RFR: 8205957: setfldw001/TestDescription.java fails with bad field value

[Serguei Spitsyn]

On Tue, 20 Aug 2024 02:15:18 GMT, Dean Long <dlong at openjdk.org> wrote:

>> src/hotspot/share/runtime/javaCalls.cpp line 358:
>> 
>>> 356: #endif
>>> 357: 
>>> 358:   CompilationPolicy::compile_if_required(method, CHECK);
>> 
>> I'd like to understand what is going to happen at the line 358. Are we going to compile method with -Xcomp even though the `interp_only` mode is required? If so, do we waist cycles in doing this?
>> This question is not to Leonid. Maybe @dean-long could answer this, please?
>
> Yes, I believe we will compile the method even in interp-only mode.  As -Xcomp is a mode to stress the compiler, I don't think it matters if we compile eagerly here or wait until interp-only mode is turned off.  Other threads not in interp-only mode will compile the method anyway.  But if it makes sense to change this behavior, then let's do that in a separate bug/RFE.

Thank you, Dean.
The issue I see is that the tests are extremely slow with the -Xcomp option.
So, it would be good to avoid unneeded proactive compilations.
Let's address it in a separate bug/RFE. I'm expecting more places where we do the same or similar compilations. Another issue is that we sometimes execute compiled methods on `interp_only` threads, so there have to be more missed checks in the code.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/20587#discussion_r1722604938