On Mon, Jan 12, 2009 at 06:55:17AM -0700, Eric Blake wrote: > According to Eric Blake on 1/9/2009 7:50 PM: > > But so far, I've discovered this > > about -0.0L on Irix 6.2: > > > > With 'cc -n32', LDBL_MIN is a compile-time constant; but with 'gcc', I get > > an error when trying to use it to initialize a global variable (note, > > however, that it can be used inside the body of a method): > > > > foo.c:9: error: initializer element is not constant > > > > > > My conclusions: cc is buggy when representing -0.0L as a global variable, > > but any variant that computes -0.0L during the body of a method works. > > I tested on Irix 6.2, but suspect 6.5 will have similar behavior. With > this patch, both cc and gcc pass all 33 tests created by this: > > gnulib-tool --create-testdir --dir=testdir2 --with-tests ceill floorl > frexpl isnan isnanl ldexpl roundl signbit snprintf-posix sprintf-posix > truncl vasnprintf-posix vasprintf-posix > I've attempted to verify this on IRIX 6.5.30. Using MIPSpro 7.4.4m in std cc mode all tests pass however in c99 mode test-roundl fails: /bin/sh[9]: 6144914 Trace/BPT trap(coredump) FAIL: test-roundl
Running it by hand gives: $ ./test-roundl Trace/BPT/RangeErr/DivZero/Ovflow trap (core dumped) This test was also failing before but only with MIPSpro in c99 mode. With gcc 4.3.1 one test fails: test-ldexpl.c:129: assertion failed /bin/sh[9]: 6244917 Abort(coredump) FAIL: test-ldexpl > whereas without the patch, cc compiled all tests but failed many of them, > and gcc failed to even compile some of the tests. I'm going ahead and > committing it. > I've verified this for IRIX 6.5.30. With MIPSpro 7.4.4m in cc mode it failed 7 tests and in c99 mode it failed 8 tests. With gcc 4.3.1 the build failed in test-ceill.c -tgc
