On Mon, 6 Nov 2023 15:49:02 GMT, Alan Bateman <al...@openjdk.org> wrote:
>> Could I have a review of a PR that removes the bytecode instrumentation for
>> the exception events.
>>
>> Testing: jdk/jdk/jfr + tier1 + tier2
>
> src/java.base/share/classes/jdk/internal/event/ThrowableTracer.java line 37:
>
>> 35: private static final AtomicLong numThrowables = new AtomicLong();
>> 36:
>> 37: public static void enable() throws NoSuchFieldException,
>> SecurityException, IllegalArgumentException, IllegalAccessException {
>
> SecurityException and IllegaArgumentException are unchecked so don't need
> them in the throws declaration.
Will fix.
> src/java.base/share/classes/jdk/internal/event/ThrowableTracer.java line 44:
>
>> 42:
>> 43: public static void traceError(Class<?> clazz, String message) {
>> 44: if (OutOfMemoryError.class.isAssignableFrom(clazz)) {
>
> StackOverflowError is likely problematic too, maybe it should be
> VirtualMachineError.
I remember asking the same question ten years ago, when Nils did the original
implementation, but I think it was only needed for OOM, because it creates an
infinite loop when the event object was allocated, which resulted in a
StackOverflowError instead OOM.
Some OOM tests failed with JFR enabled.
The event object allocation has been removed, but I think we can run into the
allocation recursion by other means. I looked into it a few years ago, but I
don't remember exactly why it failed.
If a SOE happens, I think we are fine. There is something that prevents
infinite recursion when the SOE object is created. Perhaps it is preallocated?
I prefer to not change the behavior, at least not in this PR.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/16493#discussion_r1384124648
PR Review Comment: https://git.openjdk.org/jdk/pull/16493#discussion_r1384124213
