AFAIR, deadlock is due to that StampedLock is not re-entrant. We can use ReentrantReadWriteLock to avoid deadlock.
Cheers, Daniel Sun On 2025/01/24 04:48:48 Paul King wrote: > Hi folks, I am wondering whether the type for sourceCache in > GroovyClassLoader should be hard-coded as the concrete type > StampedCommonCache<String, Class> but instead be the interface > EvictableCache<String, Class>? > > https://github.com/apache/groovy/blob/master/src/main/java/groovy/lang/GroovyClassLoader.java#L102 > > This would let folks creating a custom class loader change to one of > the other caches, e.g. ConcurrentCommonCache if facing things like the > deadlock mentioned in StampedCommonCache (see also GROOVY-9742). That > is known to be slower but doesn't have the reentrancy issues. > > There is one place where a method only in StampedCommonCache is called > but using instanceof, we could fallback to only using the interface > methods (with some reduced functionality). > > This would break binary compatibility so I would only suggest it for Groovy 5. > > In addition to the type change, what do folks think about the proposed > workaround mentioned in GROOVY-9742, e.g. change > GroovyClassLoader#doParseClass() from private to public or protected. > This could potentially be done for an earlier version, e.g. Groovy 4. > > Cheers, Paul. >
