RFR: JDK-8214063: OpenJDK will not build on AIX while using the xlc 13.1 compiler

Wed Nov 21 14:14:05 UTC 2018

On Wed, Nov 21, 2018 at 12:44 PM Magnus Ihse Bursie
<magnus.ihse.bursie at oracle.com> wrote:
>
>
>
> On 2018-11-20 18:50, Volker Simonis wrote:
> > On Tue, Nov 20, 2018 at 6:15 PM Thomas Stüfe <thomas.stuefe at gmail.com> wrote:
> >> On Tue, Nov 20, 2018 at 6:12 PM Adam Farley8 <adam.farley at uk.ibm.com> wrote:
> >>> Heya Tom,
> >>>
> >>> "In JDK11 and JDK12, source files are compiled with -qvisibility=hidden
> >>> when using xlc version other than 12.1. That doesn't seem to play well
> >>> with link option -bexpall. "
> >>>
> >>> Found that buried in one of the associated Git issues. It appears that
> >>> it's OpenJDK's use of that option that's causing the problem, though
> >>> I couldn't speculate as to why it was added in the first place.
> >>>
> >>> I see this has also been noted in https://bugs.openjdk.java.net/browse/JDK-8204541
> >>>
> >>> Does that answer your question?
> >>>
> >> Yes, Thank you. Odd. Will have to do archeology on that one.
> >>
> > No I begin to understand the problem as well :)
> >
> > It was actually change "8202322: AIX: symbol visibility flags not
> > support on xlc 12.1" [1] which introduced "-qvisibility=hidden" for
> > XLC version not equal to 12.1. That's kind of a weak check and I
> > suppose nobody has ever tested this change with an XLC version other
> > than 12.1 (until you came along :). Maybe that check should be a more
> > precisly check for >= 13.1 (but I know such version checks are hard to
> > do in Makefile syntax)?
> In configure (where, ideally, all version checks should be made),
> there's the TOOLCHAIN_CHECK_COMPILER_VERSION function, which supports
> #   IF_AT_LEAST:   block to run if the compiler is at least this version
> (>=)
> #   IF_OLDER_THAN:   block to run if the compiler is older than this
> version (<)
> for normal, dot-separated version number schemes.
>

Thanks for the pointer. I was looking for this functionality yesterday
but couldn't find it :)

