On Thu, 22 Sep 2022 09:35:47 GMT, Andrew Haley <a...@openjdk.org> wrote:
>> This warning seems to be a false positive, because 1) array "fq" with >> elements from index 0 to "jz" has already been initialized as "fw" at line >> 290 [1], and 2) variable "jz" should be non-negative from the comment at >> line 99 [2]. >> >> Note-1: GCC warning option -Wmaybe-uninitialized is not a new one. Note-2: >> x86-64 build with GCC 12 on Ubuntu 22.04 passed in my local test. >> >> This warning is raised only on GCC 12 + AArch64. I suspect it might be some >> GCC 12 bug, so I reported it to GCC community [3]. >> >> Since it involves third party code, I think it's better to suppress the >> warning by simply disabling this warning option in the makefile. >> >> Testing: Release builds with GCC 9, GCC 11 and GCC 12 passed on Ubuntu >> 22.04/AArch64 system. >> >> [1] >> https://github.com/openjdk/jdk/blob/master/src/java.base/share/native/libfdlibm/k_rem_pio2.c#L290 >> >> [2] >> https://github.com/openjdk/jdk/blob/master/src/java.base/share/native/libfdlibm/k_rem_pio2.c#L99 >> >> [3] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106992 > > make/modules/java.base/lib/CoreLibraries.gmk line 52: > >> 50: CFLAGS_aix := -qfloat=nomaf, \ >> 51: DISABLED_WARNINGS_gcc := sign-compare misleading-indentation \ >> 52: array-bounds maybe-uninitialized, \ > > Something like this would be better: > > #pragma GCC diagnostic push > #pragma GCC diagnostic ignored "-Wmaybe-uninitialized" > write(foo, bar, baz); > #pragma GCC diagnostic pop Thanks for your review. Your mentioned solution should work and we can also enable the pragma only for aarch64 + gcc>=12 condition. I considered such a solution when preparing this patch. But I personally prefer to suppress the waring in the makefile, mainly because fdlibm is 3rd party library and we'd better not changing the source code. I think it is a long-standing policy. ------------- PR: https://git.openjdk.org/jdk/pull/10386
