On Tue, Jun 7, 2016 at 3:37 PM, Erik Joelsson <erik.joels...@oracle.com> wrote: > Hello Volker, > > On 2016-06-07 15:10, Volker Simonis wrote: >> >> Hi, >> >> I've just realized the the override mechanism from >> JvmOverrideFiles.gmk isn't really 'overriding' options but just >> 'appending' new ones. >> >> So for example: >> >> BUILD_LIBJVM_sharedRuntimeTrig.cpp_CXXFLAGS := -DNO_PCH $(CXX_O_FLAG_NONE) > > That's a bad use of *_<file>_CXXFLAGS. The optimization level can be > specifically overridden on a per file bases using the same notation (or at > least something similar too): > > BUILD_LIBJVM_sharedRuntimeTrig.cpp_OPTIMIZATION := NONE
Great, that's exactly what I was looking for! > > In that case it is a proper override and not just append. Feel free to fix > this. I'm preparing a fix for various problems with the new hotspot build on AIX, so I'd like to leave the general fix for another change. Especially when I see that all current 'overrides' in JvmOverrideFiles.gmk are using _CXXFLAGS when probably most of them should use _OPTIMIZATION. And I think the valid values for OPTIMIZATION should be synchronized with the corresponding option names (currently LOW is mapped to CXX_O_FLAG_NORM and HIGH to CXX_O_FLAG_HI without any good reason). Thank you and best regards, Volker > > For flags in general, I don't see how we could do anything else but just > append. > > /Erik > >> will not replace the CXXFLAGS flags for sharedRuntimeTrig.cpp but only >> append the new ones. >> >> The command line will look like this: >> >> .. -O3 -DNO_PCH -O0 .. >> >> Is this the way it is supposed to work? >> >> I think that's a little problematic because if CXX_O_FLAG_NONE is >> empty (which seems plausible for "doing no optimizations", the >> override mechanism will actually do nothing at all. >> >> Even worse, on AIX we have CXX_O_FLAG_HIGHEST_JVM="-O3 -qhot=level=1 >> -qinline -qinlglue" and CXX_O_FLAG_NORM="-O2". So even if "-O2" >> overrides "-O3", we still have "-qhot=level=1 -qinline -qinlglue" in >> place. >> >> Do we actually have to negate every possible option in >> JvmOverrideFiles.gmk if we want to be sure to have correct results. >> This makes it impossible to use shortcuts like CXX_O_FLAG_NORM in an >> override. >> >> Regards, >> Volker > >
