[RFR] JDK-8156980: Hotspot build doesn't have -std=gnu++98 gcc option

Fri Sep 14 02:00:25 UTC 2018

Hi Andrew,

Thanks for your kind response!

在 2018年09月11日 01:19, Andrew Hughes 写道:
> On Wed, 5 Sep 2018 at 10:52, Leslie Zhai <zhaixiang at loongson.cn> wrote:
>> Hi Andrew,
>>
>> Thanks for your response!
>>
>> I just quote it from here:
>>
>> http://mail.openjdk.java.net/pipermail/build-dev/2016-July/017464.html
>>
>> I spotted that jsig is just a single C file and so doesn't
>> need the -std flag. In fact, it complains about it:
>>
>> Compiling jsig.c (for libjsig.so)
>> ( ( /usr/bin/gcc -fPIC -D_GNU_SOURCE -D_REENTRANT -O2 -pipe -march=core2
>> -std=gnu++98 -m64 -g -DTHIS_FILE='"jsig.c"' -c -MMD -\
>> MF /home/andrew/builder/dev/hotspot/libjsig/objs/jsig.d -o
>> /home/andrew/builder/dev/hotspot/libjsig/objs/jsig.o /home/andrew/p\
>> rojects/openjdk/upstream/dev/hotspot/src/os/linux/vm/jsig.c >
>>   >(/usr/bin/tee /home/andrew/builder/dev/hotspot/libjsig/objs/jsi\
>> g.o.log) 2> >(/usr/bin/tee
>> /home/andrew/builder/dev/hotspot/libjsig/objs/jsig.o.log >&2) || (
>> exitcode=$? && /bin/cp /home/and\
>> rew/builder/dev/hotspot/libjsig/objs/jsig.o.log
>> /home/andrew/builder/dev/make-support/failure-logs/hotspot_libjsig_objs_jsig.o\
>> .log && exit $exitcode ) ) && wait )
>> cc1: warning: command line option '-std=gnu++98' is valid for C++/ObjC++
>> but not for C
>>
>> It is still able to reproducible the warning for OpenJDK8 mips64el with
>> GCC and LLVM toolchains.  And gcc treat it as warning,  so it is often
>> be ignored,  but clang treat it as error.
>> ----- 8< -------- 8< -------- 8< -------- 8< -------- 8< -------- 8< ---
>>
>> Because CXXSTD_CXXFLAG="-std=gnu++98",  such flag might be effect
>> others,  so workaround for LLVM toolchain:
>> diff -r 1a87e769fb7f hotspot/make/linux/makefiles/jsig.make
>> --- a/hotspot/make/linux/makefiles/jsig.make    Mon Sep 03 18:02:35 2018
>> +0800
>> +++ b/hotspot/make/linux/makefiles/jsig.make    Wed Sep 05 15:53:22 2018
>> +0800
>> @@ -54,7 +54,7 @@
>>    $(LIBJSIG): $(JSIGSRCDIR)/jsig.c $(LIBJSIG_MAPFILE)
>>        @echo Making signal interposition lib...
>>        $(QUIETLY) $(CC) $(SYMFLAG) $(ARCHFLAG) $(SHARED_FLAG) $(PICFLAG) \
>> -                         $(LFLAGS_JSIG) $(JSIG_DEBUG_CFLAGS)
>> $(EXTRA_CFLAGS) -o $@ $< -ldl
>> +                         $(LFLAGS_JSIG) $(JSIG_DEBUG_CFLAGS)
>> -fintegrated-as -o $@ $< -ldl
>>    ifeq ($(ENABLE_FULL_DEBUG_SYMBOLS),1)
>>        $(QUIETLY) $(OBJCOPY) --only-keep-debug $@ $(LIBJSIG_DEBUGINFO)
>>        $(QUIETLY) $(OBJCOPY) --add-gnu-debuglink=$(LIBJSIG_DEBUGINFO) $@
>> diff -r 1a87e769fb7f hotspot/make/linux/makefiles/saproc.make
>> --- a/hotspot/make/linux/makefiles/saproc.make    Mon Sep 03 18:02:35
>> 2018 +0800
>> +++ b/hotspot/make/linux/makefiles/saproc.make    Wed Sep 05 15:53:22
>> 2018 +0800
>> @@ -118,7 +118,7 @@
>>                   $(SASRCFILES) \
>>                   $(SA_LFLAGS) \
>>                   $(SA_DEBUG_CFLAGS) \
>> -               $(EXTRA_CFLAGS) \
>> + -fintegrated-as                                      \
>>                   -o $@ \
>>                   -lthread_db
>>    endif
>>
>> Please give me some suggestion,  thanks a lot!
>>
>> --
>> Regards,
>> Leslie Zhai
>>
>>
> Is this the OpenJDK 8 build? It's quite different to the OpenJDK 9+ build I was
> referring to in my e-mail.
Yes, I am compiling OpenJDK 8 with LLVM for mips64el
http://mail.openjdk.java.net/pipermail/build-dev/2018-September/023074.html

>
> There are effectively three different OpenJDK build systems:
>
> 1. OpenJDK 6 & 7 - the old legacy make & Ant build throughtout
> 2. OpenJDK 8 - new configure & make build for most repositories, but invoking
> the old legacy build for HotSpot
> 3. OpenJDK 9 - full  new configure & make build across the board
>
> This leaves OpenJDK 8 as unique and quite confusing. It means when backporting
> build changes from 9+, they essentially have to be duplicated to apply
> to the HotSpot
> part of the build as well.
>
> I'm aware of the build warning due to -std in the backport to 8. It
> was unavoidable
> as the older build only uses CFLAGS, and primarily for C++ code. There is simply
> no variable to add it to instead. CXXFLAGS is used there as a variable for some
> preprocessor directives :
>
> hotspot/make/linux/makefiles/rules.make:CC_COMPILE       = $(CC)
> $(CXXFLAGS) $(CFLAGS)
> hotspot/make/linux/makefiles/rules.make:CXX_COMPILE      = $(CXX)
> $(CXXFLAGS) $(CFLAGS)
>
> Trying to alter this at this stage in 8u's development is really too
> dangerous, considering
> the range of systems relying on being able to build 8u. I'm already a
> little worried by the
> number of build system changes coming into 8u. I would regard the
> build system for
> a legacy release as needing the same stability guarantees as the code itself.
>
> Certainly removing EXTRA_CFLAGS is not appropriate as this code will
> then be built
> without flags passed in by configure and other additions made by
> configure checks.
Sorry for my monkey patch, I will rebase the workaround compile with 
llvm patch.

> Being able to build without that is one thing. Being able to ensure
> there are no regressions
> on all builds is quite another and we've already seen an issue where
> flags for GCC >= 6
> were dropped because of a typo [0].
>
> My suggestion would be that -std is filtered out locally where it is
> known that only
> C code is being compiled (jsig.make by the looks of it). It's not
> pretty, but it's the option
> which is the least risky.
>
> Incidentally, unless things have changed, I wasn't aware that clang
> was even supported
> for OpenJDK 8, which probably accounts for your issues.
>
> [0] http://mail.openjdk.java.net/pipermail/jdk8u-dev/2018-September/007802.html
Thank you so much for your patch!

>
> Thanks,