Proposed API for JEP 259: Stack-Walking API

Mandy Chung mandy.chung at oracle.com
Tue Nov 17 20:50:32 UTC 2015


> On Nov 17, 2015, at 11:54 AM, Peter Levart <peter.levart at gmail.com> wrote:
> 
> 
> 
> On 11/16/2015 08:16 PM, Mandy Chung wrote:
>>> On Nov 15, 2015, at 10:59 AM, Peter Levart <peter.levart at gmail.com>
>>>  wrote:
>>> 
>>> OTOH in the described cases, a caller of walker.getCallerClass() is actually expecting to be called by a Java method, right? So what would it be if such "caller-sensitive" method demanded to be called by a Java method and throw IllegalArgumentException("Not called by Java method") otherwise?
>>> 
>>> Have you thought of that possibility?
>>> 
>> Are you thinking about a JNI method calling getCallerClass?  Or refer getCallerClass being the bottom of the stack?
>> 
>> I don’t see any issue for a JNI method calling getCallerClass and it’s doing:
>> *     Class<?> caller = walker.walk(s ->
>> *         s.map(StackFrame::getDeclaringClass)
>> *          .skip(2)
>> *          .findFirst());
>> 
>> Mandy
>> 
> 
> Hi Mandy,
> 
> No, I was not thinking of StackWalker::getCallerClass being called by JNI. I was thinking of a "caller-sensitive" method calling getCallerClass() and that "caller-sensitive" method being called by JNI from a newly attached thread. 
> 
> 1st, I don't think anyone will attempt calling getCallerClass() from the static main() method invoked by java launcher. That's silly and need not be supported.
> 2nd, I don't think anyone will attempt calling getCallerClass() from overridden Thread::run() method. That too need not be supported.
> 
> So what we are left with are other "caller-sensitive" methods calling getCallerClass(). Like for example:
> 
> public class Strings {
>     public static ResourceBundle getBundle() {
>         Class<?> cc = stackWalker.getCallerClass();
>         return ResourceBundle.getBundle(cc.getName() + "$Strings", Locale.getDefault(), cc.getClassLoader());
>     }
> }
> 
> Such method obviously expects to be called by some Java method. So it may as well demand it. By throwing exception if it is called by JNI from newly attached thread.
> 
> "caller-sensitive" methods could be viewed as taking an implicit parameter - their caller. If that parameter is not specified, they could behave like when some other explicit parameter was not specified or invalid - throw new IllegalArgumentException("Not called by Java method") …
> 

I thought about this case and I consider it the same category as the static main method case which is a method being invoked by newly JNI attached thread where the entry point is the last frame on the stack.  I think common cases would have the entry point to invoke other java methods including caller-sensitive ones.

I see your point of those methods being abused.

The other alternative I considered is throwing UnsupportedOperationException if there is no caller frame, i.e. skip(2) doesn’t exist.  I agree that it’s a better option than returning (n-1)-th frame if it’s the last frame on the stack.

I will keep returning the thread’s entry point case to return the class of the runnable instead of returning Thread.class.

Mandy

> So instead of "pretending" they were called by themselves, which might not be right, simply refuse the invocation with exception. I don't think this would hinder any such caller-sensitive method in it's utility. More importantly, it would prevent such methods from being abused.

> 
> You said about possible use-cases for getCallerClass:
> 
>> I have been thinking what the users would do when there is no caller.
>> 
>> The JDK use  of getCallerClass are to:
>> 1. find the caller's class loader for class loader hierarchy check
> 
> Abuse: calling such "caller-sensitive" method by JNI from newly-attached thread would use the caller-sensitive method's declaring class ClassLoader for hierarchy check - wrong!
> 
>> 2. find the caller’s class loader to load something on behalf of the caller (visibility)
> 
> Abuse: ... would use the caller-sensitive method's declaring class ClassLoader to load something - might work, might not
> 
>> 3. looking forward to work with modules, one would want to get the caller’s module
>> - to load something on behalf of the caller's module
> 
> Abuse: ...would load something from caller-sensitive method's declaring class module - wrong.
> 
>> - to add read edge to the caller’s module before accessing its type
> 
> Abuse: ...would add a read edge to itself before accessing some type - wrong.
> 
> 
> In all those cases throwing exception would be the right thing to do.
> 
> 
> What do you think?



More information about the core-libs-dev mailing list