Problems in Dist.gmk with --enable-new-hotspot-build

[Erik Joelsson]

I've just pushed your patch.

/Erik

On 2015-04-30 12:20, Volker Simonis wrote:
> On Wed, Apr 29, 2015 at 10:25 PM, Erik Joelsson
> <erik.joelsson at oracle.com> wrote:
>> Hello Volker,
>>
>> Very nice to see you are on top of this!
>>
>> I agree with your assessment of the failure and it is indeed an error to use
>> SetupCopyFiles in that way. Splitting into two calls is the right solution.
>>
> OK, great! Could you then push this fix until I get committer status?
> I've attached a patch to this mail for your convenience.
>
>> I also hope you aren't too scared by the current state of the new makefiles,
>> there is still a lot of work to be done to make them nice and tidy.
>>
> No problem - I've noticed that there are still some rough edges :)
>
> I saw for example, that we use "JVM_EXCLUDE_PATTERNS += x86_32." in
> the solaris-sparcv9 block. I think that's not necessary, right?
>
>> About making you committer, I would love to. I suppose we have to propose
>> and vote about it just like with any other project?
>>
> I think so :)
>
>> /Erik
>>
>>
>> On 2015-04-29 10:07, Volker Simonis wrote:
>>> Hi,
>>>
>>> I'm currently trying to adopt the new HotSpot make system to ppc64.
>>>
>>> Linux/ppc64 seems quite simple and I get up to the COPY_INCLUDE step in
>>> Dist.gmk where it fails with:
>>>
>>> SetupCopyFiles COPY_INCLUDE
>>>    [2] SRC := /usr/work/d046063/OpenJDK/jdk9-build
>>>    [3] DEST :=
>>> /priv/d046063/OpenJDK/output-jdk9-build-dbg/hotspot/dist/include
>>>    [4] FLATTEN := true
>>>    [5] FILES :=
>>> /usr/work/d046063/OpenJDK/jdk9-build/hotspot/src/share/vm/prims/jni.h
>>>
>>> /priv/d046063/OpenJDK/output-jdk9-build-dbg/hotspot/gensrc/jvmtifiles/jvmti.h
>>> /usr/work/d046063/OpenJDK/jdk9-build/hotspot/src/share/vm/code/jvmticmlr.h
>>> /usr/work/d046063/OpenJDK/jdk9-build/hotspot/src/share/vm/services/jmm.h
>>>
>>> gmake[4]: Entering directory
>>> `/net/usr.work/d046063/OpenJDK/jdk9-build/hotspot/makefiles'
>>> /bin/echo Copying files to dist directory
>>> Copying files to dist directory
>>> gmake[4]: *** No rule to make target
>>>
>>> `/usr/work/d046063/OpenJDK/jdk9-build//priv/d046063/OpenJDK/output-jdk9-build-dbg/hotspot/gensrc/jvmtifiles/jvmti.h',
>>> needed by
>>>
>>> `/priv/d046063/OpenJDK/output-jdk9-build-dbg/hotspot/dist/include/jvmti.h'.
>>> Stop.
>>>
>>> But this also fails in the same way if I build on Linux/amd64.
>>>
>>> The problem seems to be in Dist.gmk:
>>>
>>> $(eval $(call SetupCopyFiles, COPY_INCLUDE, \
>>>       SRC := $(SRC_ROOT), \
>>>       DEST := $(DIST_OUTPUTDIR)/include, \
>>>       FLATTEN := true, \
>>>       FILES := $(HOTSPOT_TOPDIR)/src/share/vm/prims/jni.h \
>>>           $(HOTSPOT_OUTPUTDIR)/gensrc/jvmtifiles/jvmti.h \
>>>           $(HOTSPOT_TOPDIR)/src/share/vm/code/jvmticmlr.h \
>>>           $(HOTSPOT_TOPDIR)/src/share/vm/services/jmm.h))
>>>
>>> where we mix files from the source directory (e.g.
>>> $(HOTSPOT_TOPDIR)/src/share/vm/prims/jni.h) with files from the output
>>> directory ($(HOTSPOT_OUTPUTDIR)/gensrc/jvmtifiles/jvmti.h) which doesn't
>>> seem to work.
>>>
>>> SetupCopyFiles has the following comment:
>>>
>>> #   FILES   : List of files to copy with absolute paths, or path relative
>>> to SRC.
>>> #             Must be in SRC.
>>>
>>> And looking at the implementation of SetupCopyFiles I can see that first
>>> SRC is substituted from every FILES argument and later on the remaining
>>> file part is appended to SRC again. However if a file in FILES isn't
>>> located under SRC this can not work and results in a non-existing path
>>> like
>>>
>>> "/usr/work/d046063/OpenJDK/jdk9-build//priv/d046063/OpenJDK/output-jdk9-build-dbg/hotspot/gensrc/jvmtifiles/jvmti.h"
>>> in my example which is a concatenation of SRC with the complete absolute
>>> file path from FILES.
>>>
>>> Maybe you haven't seen this because you always build into a subdirectory
>>> of
>>> the source tree?
>>> Nevertheless I think this should be fixed.
>>>
>>> The easiest fix I can come up is to just divide the COPY_INCLUDE step into
>>> two steps like:
>>>
>>> $(eval $(call SetupCopyFiles, COPY_INCLUDE, \
>>>       SRC := $(SRC_ROOT), \
>>>       DEST := $(DIST_OUTPUTDIR)/include, \
>>>       FLATTEN := true, \
>>>       FILES := $(HOTSPOT_TOPDIR)/src/share/vm/prims/jni.h \
>>>           $(HOTSPOT_TOPDIR)/src/share/vm/code/jvmticmlr.h \
>>>           $(HOTSPOT_TOPDIR)/src/share/vm/services/jmm.h))
>>>
>>> TARGETS += $(COPY_INCLUDE)
>>>
>>> $(eval $(call SetupCopyFiles, COPY_GENERATED_INCLUDE, \
>>>       DEST := $(DIST_OUTPUTDIR)/include, \
>>>       FLATTEN := true, \
>>>       FILES := $(HOTSPOT_OUTPUTDIR)/gensrc/jvmtifiles/jvmti.h))
>>>
>>> TARGETS += $(COPY_GENERATED_INCLUDE)
>>>
>>> With this change I could successfully build HotSpot with the new build
>>> system on Linux/amd64 (and with my other ppc-related changes also on
>>> Linux/ppc64).
>>>
>>> My configure line is:
>>>
>>> /priv/d046063/OpenJDK/output-jdk9-build-dbg$ bash
>>> /usr/work/d046063/OpenJDK/jdk9-build/configure
>>>
>>> --with-boot-jdk=/usr/work/openjdk/nb/linuxppc64/last_known_good/output-jdk8/images/j2sdk-image
>>> --with-jvm-variants=server --with-target-bits=64
>>> --with-debug-level=slowdebug --disable-zip-debug-info
>>> --enable-new-hotspot-build
>>>
>>> Notice that the build and the source directory have no common root!
>>>
>>> My build command is:
>>>
>>> make hotspot-only LOG=debug
>>>
>>> If you'd make me a committer of the build-infr project I'd be happy to
>>> submit this fix along with the up-coming ppc fixes right into the
>>> build-infr repo :)
>>>
>>> Regards,
>>> Volker
>>