On Wed, 12 May 2021 21:03:27 GMT, Ioi Lam <ik...@openjdk.org> wrote: >> This bug was discovered during the development of >> [JDK-6824466](https://bugs.openjdk.java.net/browse/JDK-6824466): when CDS is >> enabled, if `BootLoader::loadClassOrNull` is called before `initPhase2`, it >> would trigger the initialization of the archived module graph in the wrong >> order. Because of unanticipated nesting of `<clinit>` methods, >> `BootLoader::SERVICES_CATALOG` becomes empty, causing future `ServiceLoader` >> operations to fail. >> >> The fix has 2 parts: >> >> - `BootLoader::loadClassOrNull` no longer calls >> `ClassLoaders::bootLoader()`. This avoids triggering the archived module >> graph initialization. Instead, it makes a direct call to >> `Classloader::findBootstrapClassOrNull()`. We don't actually need a >> `ClassLoader` instance for this call, so I changed >> `Classloader::findBootstrapClassOrNull()` to be a static method. >> - The object returned by `BootLoader::getServicesCatalog()` is now >> maintained inside `jdk.internal.loader.ClassLoaders`. Although not strictly >> required for the fix, this simplifies the initialization of the archived >> module graph. It also makes the logic consistent for the 3 built-in loaders >> (boot/platform/app). >> >> Testing: tiers1-4 in progress. I also verified that the bug reported by >> Mandy is no longer reproducible after I applied this patch on her branch. > > Ioi Lam has updated the pull request incrementally with one additional commit > since the last revision: > > cleaned up ClassLoaders.setArchivedServicesCatalog
Marked as reviewed by alanb (Reviewer). ------------- PR: https://git.openjdk.java.net/jdk/pull/3992