> /Magnus
>
> >
> > The thing I don't understand about your patch (the changes in
> > "jni_md.h" look good although I haven't tested them) is why you need
> > the extra changes in NativeImageBuffer.cpp?
> > "jdk_internal_jimage_NativeImageBuffer.h" is a plain, generated JNI
> > header file. If XLC 13 has problems to parse it, there should be much
> > more places which need fixing. I think that part of your change needs
> > a closer evaluation.
> >
> > Thank you and best regards,
> > Volker
> >
> > [1] https://bugs.openjdk.java.net/browse/JDK-8202322
> >
> >> ..Thomas
> >>
> >>> Best Regards
> >>>
> >>> Adam Farley
> >>> IBM Runtimes
> >>>
> >>>
> >>> "Thomas Stüfe" <thomas.stuefe at gmail.com> wrote on 20/11/2018 16:44:07:
> >>>
> >>>> From: "Thomas Stüfe" <thomas.stuefe at gmail.com>
> >>>> To: Adam Farley8 <adam.farley at uk.ibm.com>
> >>>> Cc: Java Core Libs <core-libs-dev at openjdk.java.net>
> >>>> Date: 20/11/2018 16:48
> >>>> Subject: Re: RFR: JDK-8214063: OpenJDK will not build on AIX while
> >>>> using the xlc 13.1 compiler
> >>>>
> >>>> Hi Adam,
> >>>>
> >>>> On Tue, Nov 20, 2018 at 5:12 PM Adam Farley8 <adam.farley at uk.ibm.com> wrote:
> >>>>> Hi Tom,
> >>>>>
> >>>>> Sounds reasonable. I've added a webex to the bug, and here's a
> >>>> link to the bug.
> >>>>> https://urldefense.proofpoint.com/v2/url?
> >>>> u=https-3A__bugs.openjdk.java.net_browse_JDK-2D8214063&d=DwIFaQ&c=jf_iaSHvJObTbx-
> >>>> siA1ZOg&r=P5m8KWUXJf-
> >>>> CeVJc0hDGD9AQ2LkcXDC0PMV9ntVw5Ho&m=z8YYwBXEfN7UtX1suPjpp9CZSHf8v0GrIMK3XGIC9VY&s=81TP9mIjhYD2Hmt8g7p2EHWRZXgiep21hxKLYRU7zIQ&e=
> >>>>> This patch is required because otherwise, when building on AIX
> >>>> using xlc 3.1,
> >>>>> the build fails with this error:
> >>>>>
> >>>>> "Visibility is not allowed on a reference to an imported symbol."
> >>>>>
> >>>>> We believe this is caused by JNIEXPORT and JNIIMPORT not being
> >>>> defined. Without
> >>>>> this, almost no symbols are exported from shared libraries due to use of
> >>>>> -qvisibility=hidden as specified in make/lib/LibCommon.gmk.
> >>>> Yes but what I try to understand is why does this happen now with
> >>>> xlc13? Did xlc change the rules for -qvisibility from v12 to v13 ?
> >>>> That would be quite a break in backward compatibility.
> >>>>
> >>>>> For convenience, here's a summary of the diffs:
> >>>>>
> >>>>> --------------------------------------
> >>>>> File 1 of 2) src/java.base/share/native/libjimage/NativeImageBuffer.cpp
> >>>>>
> >>>>>   #include "osSupport.hpp"
> >>>>>
> >>>>> +#if defined(__xlC__) && (__xlC__ >= 0x0d01)
> >>>>> +/*
> >>>>> + * Version 13.1.3 of xlc seems to have trouble parsing the `__attribute__`
> >>>>> + * annotation in the generated header file we're about to
> >>>> include. Repeating
> >>>>> + * the forward declaration (without the braces) here avoids the diagnostic:
> >>>>> + *   1540-0040 (S) The text "void" is unexpected.  "visibility"
> >>>> may be undeclared or ambiguous.
> >>>>> + */
> >>>>> +extern "C" JNIEXPORT jobject JNICALL
> >>>> Java_jdk_internal_jimage_NativeImageBuffer_getNativeMap(JNIEnv *,
> >>>> jclass, jstring);
> >>>>> +#endif
> >>>>> +
> >>>>> #include "jdk_internal_jimage_NativeImageBuffer.h"
> >>>>> --------------------------------------
> >>>>> File 2 of 2) src/java.base/unix/native/include/jni_md.h
> >>>>>
> >>>>>       #define JNIIMPORT     __attribute__((visibility("default")))
> >>>>>    #endif
> >>>>> +#elif defined(__xlC__) && (__xlC__ >= 0x0d01) /* xlc version 13.1
> >>>> or better required */
> >>>>> +  #define JNIEXPORT       __attribute__((visibility("default")))
> >>>>> +  #define JNIIMPORT       __attribute__((visibility("default")))
> >>>>> #else
> >>>>>    #define JNIEXPORT
> >>>>> --------------------------------------
> >>>>>
> >>>> Thank you.
> >>>>
> >>>> Cheers, Thomas
> >>>>
> >>>>> Best Regards
> >>>>>
> >>>>> Adam Farley
> >>>>> IBM Runtimes
> >>>>>
> >>>>>
> >>>>> "Thomas Stüfe" <thomas.stuefe at gmail.com> wrote on 19/11/2018 18:11:34:
> >>>>>
> >>>>>> From: "Thomas Stüfe" <thomas.stuefe at gmail.com>
> >>>>>> To: Adam Farley8 <adam.farley at uk.ibm.com>
> >>>>>> Cc: Java Core Libs <core-libs-dev at openjdk.java.net>
> >>>>>> Date: 19/11/2018 18:12
> >>>>>> Subject: Re: RFR: JDK-8214063: OpenJDK will not build on AIX while
> >>>>>> using the xlc 13.1 compiler
> >>>>>>
> >>>>>> Hi Adam,
> >>>>>>
> >>>>>> could you please include link to the JBS issue and either link to the
> >>>>>> patch/webrev or link to the webrev, or at the very least the patch
> >>>>>> verbatim?
> >>>>>>
> >>>>>> As for the issue itself: could you please elaborate why this
> >>>> fails with xlc13?
> >>>>>> Also, a real patch would be helpful instead here of yet another link
> >>>>>> to some J9 issue. We are really strapped for manpower and the AIX port
> >>>>>> eats up enough time as it is.
> >>>>>>
> >>>>>> Thanks, Thomas
> >>>>>>
> >>>>>> On Mon, Nov 19, 2018 at 6:28 PM Adam Farley8
> >>>> <adam.farley at uk.ibm.com> wrote:
> >>>>>>> Hi All
> >>>>>>>
> >>>>>>> Both the problem and the solution appear straight-forward enough.
> >>>>>>>
> >>>>>>> Details included in the bug description.
> >>>>>>>
> >>>>>>> Thoughts and opinions welcome.
> >>>>>>>
> >>>>>>> Best Regards
> >>>>>>>
> >>>>>>> Adam Farley
> >>>>>>> IBM Runtimes
> >>>>>>>
> >>>>>>> Unless stated otherwise above:
> >>>>>>> IBM United Kingdom Limited - Registered in England and Wales with number
> >>>>>>> 741598.
> >>>>>>> Registered office: PO Box 41, North Harbour, Portsmouth,
> >>>> Hampshire PO6 3AU
> >>>>> Unless stated otherwise above:
> >>>>> IBM United Kingdom Limited - Registered in England and Wales with
> >>>> number 741598.
> >>>>> Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
> >>> Unless stated otherwise above:
> >>> IBM United Kingdom Limited - Registered in England and Wales with number 741598.
> >>> Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
>