RFR(M): 8199852: Print more information about class loaders in LinkageErrors.

David Holmes david.holmes at oracle.com
Thu Mar 22 21:20:49 UTC 2018


On 23/03/2018 2:21 AM, Lois Foltan wrote:
> On 3/21/2018 4:25 PM, Lindenmaier, Goetz wrote:
> 
>> ... wherever I report the class and it's loader I
>> will use class_loader_and_module_name() if I manage
>> to usefully extend it's information about the classloaders. ...
> Thank you Goetz, I did take a look at the 2 bugs you mentioned below. 
> Especially in the case of improving the IAE message, I think reporting 
> the module name is a must since lack of readability between 2 modules or 
> failure to export a package within a module can affect access.

Just on that note for the other bug ... ideally the IAE message relates 
more directly to the reason the access check failed. I would not want to 
see module and loader information added to the message for every single 
IAE, but only for those where module/package access is an issue. Having 
said that, loader name/description doesn't enter into it at all - the 
loader is not relevant to access checks.

David
-----

> Lois
> 
>>
>>> -----Original Message-----
>>> From: Lindenmaier, Goetz
>>> Sent: Wednesday, March 21, 2018 9:21 PM
>>> To: 'Lois Foltan' <lois.foltan at oracle.com>; hotspot-runtime-
>>> dev at openjdk.java.net
>>> Subject: RE: RFR(M): 8199852: Print more information about class 
>>> loaders in
>>> LinkageErrors.
>>>
>>> Hi Lois,
>>>
>>> The intention to normalize the output makes completely sense.
>>> It starts with using external_name everywhere ��.
>>>
>>> But you need a method that reports about a classloader that does
>>> not belong to a class. There are places where I want to report three
>>> class loaders, but only one class name.
>>>
>>> For example: "classloader A can not load class B because ..."
>>>
>>> Also, as I understand, the class loader can be in a different module
>>> than the loaded class. So you might want to report both. (Well,
>>> this could be handled by the enum you describe).
>>>
>>> I'll read the docu you mention.  I already had a look at the test
>>> you mention, but it seemed not to fit for the messages here.
>>> If the tests pass, I will post another change for IllegalAccessError
>>> that uses the same method. I wanted to hold that back until
>>> this one is finished, but maybe it makes sense to look at all the
>>> messages at a time.
>>>
>>> Please also have a look at the wording of the new messages in the
>>> bug description.
>>> https://bugs.openjdk.java.net/browse/JDK-8199852
>>> https://bugs.openjdk.java.net/browse/JDK-8199940
>>>
>>> Best regards,
>>>    Goetz.
>>>
>>>
>>>> -----Original Message-----
>>>> From: Lois Foltan [mailto:lois.foltan at oracle.com]
>>>> Sent: Wednesday, March 21, 2018 8:27 PM
>>>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; hotspot-runtime-
>>>> dev at openjdk.java.net
>>>> Subject: Re: RFR(M): 8199852: Print more information about class 
>>>> loaders in
>>>> LinkageErrors.
>>>>
>>>> On 3/21/2018 3:06 PM, Lois Foltan wrote:
>>>>
>>>>> On 3/21/2018 9:03 AM, Lindenmaier, Goetz wrote:
>>>>>
>>>>>> Hi Lois,
>>>>>>
>>>>>> Thanks for looking at this!
>>>>>>
>>>>>> I had a look at Klass::class_loader_and_module_name(), but there are
>>>>>> places
>>>>>> where you only want to report the class loader information, 
>>>>>> therefore I
>>>>>> decided to do it this way. (If you like this, I have more of these in
>>>>>> work :)).
>>>>>> For example see expectedErrorMessageDuplicate1 in the test.
>>>>>> Also, there might just be a classloader at hand, not a class as 
>>>>>> needed
>>>>>> for that method.
>>>>>>
>>>>>> But maybe it would make sense to call describe_external() in
>>>>>> class_loader_and_module_name() to get more information about
>>>>>> the classloader?
>>>>> Hi Goetz,
>>>>>
>>>>> In the JDK 9 time frame we strove to construct the string representing
>>>>> the class loader, module name & module version in a defined,
>>>>> structured manner.  This allows for consistency across error messages,
>>>>> stack dumps, etc.  Use of Klass::class_loader_and_name() gives us this
>>>>> consistency and you can see some examples of its use in
>>>>> ClassCastException messages. See,
>>>>> https://bugs.openjdk.java.net/browse/JDK-8165550 and the test,
>>>>> test/hotspot/jtreg/runtime/modules/CCE_module_message.java. I
>>>> strongly
>>>>> encourage you to use Klass::class_loader_and_name() and not add
>>>>> describe_external().  Klass::class_loader_and_name() could be enhanced
>>>>> to take an enum parameter that allows for tailoring of the message
>>>>> itself, for example including/not including the module name when
>>>>> available, including the parent loader, etc.  It could be pulled out
>>>>> of Klass for use to obtain a class loader's name when a Klass is not
>>>>> available, although Coleen just recently moved it to Klass.  I would
>>>>> support this type of change.
>>>> See the javadoc for java.lang.StackTraceElement,
>>>>
>>> https://docs.oracle.com/javase/9/docs/api/java/lang/StackTraceElement.ht
>>>> ml.
>>>> Specifically the API Note in the StackTraceElement.toString()
>>>> description which documents the format and contains examples.
>>>> Lois
>>>>
>>>>> Thanks,
>>>>> Lois
>>>>>
>>>>>> Best regards,
>>>>>>     Goetz.
>>>>>>
>>>>>>
>>>>>>
>>>>>>> -----Original Message-----
>>>>>>> From: Lois Foltan [mailto:lois.foltan at oracle.com]
>>>>>>> Sent: Mittwoch, 21. März 2018 13:22
>>>>>>> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; hotspot-
>>>> runtime-
>>>>>>> dev at openjdk.java.net
>>>>>>> Subject: Re: RFR(M): 8199852: Print more information about class
>>>>>>> loaders in
>>>>>>> LinkageErrors.
>>>>>>>
>>>>>>> On 3/21/2018 7:55 AM, Lindenmaier, Goetz wrote:
>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> ClassLoaders have a field containing a name since Java 9.
>>>>>>>>
>>>>>>>> This change improves some of the messages of LinkageErrors to
>>>>>>>> print the name of the loader involved, as well as the parent of the
>>>>>>>> loader.
>>>>>>>> This simplifies tracking down the component causing an error in 
>>>>>>>> large
>>>>>>>> Java systems.
>>>>>>>>
>>>>>>>> Please review:
>>>>>>>> http://cr.openjdk.java.net/~goetz/wr18/8199852-
>>>>>>> exMsg_Linkage/01/index.html
>>>>>>>> I ran this through all our tests (jck, jtreg, spec etc).  I will
>>>>>>>> run it through
>>>>>>> submit-hs
>>>>>>>> (or jdk/submit) once it's reviewed.
>>>>>>>>
>>>>>>>> Best regards,
>>>>>>>>      Goetz.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>> Hi Goetz,
>>>>>>>
>>>>>>> This looks promising and something that we started to do in the 
>>>>>>> JDK 9
>>>>>>> time frame.  Instead of adding the new method
>>>>>>> java_lang_ClassLoader::describe_external() can you possible
>>>> use/improve
>>>>>>> upon Klass::class_loader_and_module_name() in order to avoid
>>>> duplicate
>>>>>>> code?
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Lois
> 


More information about the hotspot-runtime-dev mailing list