On 2018-09-25 16:33, Baesken, Matthias wrote:
Hello, it looks like
8211029: Have a common set of enabled warnings for all native libraries
breaks a lot of our Linux based builds.
Our gcc 4.8 misses some of the introduced command line options :
cc1plus: error: unrecognized command line option "-Wno-misleading-indentation"
[-Werror]
cc1plus: error: unrecognized command line option "-Wno-implicit-fallthrough"
[-Werror]
cc1plus: error: unrecognized command line option "-Wno-int-in-bool-context"
[-Werror]
You can just ignore this. It's a red herring.
The way gcc error handling works is that if you pass -Wno-FOOBAR, and
that version of gcc does not recoginze FOOBAR as a warning, it will just
silently ignore it -- as long as there is no other warnings/errors in
the file.
If, on the other hand, something else causes a warning or an error (I'm
not sure if a warning is enough, it might need to be an error or a
warning promoted to error), gcc will then *also* print out the
-Wno-FOOBARs it does not recognize.
Additionally , the added -Werror=switch triggers a LOT of errors on our
porting platforms, e.g. linux ppc64 le :
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value '_Double_valueOf' not handled in switch
[-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value '_forEachRemaining' not handled in switch
[-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value 'ID_LIMIT' not handled in switch [-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value 'LAST_COMPILER_INLINE' not handled in switch
[-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value 'FIRST_MH_SIG_POLY' not handled in switch
[-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value 'FIRST_MH_STATIC' not handled in switch
[-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value 'LAST_MH_SIG_POLY' not handled in switch
[-Werror=switch]
/build_ci_jdk_jdk_linuxppc64le/src/hotspot/cpu/ppc/c1_LIRGenerator_ppc.cpp:719:10:
error: enumeration value 'FIRST_ID' not handled in switch [-Werror=switch]
Could we get rid of the -Werror=switch at least for now ?
Maybe it should be disabled for ppc64 / ppc64le / s390x like it has been
done for zero ?
It is of course possible to disable the warning for certain build
setups, yes. This could be done on a per OS, CPU or gcc version basis
(or a combination thereof).
However, you might want to make certain that it does not make more sense
to actually fix the code. The "switch" warning detected code for s390
that was described as "sneaky" and a "fun source of bugs". You should be
able to use --disable-warnings-as-errors as an intermediate fix to stop
this from breaking the build.
Some warnings has already been fixed, or is in the process of being
fixed, see e.g. JDK-8211145, JDK-8211207 and JDK-8211071. All of which,
I believe, choose to fix the code instead of disabling the warning.
4.13+
4.14 ifeq ($(call check-jvm-feature, zero), true)
4.15- DISABLED_WARNINGS_gcc += return-type
4.16+ DISABLED_WARNINGS_gcc += return-type switch
4.17 endif
Regarding the unrecognized command line options , I suggest to still support
older gcc versions;
what would be the minimal gcc version that supports 8211029 ?
So, I tested this patch with gcc 4.8. However, I don't have access to a
ppc64 nor s390x testing environment. The problem here, I think, is not
in terms of gcc version but on target CPU. The intention was at least
not to change the support for any currently supported compiler, but was
merely in the logistics of me being able to test all kinds of build
environments before the push.
/Magnus
Best regards, Matthias
