[ https://issues.jboss.org/browse/WELD-1100?page=com.atlassian.jira.plugin.... ] Stuart Douglas commented on WELD-1100: -------------------------------------- So I have had some thoughts about how best to do this, and some idea's of the potential benefits it can provide. As much as possible we should be avoiding using java.lang.reflect.*, as it is really slow. Even things that you would expect to be fast such as accessing a member on an annotation are actually relatively slow. At boot time we use reflection for a few different things: *Identifying CDI scopes, produces, injection points etc* This can be completely replaced with Jandex, as there is no spec requirement to use java.lang.reflect.*. I think in practice this would mean a move away from using WeldAnnotated* to determine this information, which I think is a good thing, as I think WeldAnnotated is doing to many things at the moment. *Building the AnnotatedType used in ProcessAnnotatedType* Even though this is spec mandated, I think there are a few different things we could explore here to improve performance. 1) May the AnnotatedType implementation lazy. This will not help much if there are portable extensions that need to examine field/method level metadata, but it is a start, and will be a big win if we come up with a solution to CDI-43. 2) Use our own annotation implementations instead of the JDK's annotations. Because we have the Jandex indexes we can tell which Annotated* have no annotations, and can avoid a reflection call to Class#getDeclaredAnnotations() that would just result in an empty set. We can also investiate using our own annotation generation (think AnnotationInstanceProvider from deltaspike) and generate annotations that do not involve any reflection, and do not involve JDK wide locks as part of the creation. The JDK reflection impl is full of locks, and even though they are not held for long they still have a noticeable impact on parallel boot performance. *Don't use annotations as part of internal data structures* I already started this with org.jboss.weld.resolution.QualifierInstance as part of the 1.1.7 performance work, which provided a big boost to performance (especially resolving observer methods at runtime) We also need to be careful with these changes that we do not impact the performance of weld-se or weld-servlet, which will be stuck using a reflection based implementation of whatever annotation provider SPI we come up with. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.jboss.org/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira