|
Thanks a lot, Jc!
Serguei
On 11/13/18 08:59, JC Beyler wrote:
Ok makes sense then: it is not specified what
happens to locals that are out of scope and therefore depending
on the compiler/modes/tiers, you could get a different return in
those cases.
Webrev looks good to me now :)
Jc
Hi Jc,
Thank you a lot for reviewing!
On 11/12/18 09:35, JC Beyler wrote:
Hi Serguei,
The fix looks good (though I never like commented
out code, why do we not just remove the lines and
add a simple comment: "Due to JDK-8213525, we do not
test X,Y, and Z because of stability isssues").
I also normally do not like commented out code.
In this particular case, I considered commented out lines
as part of comment.
They explain what is removed better than any words. :)
Okay, I've removed these lines with the comment.
Forgot to tell that I've updated the same webrev:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2018/8213525-unstable-test.1/
Thanks,
Serguei
But the underlying question I have that is not
really explained is : "why is it failing?"; is the
spec not specific in these cases? is it a bug in the
compiler/runtime that is not yet fixed to conform to
the spec? I ask because I would imagine that it
might be something we would like to fix, no?
No.
There is no information from the JIT compiler to return
errors when the LVT is absent.
Moreover, different compilers, modes or tiers differently
represent local values that are out of scope.
Thanks,
Serguei
Please, review a fix for:
https://bugs.openjdk.java.net/browse/JDK-8213525
Webrev:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2018/8213525-unstable-test.1/
Summary:
A couple of the checks in new unit test
developed for JDK-8080406
is not stable.
It is expected that the type of the local
intLoc returned by the StackValueCollection
has
to be T_CONFLICT as it is out of scope at the
point where the testLocals() is called:
int staticMeth(byte byteArg, Object objArg, double dblArg, int intArg) {
testLocals(Thread.currentThread());
{
int intLoc = 9999;
intArg = intLoc;
}
return intArg;
}
But sometimes the type T_INT is returned instead of T_CONFLICT.
The fix is to disable the checks that can fail because of it.
Thanks,
Serguei
--
--
|
