Fix looks good to me then.
Do you need sponsoring help? Is there a bug number for this yet?
/Erik
On 2020-06-01 01:53, Claes Redestad wrote:
Hi Fedor,
thanks for verifying!
/Claes
On 2020-06-01 10:39, Fedor wrote:
Hi Claes,
The content of classlist files is the same in both cases (before and
after change).
And the change with updated file name $@.classlist => $@.interim as
suggested before.
diff --git a/make/GenerateLinkOptData.gmk b/make/GenerateLinkOptData.gmk
--- a/make/GenerateLinkOptData.gmk
+++ b/make/GenerateLinkOptData.gmk
@@ -69,10 +69,10 @@
-Duser.language=en -Duser.country=US \
-cp $(SUPPORT_OUTPUTDIR)/classlist.jar \
build.tools.classlist.HelloClasslist $(LOG_DEBUG)
- $(GREP) -v HelloClasslist $@.raw >
$(INTERIM_IMAGE_DIR)/lib/classlist
- $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java -Xshare:dump \
+ $(GREP) -v HelloClasslist $@.raw > $@.interim
+ $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java -Xshare:dump
-XX:SharedClassListFile=$@.interim -XX:SharedArchiveFile=$@.jsa \
-Xmx128M -Xms128M $(LOG_INFO)
- $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java
-XX:DumpLoadedClassList=$@.raw \
+ $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java
-XX:DumpLoadedClassList=$@.raw -XX:SharedClassListFile=$@.interim
-XX:SharedArchiveFile=$@.jsa \
-Djava.lang.invoke.MethodHandle.TRACE_RESOLVE=true \
-Duser.language=en -Duser.country=US \
--module-path $(SUPPORT_OUTPUTDIR)/classlist.jar \
Thanks,
Fedor
On 26.05.2020 18:39, Claes Redestad wrote:
Hi,
I think the proposed flags should do the right thing here. It should be
straightforward to verify since the contents of the generated files
should ideally not change at all on a standard build.
/Claes
On 2020-05-26 17:25, Erik Joelsson wrote:
Hello,
On 2020-05-26 08:00, Magnus Ihse Bursie wrote:
On 26/05/2020 11:14 pm, Fedor wrote:
Hello all!
I've tried to crossbuild jdk using current
http://hg.openjdk.java.net/jdk/jdk/ sources and noticed that
build requires write permissions to bootstrap jdk.
Your patch seems to be doing the right thing. I would just like to
have Claes verify that the JVM options you are setting are actually
having the desired effect. We really shouldn't be putting any files
in the INTERIM_IMAGE_DIR as that can possibly be located outside
the build directory. Only thing I would want to change is the
suffix for the file you call $@.classlist, since $@ is already
named "classlist", I would prefer something like $@.tmp or
$@.interim, to signal that this is the interim result between the
two generation steps.
Then just to clarify, it's not the BOOT_JDK, but the BUILD_JDK,
which is a different concept. This can get tricky so it's important
to keep the definitions correct. The BOOT_JDK, which is required to
be of version N-1, or possibly N, is used to build a bootstrap
version of the javac compiler, which we can then run on the
BOOT_JDK to compiler the JDK N java code. We also use it to run
most build tools in the build.
The BUILD_JDK is a concept introduced for cross compiling. In a
native build, certain build steps are required to be performed with
the newly built JDK. This includes running jmod and jlink as well
as generating data used to optimize the JDK image in the jlink step
(which is where you are seeing this issue). For a native build, the
BUILD_JDK is simply the exploded image jdk we just built. When
cross compiling, the default behavior is to just also build enough
of the native parts of the JDK for the build platform and use that.
You may also supply a BUILD_JDK when configuring a cross
compilation build. When doing so you must ensure that it's built
from the exact same sources as the JDK you are cross compiling,
otherwise results are unpredictable.
In our internal build setups, we don't use the external BUILD_JDK
feature, which is why we have not run into the problem you are
reporting here.
/Erik
The problem is it tries to write/rewrite class list into
bootstrap jdk directory.
No, it should write to the INTERIM_IMAGE_DIR.
However, I now see that we replace INTERIM_IMAGE_DIR with
BUILD_JDK in GenerateLinkOptData.gmk if we have an external
build-jdk. I assume that is how you have configured. I also assume
you are not cross-compiling.
I'm not entirely sure about the thinking here. I can't say for
certain that we really should run HelloClasslist if we have an
external build-jdk. I think I need Erik or Claes to chime in.
/Magnus
I would like to suggest the fix below to solve this problem:
diff --git a/make/GenerateLinkOptData.gmk
b/make/GenerateLinkOptData.gmk
--- a/make/GenerateLinkOptData.gmk
+++ b/make/GenerateLinkOptData.gmk
@@ -69,10 +69,10 @@
-Duser.language=en -Duser.country=US \
-cp $(SUPPORT_OUTPUTDIR)/classlist.jar \
build.tools.classlist.HelloClasslist $(LOG_DEBUG)
- $(GREP) -v HelloClasslist $@.raw >
$(INTERIM_IMAGE_DIR)/lib/classlist
- $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java -Xshare:dump \
+ $(GREP) -v HelloClasslist $@.raw > $@.classlist
+ $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java -Xshare:dump
-XX:SharedClassListFile=$@.classlist -XX:SharedArchiveFile=$@.jsa \
-Xmx128M -Xms128M $(LOG_INFO)
- $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java
-XX:DumpLoadedClassList=$@.raw \
+ $(FIXPATH) $(INTERIM_IMAGE_DIR)/bin/java
-XX:DumpLoadedClassList=$@.raw
-XX:SharedClassListFile=$@.classlist -XX:SharedArchiveFile=$@.jsa \
-Djava.lang.invoke.MethodHandle.TRACE_RESOLVE=true \
-Duser.language=en -Duser.country=US \
--module-path $(SUPPORT_OUTPUTDIR)/classlist.jar \
Please correct me in case if it is wrong alias, or I need to
file bug first, or something else.
Best Regards,
Fedor
