On 12/07/2016 1:40 PM, Andrew Hughes wrote:
----- Original Message -----
Catching up ...
On 11/07/2016 7:05 AM, Andrew Hughes wrote:
----- Original Message -----
On Jul 8, 2016, at 2:38 AM, Erik Joelsson <erik.joels...@oracle.com>
wrote:
Hello,
This looks good except for the change in toolchain.m4, which looks like
it
might actually break cross compilation by overriding the value for
compiler version for the build compiler using the target compiler. With
this change we basically have:
if cross compilation
TOOLCHAIN_PREPARE_FOR_VERSION_COMPARISONS([BUILD_], [OPENJDK_BUILD_])
else
...
fi
TOOLCHAIN_PREPARE_FOR_VERSION_COMPARISONS([], [OPENJDK_BUILD_])
The problem you are trying to solve is that in the case of not cross
compilation, the TOOLCHAIN_PREPARE_FOR_VERSION_COMPARISONS macro wasn't
called with "OPENJDK_BUILD_". Kim's suggested patch was to add the call
in
the else clause.
Good catch! I totally missed that when reviewing.
Yes, good catch! The indenting on the bug report must have confused me.
Fixed version:
http://cr.openjdk.java.net/~andrew/8156980/webrev.06/root/
http://cr.openjdk.java.net/~andrew/8156980/webrev.06/hotspot
I was glad to see you realized that parts of the hotspot build (libjsig,
libdtrace) only deal with C programs not C++ and so don't want and
should not have a C++ setting passed. But that then begs the question in
the main build in flags.m4 why we have:
$2JVM_CFLAGS="${$2JVM_CFLAGS} ${$2CXXSTD_CXXFLAG}"
which again adds a C++ related flags to what should be the C compilation
flags ??
You'd think so, but for some reason, the HotSpot build has always called the
C++ compiler flags 'CFLAGS'. There is no JVM_CXXFLAGS.
Ah now I recall - this is the CPP -> CXX mess. CFLAGS meant
compiler-flags; CPPFLAGS meant pre-processor-flags. People thought CPP
meant C++ and changed to CXX.
The same behaviour was true with the old build; we had to pass these options
down to HotSpot in EXTRA_CFLAGS.
JVM_CFLAGS is the main variable used in the HotSpot build (see
make/lib/CompileJvm.gmk).
The only C files in the HotSpot tree are for libsaproc, jsig and DTrace,
all of which are handled by separate makefiles. The only one to use JVM_CFLAGS
is
make/lib/CompileDtracePostJvm.gmk, where it is used for a single C++ file,
generateJvmOffsets.cpp:
generateJvmOffsets.cpp_CXXFLAGS := $(JVM_CFLAGS) -mt -xnolib
-norunpath, \
generateJvmOffsetsMain.c_CFLAGS := -library=%none -mt -m64 -norunpath
-z nodefs, \
Thus, it seems JVM_CFLAGS is JVM_CXXFLAGS in all but name.
Thanks for stepping through that.
David
If -std=gnu++98 does get passed to the C compiler, it produces a warning:
cc1: warning: command line option '-std=gnu++98' is valid for C++/ObjC++ but
not for C
I don't see this at all for this patch on OpenJDK 9. It does appear in the
backport
to 8u [0], because the aforementioned C code in the old HotSpot build does use
the
same CFLAGS there.
Thanks,
David
-----
HotSpot webrev is unchanged.
Thanks,
[0] http://mail.openjdk.java.net/pipermail/jdk8u-dev/2016-July/005690.html
