diff options
| author | Stephan Bergmann <sbergman@redhat.com> | 2022-02-25 13:20:14 +0100 |
|---|---|---|
| committer | Stephan Bergmann <sbergman@redhat.com> | 2022-02-25 22:47:21 +0100 |
| commit | 7025ca508b427fdf5eee7410d7e636adac9f0a94 (patch) | |
| tree | 27e22d4d8ea7e3660e89801d14c605e87f7e81b2 /sal/rtl | |
| parent | b14bb255199e7d6db6ec9155b5d9237cb35fdba7 (diff) | |
Make an -fsanitize=undefined workaround conditional
...that had been introduced with b5cb4935c268f12e63b61e035b455b0a59e67aa2 "Work
around undef conversion of large double to float" but should no longer be
necessary with
<https://github.com/llvm/llvm-project/commit/9e52c43090f8cd980167bbd2719878ae36bcf6b5>
"Treat the range of representable values of floating-point types as [-inf, +inf]
not as [-max, +max]" added towards Clang 9.
Thanks to Mike Kaganski for pointing me at this old code and at Richard Smith's
comment at
<https://cplusplusmusings.wordpress.com/2013/03/26/testing-libc-with-fsanitizeundefined/>.
Change-Id: I8ecf115fcf6b1ebf621cb4567f8d31ac9b10ef1c
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/130531
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Diffstat (limited to 'sal/rtl')
| -rw-r--r-- | sal/rtl/ustring.cxx | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/sal/rtl/ustring.cxx b/sal/rtl/ustring.cxx index aa3a7263d1b7..1f81dc3fd902 100644 --- a/sal/rtl/ustring.cxx +++ b/sal/rtl/ustring.cxx @@ -25,6 +25,7 @@ #include <stdexcept> #include <string> +#include <config_options.h> #include <osl/diagnose.h> #include <osl/interlck.h> #include <osl/mutex.h> @@ -133,14 +134,22 @@ sal_Int32 SAL_CALL rtl_ustr_valueOfDouble(sal_Unicode * pStr, double d) namespace { // Avoid -fsanitize=undefined warning e.g. "runtime error: value 1e+99 is -// outside the range of representable values of type 'float'": +// outside the range of representable values of type 'float'" with Clang prior to +// <https://github.com/llvm/llvm-project/commit/9e52c43090f8cd980167bbd2719878ae36bcf6b5> "Treat the +// range of representable values of floating-point types as [-inf, +inf] not as [-max, +max]" +// (ENABLE_RUNTIME_OPTIMIZATIONS is an approximation for checking whether building is done without +// -fsanitize=undefined): float doubleToFloat(double x) { +#if !defined __clang__ || __clang_major__ >= 9 || ENABLE_RUNTIME_OPTIMIZATIONS + return static_cast<float>(x); +#else return x < -std::numeric_limits<float>::max() ? -std::numeric_limits<float>::infinity() : x > std::numeric_limits<float>::max() ? std::numeric_limits<float>::infinity() : static_cast<float>(x); +#endif } } |