On 2/07/2016 12:07 AM, Erik Joelsson wrote:
Hello,
Again we are hit with a build problem involving -Xmx and only building
server on Windows 32bit. The module summary generation build step is run
on the just built jdk (just like a bootcycle build). Now we face a
similar issue with this summary generation, which uses a fixed setting
of -Xmx1100M. In JDK-8160571 we learned that this is sometimes too much
for the server JVM on Windows 32bit and now we see failures because of it.
My proposed fix is to set the jvm arguments for this to the same as is
now being calculated for the bootcycle jvm args since JDK-8160571.
Bug: https://bugs.openjdk.java.net/browse/JDK-8160722
Patch:
diff -r 31261e7238af common/autoconf/spec.gmk.in
--- a/common/autoconf/spec.gmk.in
+++ b/common/autoconf/spec.gmk.in
@@ -578,7 +578,7 @@
JAVAC_FLAGS?=@JAVAC_FLAGS@
-BUILD_JAVA_FLAGS:=-Xms64M -Xmx1100M
+BUILD_JAVA_FLAGS := @BOOTCYCLE_JVM_ARGS_BIG@
BUILD_JAVA=@FIXPATH@ $(BUILD_JDK)/bin/java $(BUILD_JAVA_FLAGS)
Seems reasonable. Any more potholes on this road we are going to hit?
I'm still worried we may see transient OOM situations due to using
server + G1 by default.
Thanks,
David
# Use ?= as this can be overridden from bootcycle-spec.gmk
/Erik
