Request for review: 7034585 Adjust fillinStackTrace filtering to assist 6998871

[Tom Rodriguez]

Performance of fillInStackTrace is somewhat important so I think you should avoid using as_C_string since that creates a copy of both fillInStackTrace and the method being checked.  Maybe something like this:

const char* fillInStackTrace = "fillInStackTrace";
int len = 16;
assert(strlen(fillInStackTrace) == len, "must agree");
if (method->name()->starts_with(fillInStackTrace, 16) &&
    throwable->is_a(method->method_holder())) {

tom

On Apr 8, 2011, at 4:33 AM, David Holmes wrote:

> webrev:
> 
> http://cr.openjdk.java.net/~dholmes/7034585/webrev/
> 
> When an exception is created, fillInStacktrace is called to populate the backtrace information. This is done in java_lang_Throwable::fill_in_stack_trace in the VM. Because the interesting part of the stacktrace is from the location where the exception was created, and upwards, filtering is applied in fill_in_stack_trace to remove the entry for fillInStackTrace() itself, and the exception constructors.
> 
> The current filtering code only expects to find a single frame for the fillInStackTrace method, so if an exception class overrides fillInStackTrace (and invokes super.fillInStackTrace) this actually disables the filtering of the constructors. Eg we see:
> 
> Exception in thread "main" MyException
>        at MyException.fillInStackTrace(MyException.java:3)
>        at java.lang.Throwable.<init>(Throwable.java:260)
>        at java.lang.Exception.<init>(Exception.java:54)
>        at java.lang.RuntimeException.<init>(RuntimeException.java:51)
>        at MyException.<init>(MyException.java:1)
>        at MyException.main(MyException.java:7)
> 
> instead of:
> 
> Exception in thread "main" MyException
>        at MyException.main(MyException.java:7)
> 
> The changes to Throwable.java for 6998871 effectively introduce an additional fillInStackTrace() frame and so this too disables the desired filtering of the backtrace.
> 
> The proposal is quite simple: to modify fill_in_stack_trace so that it expects one or more fillInStackTrace frames, followed by zero or more constructor frames. This addresses the needs of 6998871 as well as fixing any user-defined overriding of fillInStackTrace.
> 
> Thanks,
> David