This is the first of a series of four changes to support -Wreturn-type in OpenJDK 8u. The -Wreturn-type warning catches instances where control flow exits a non-void function without returning a value. This can combine with compiler optimisations in some cases to cause runtime crashes. The warning is turned on by default in GCC 8 [0], causing the build to fail if it is not explicitly disabled.
The subsequent changes are JDK-8062808 to apply the warning and fix issues in the default build, JDK-8143245 to fix issues in the Zero build and JDK-8197981 to fix one remaining Zero issue on 32-bit platforms only. This change ensures the Zero build is not broken by 8062808 by disabling -Wreturn-type in the Zero makefile. Changes in the backport process: * The G1 changes are dropped as 8u does not contain JDK-8073013, which introduces g1EvacStats.* * I've changed the ifeq ($(USE_CLANG), true) to ifeq ($(JVM_VARIANT_ZEROSHARK), true). There seems to have been a mistake in the original patch which equated LLVM with clang. What is actually referred to by "some version of llvm do not like -Wundef" is the LLVM headers that Shark is built against, not the clang compiler which also uses LLVM. This made its way through to the new HotSpot build in OpenJDK 9 & 10, but is gone as of JDK-8198724 in OpenJDK 11 and later. This backport has been in IcedTea since 3.8.0 (2018-05-29) [1] and our RHEL RPMs since 2018-06-16 without issue. Builds of current 8u-dev on x86_64 with Zero release and Zero slowdebug succeeded. [0] https://gcc.gnu.org/gcc-8/changes.html [1] https://icedtea.classpath.org/bugzilla/show_bug.cgi?id=3548 -- Andrew :) Senior Free Java Software Engineer Red Hat, Inc. (http://www.redhat.com) PGP Key: ed25519/0xCFDA0F9B35964222 (hkp://keys.gnupg.net) Fingerprint = 5132 579D D154 0ED2 3E04 C5A0 CFDA 0F9B 3596 4222 https://keybase.io/gnu_andrew
