Ian Lance Taylor <i...@google.com> writes:
> This patch to libgo, from Peter Collingbourne, changes some of the C
> code to make it easier to compile libgo with a compiler other than GCC.
> Bootstrapped and ran Go testsuite on x86_64-unknown-linux-gnu.
> Committed to mainline.
[...]
> diff -r d73f07d002ef libgo/runtime/print.c
> --- a/libgo/runtime/print.c Tue May 27 15:00:31 2014 -0700
> +++ b/libgo/runtime/print.c Wed May 28 16:09:25 2014 -0700
> @@ -2,6 +2,8 @@
> // Use of this source code is governed by a BSD-style
> // license that can be found in the LICENSE file.
>
> +#include <complex.h>
> +#include <math.h>
> #include <stdarg.h>
> #include "runtime.h"
> #include "array.h"
> @@ -174,13 +176,12 @@
> gwrite("NaN", 3);
> return;
> }
> - i = __builtin_isinf_sign(v);
> - if(i > 0) {
> - gwrite("+Inf", 4);
> - return;
> - }
> - if(i < 0) {
> - gwrite("-Inf", 4);
> + if(isinf(v)) {
> + if(signbit(v)) {
> + gwrite("-Inf", 4);
> + } else {
> + gwrite("+Inf", 4);
> + }
> return;
> }
>
This change broke Solaris Go bootstrap:
/vol/gcc/src/hg/trunk/local/libgo/runtime/print.c: In function
'__go_print_double':
/vol/gcc/src/hg/trunk/local/libgo/runtime/print.c:200:3: error: dereferencing
type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing]
if(signbit(v)) {
^
cc1: all warnings being treated as errors
make: *** [print.lo] Error 1
It's an issue in the gcc version of signbit in <iso/math_c99.h>,
ultimately:
#undef signbit
#if defined(__sparc)
#define signbit(x) __extension__( \
{ __typeof(x) __x_s = (x); \
(int) (*(unsigned *) &__x_s >> 31); })
#elif defined(__i386) || defined(__amd64)
#define signbit(x) __extension__( \
{ __typeof(x) __x_s = (x); \
(sizeof (__x_s) == sizeof (float) ? \
(int) (*(unsigned *) &__x_s >> 31) : \
sizeof (__x_s) == sizeof (double) ? \
(int) (((unsigned *) &__x_s)[1] >> 31) : \
(int) (((unsigned short *) &__x_s)[4] >> 15)); })
#endif
Solaris lacks __signbit{f,,l} in libm, like some other Unices do, and
cannot use __builtin_signbit since that is missing in gcc 3.4 still
bundled with Solaris 10 and 11. I'll probably go for handling this in
fixincludes on mainline, though, where the gcc 3 issue is irrelevant.
Rainer
--
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University
