fact hidden.
>>
> 
> Yes, it could have checked if method->is_hidden() give the special line 
> number with bci encoded.   Maybe tools can print something useful with 
> that information. StackTraceElement.toString() doesn't though.
> 
>> If we want to report the BCI when no line number table exists then 
>> that should be a specific enhancement to the API.
> 
> I agree.
>>
>> ---
>>
>> src/hotspot/share/classfile/javaClasses.cpp
>>
>> Okay - this code was changing a NULL source file to "unknown" but only 
>> if ShowHiddenFrames is true. Again getting rid of this is good as it 
>> again has nothing to do with hidden frames.
>>
>> If we want to report "unknown" rather than NULL, then we should make 
>> that change independent of ShowHiddenFrames.
>>
> 
> Calling it <Unknown> makes it inconsistent with the case where the 
> source file is unknown, which java.lang.StackFrame prints.
>> ---
>>
>> test/hotspot/jtreg/runtime/StackTrace/HiddenFrameTest.java
>>
>> Can you add a comment explaining why String::concat meets the criteria 
>> of "null source file" and "no line number table", as it is not at all 
>> obvious to me what that should be the case. Also is that true if we 
>> disable the String concatenation use of lambda? If not we should add 
>> the necessary -XX:+ flag to the @run line.
> 
> I don't know what that flag is.  It might be a property.  I don't think 
> we'd run all the tests with such a flag though so I think this is safe.

Sorry was confusing different things. You're just using a method 
reference to String::concat which will use lambda expressions under the 
hood. I was thinking about the lambdification of string concatenation 
which is completely different.

>>
>> For completeness can you check that the NPE only has one frame if 
>> -XX:-ShowHiddenFrames is used.
> 
> Sure:
> 
> http://cr.openjdk.java.net/~coleenp/2019/8216977.02/webrev/test/hotspot/jtreg/runtime/StackTrace/HiddenFrameTest.java.html 

Thanks! One nit:

             String arg = (String)args[0];

No need for a cast there - args is an array of String.

No need for new webrev.

Thanks,
David


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