RFR: 8081833: Clean up JVMFlag getter/setter code

[Ioi Lam]

On Tue, 15 Sep 2020 16:57:12 GMT, Gerard Ziemski <gziemski at openjdk.org> wrote:

>> This is a follow to [JDK-8243208 Clean up JVMFlag implementation](https://bugs.openjdk.java.net/browse/JDK-8243208).
>> 
>> I try to use templates and inheritance (and a bit of macros) to get rid of the large amount of duplicated code in the
>> JVMFlag getters/setters.
>> Stefan Karlsson and I have tried this several times already, but I think the current attempt probably has the least
>> amount of hacks and most functionality:
>> - Type safety is enforced up to the point of loading/storing `JVMFlag::_addr`, without any unchecked typecasts.
>> - Runtime type checking is implemented by the `JVM_FLAG_TYPE(t)` macro (see jvmFlagAccess.hpp), by passing along an
>>   integer (`JVMFlag::TYPE_int`, etc). This is more efficient than previous attempts that passed along a function pointer
>>   (see http://cr.openjdk.java.net/~stefank/8081833/webrev.04/)
>> - The old `JVMFlag::xxxAtPut` functions were undocumented. There are actually two groups that have different usage. I
>>   added documentation in jvmFlagAccess.hpp to explain this.
>> - I got rid of the `JVMFlagEx` class and moved the code into `JVMFlag`. This is possible with C++11 which allows forward
>>   declaration of `enum JVMFlagsEnum : int;`
>> - I changed `JVMFlag::_type` from a string into an enum. I am surprised this had lasted for over 20 years.
>> - I also changed `JVMFlag` from a struct into a class and made some fields private. I probably will do more cleanup in a
>>   separate RFE.
>> 
>> Please start with jvmFlagAccess.hpp, jvmFlagAccess.hpp and then globals_extension.hpp.
>> 
>> Probably not everyone is in love with the `JVM_FLAG_TYPE(t)` macro. I'll be happy to try out alternatives.
>
> src/hotspot/share/runtime/flags/jvmFlagLimit.cpp line 156:
> 
>> 154:   for (int i = 0; i < NUM_JVMFlagsEnum; i++) {
>> 155:     JVMFlagsEnum flag_enum = static_cast<JVMFlagsEnum>(i);
>> 156:     if (get_range_at(flag_enum) != NULL &&
> 
> VMFlagAccess::check_range(JVMFlag::flag_from_enum(flag_enum) will check for "get_range_at(flag_enum) != NULL", so no
> need for this check here?

This function is called during VM start-up, so I want to short-circuit the check, and avoid calling
VMFlagAccess::check_range if the flag has no ranges.

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

PR: https://git.openjdk.java.net/jdk/pull/163