@Mark: so it means my proposal work, right? ;). Joke apart my proposal was not to have 1 proxy per type but 1 stable name per proxy without changing the number of classes and reuse the existing one if it already exists in the classloader. The only part which can fail is this last one since you can start twice a container in the same classloader and change the meta so the proxy but since the proxy contains the intercepted method markers in its name the reuse is safe.
So I think it is ok even if the naming convention can likely be refined a lot, no? Am I missing something? Romain Manni-Bucau @rmannibucau <https://twitter.com/rmannibucau> | Blog <https://rmannibucau.metawerx.net/> | Old Blog <http://rmannibucau.wordpress.com> | Github <https://github.com/rmannibucau> | LinkedIn <https://www.linkedin.com/in/rmannibucau> | Book <https://www.packtpub.com/application-development/java-ee-8-high-performance> Le lun. 11 nov. 2019 à 20:28, Mark Struberg <strub...@yahoo.de.invalid> a écrit : > Sure! > > Think about having an interface and 3 producer methods. Each with > different proxy configuration. Means 3 different bytecode optimisations for > the same class. > Or having an interface in an EAR and 4 WARs which have their own impls. > Still the proxy gets generated and loaded via the shared EarClassLoader for > class visibility reasons. > > So there are scenarios where you end up with n proxies - each with > different bytecode - for the same class. > > LieGrue, > strub > > > Am 11.11.2019 um 20:16 schrieb Romain Manni-Bucau <rmannibu...@gmail.com > >: > > > > @Mark: not sure what you mean and which case you have in mind. Can you > > precise your answer please and what prevents my patch to work? > > > > Le lun. 11 nov. 2019 à 20:13, Mark Struberg <strub...@yahoo.de.invalid> > a > > écrit : > > > >> no we still need it imo > >> > >>> Am 09.11.2019 um 16:30 schrieb Romain Manni-Bucau < > rmannibu...@gmail.com > >>> : > >>> > >>> Hi everyone, > >>> > >>> I wonder if we can drop the loop > >>> in > org.apache.webbeans.proxy.AbstractProxyFactory#getUnusedProxyClassName > >>> to ensure we have stable names for proxies. > >>> The rational behind that is to: > >>> > >>> 1. Stop creating the same class again and again when we > >>> start/stop/restart/restop/... a container in the same classloader > >>> 2. Ensure we have deterministic names for the same proxy (currently, > when > >>> the name are conflicting, it depends what was executed before, even in > >> the > >>> same app!). > >>> > >>> From a quick review we only need to mark the intercepted methods in the > >>> proxy name to ensure we can reuse the existing class. Therefore a quick > >>> proposal for the proxy name suffix is: > >>> > >>> + protected String getProxySuffix(final Method[] interceptedMethods, > >>> + final Method[] > >> nonInterceptedMethods) { > >>> + return "__i_" + getMethodsId(interceptedMethods); > >>> + } > >>> + > >>> + private String getMethodsId(final Method[] methods, final > >>> Predicate<Method> filter) { > >>> + return Stream.of(methods) > >>> + .filter(filter) > >>> + .map(m -> m.getName() + > >>> + > >>> > >> > Stream.of(m.getParameterTypes()).map(Class::getSimpleName).collect(joining())) > >>> + .collect(joining("_")); > >>> + } > >>> > >>> Full patch is available here: > >>> https://gist.github.com/rmannibucau/6f6cdf8d605387ac8820dcbc76e67909 > >>> > >>> Wdyt? > >>> > >>> If not desired, should I > >>> enrich org.apache.webbeans.spi.DefiningClassService with that feature? > >>> > >>> > >>> Romain Manni-Bucau > >>> @rmannibucau <https://twitter.com/rmannibucau> | Blog > >>> <https://rmannibucau.metawerx.net/> | Old Blog > >>> <http://rmannibucau.wordpress.com> | Github < > >> https://github.com/rmannibucau> | > >>> LinkedIn <https://www.linkedin.com/in/rmannibucau> | Book > >>> < > >> > https://www.packtpub.com/application-development/java-ee-8-high-performance > >>> > >> > >> > >
