RFR: 8267952: async logging supports to dynamically change tags and decorators [v2]

[David Holmes]

On Thu, 10 Jun 2021 12:53:55 GMT, David Holmes <dholmes at openjdk.org> wrote:

>> Could you take a look at my argument? IMHO, C++ can guarantee the following code is atomic. 
>> Alignment shouldn't an issue. 
>> 
>> 
>> void LogOutput::set_decorators(const LogDecorators& decorators) {
>>     _decorators = decorators;
>> }
>> 
>> 
>> I manage to support the following code. The generated code is exactly same, but it needs some tricks of c++ meta-programming.  Is it worthy?  if you still insist we should use Atomic::store, how about I create a separate JBS issue and submit a patch?  My concern is that Atomic::load/store makes current code messy. 
>> 
>> 
>> void LogOutput::set_decorators(const LogDecorators& decorators) {
>>   Atomic::store(&_decorators, decorators);
>> }
>
> C++ makes no guarantees about atomicity unless using C++ atomic operations. Yes they probably are atomic but we don't rely on "probably". If a variable is being accessed lock-free and can take part in data-race then please just use Atomic::load and store for simple accesses (assuming they are correct in the context of how the variable can be used concurrently).

Sorry I just realized what you are saying here. This is a LogDecorator type not a primitive type so the notion of atomic load/store doesn't make sense. The comments about it being a bitfield threw me.

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

PR: https://git.openjdk.java.net/jdk/pull/4408