Re: [PATCH 1/2] libstdc++: Robustify long double std::to_chars testcase [PR98384]
On 24/02/21 12:04 -0500, Patrick Palka via Libstdc++ wrote: On Wed, 24 Feb 2021, Jonathan Wakely wrote: On 23/02/21 11:30 -0500, Patrick Palka via Libstdc++ wrote: > On Mon, 22 Feb 2021, Patrick Palka wrote: > > > This makes the hexadecimal section of the long double std::to_chars > > testcase more robust by avoiding false-negative FAILs due to printf > > using a different leading hex digit than us, and by additionally > > verifying the correctness of the hexadecimal form via round-tripping > > through std::from_chars. > > > > Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this > > look OK for trunk? > > The commit message could explain the issue better, so here's v2 with a > more detailed commit message. > > -- >8 -- > > Subject: [PATCH] libstdc++: Robustify long double std::to_chars testcase > [PR98384] > > The long double std::to_chars testcase currently verifies the > correctness of its output by comparing it to that of printf, so if > there's a mismatch between to_chars and printf, the test FAILs. This > works well for the scientific, fixed and general formatting modes, > because the corresponding printf conversion specifiers (%e, %f and %g) > are rigidly specified. > > But this doesn't work so well for the hex formatting mode because the > corresponding printf conversion specifier %a is more flexibly specified. > For instance, the hexadecimal forms 0x1p+0, 0x2p-1, 0x4p-2 and 0x8p-3 > are all equivalent and valid outputs of the %a specifier for the number > 1. The apparent freedom here is the choice of leading hex digit -- the > standard just requires that the leading hex digit is nonzero for > normalized numbers. > > Currently, our hexadecimal formatting implementation uses 0/1/2 as the > leading hex digit for floating point types that have an implicit leading > mantissa bit which in practice means all supported floating point types > except x86 long double. The latter type has a 64 bit mantissa with an > explicit leading mantissa bit, and for this type our implementation uses > the most significant four bits of the mantissa as leading hex digit. > This seems to be consistent with most printf implementations, but not > all, as PR98384 illustrates. > > In order to avoid false-positive FAILs due to arbitrary disagreement > between to_chars and printf about the choice of leading hex digit, this > patch makes the testcase's verification via printf conditional on the > leading hex digits first agreeing. An additional verification step is > also added: round-tripping the output of to_chars through from_chars > should yield the original value. > > Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this > look OK for trunk? > @@ -50,6 +51,38 @@ namespace detail > void > test01() > { > + // Verifies correctness of the hexadecimal form [BEGIN,END) for VALUE by > + // round-tripping it through from_chars (if available). > + auto verify_via_from_chars = [] (char *begin, char *end, long double > value) { > +#if __cpp_lib_to_chars >= 201611L || _GLIBCXX_HAVE_USELOCALE This is currently going to fail, because we don't actually define __cpp_lib_to_chars yet (we should fix that!) Is checking the feature test macro here useful? We know that floating-point from_chars was committed before to_chars, so if this test is running, we should have from_chars (modulo uselocale being available, so that check is right). Is this to make the test usable for other C++ std::lib implementations? This preprocessor check is copied from from_chars/{5,6}.cc, which I I was going to say "which idiot wrote that then?" and then I realised that the check is fine and I just misread the || as &&. Doh. figured should be appropriate to use here as well. I figured we'd want to adjust each of these checks after we define __cpp_lib_to_chars appropriately anyway (e.g. if __cpp_lib_to_chars is conditioned on uselocale being available, then the three tests should be changed just look at __cpp_lib_to_chars, IIUC). Agreed. The patch is fine for trunk, sorry for the noise. Thanks.
Re: [PATCH 1/2] libstdc++: Robustify long double std::to_chars testcase [PR98384]
On Wed, 24 Feb 2021, Jonathan Wakely wrote: > On 23/02/21 11:30 -0500, Patrick Palka via Libstdc++ wrote: > > On Mon, 22 Feb 2021, Patrick Palka wrote: > > > > > This makes the hexadecimal section of the long double std::to_chars > > > testcase more robust by avoiding false-negative FAILs due to printf > > > using a different leading hex digit than us, and by additionally > > > verifying the correctness of the hexadecimal form via round-tripping > > > through std::from_chars. > > > > > > Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this > > > look OK for trunk? > > > > The commit message could explain the issue better, so here's v2 with a > > more detailed commit message. > > > > -- >8 -- > > > > Subject: [PATCH] libstdc++: Robustify long double std::to_chars testcase > > [PR98384] > > > > The long double std::to_chars testcase currently verifies the > > correctness of its output by comparing it to that of printf, so if > > there's a mismatch between to_chars and printf, the test FAILs. This > > works well for the scientific, fixed and general formatting modes, > > because the corresponding printf conversion specifiers (%e, %f and %g) > > are rigidly specified. > > > > But this doesn't work so well for the hex formatting mode because the > > corresponding printf conversion specifier %a is more flexibly specified. > > For instance, the hexadecimal forms 0x1p+0, 0x2p-1, 0x4p-2 and 0x8p-3 > > are all equivalent and valid outputs of the %a specifier for the number > > 1. The apparent freedom here is the choice of leading hex digit -- the > > standard just requires that the leading hex digit is nonzero for > > normalized numbers. > > > > Currently, our hexadecimal formatting implementation uses 0/1/2 as the > > leading hex digit for floating point types that have an implicit leading > > mantissa bit which in practice means all supported floating point types > > except x86 long double. The latter type has a 64 bit mantissa with an > > explicit leading mantissa bit, and for this type our implementation uses > > the most significant four bits of the mantissa as leading hex digit. > > This seems to be consistent with most printf implementations, but not > > all, as PR98384 illustrates. > > > > In order to avoid false-positive FAILs due to arbitrary disagreement > > between to_chars and printf about the choice of leading hex digit, this > > patch makes the testcase's verification via printf conditional on the > > leading hex digits first agreeing. An additional verification step is > > also added: round-tripping the output of to_chars through from_chars > > should yield the original value. > > > > Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this > > look OK for trunk? > > > @@ -50,6 +51,38 @@ namespace detail > > void > > test01() > > { > > + // Verifies correctness of the hexadecimal form [BEGIN,END) for VALUE by > > + // round-tripping it through from_chars (if available). > > + auto verify_via_from_chars = [] (char *begin, char *end, long double > > value) { > > +#if __cpp_lib_to_chars >= 201611L || _GLIBCXX_HAVE_USELOCALE > > This is currently going to fail, because we don't actually define > __cpp_lib_to_chars yet (we should fix that!) > > Is checking the feature test macro here useful? We know that > floating-point from_chars was committed before to_chars, so if this > test is running, we should have from_chars (modulo uselocale being > available, so that check is right). Is this to make the test usable > for other C++ std::lib implementations? This preprocessor check is copied from from_chars/{5,6}.cc, which I figured should be appropriate to use here as well. I figured we'd want to adjust each of these checks after we define __cpp_lib_to_chars appropriately anyway (e.g. if __cpp_lib_to_chars is conditioned on uselocale being available, then the three tests should be changed just look at __cpp_lib_to_chars, IIUC). > > > +long double roundtrip; > > +auto result = from_chars(begin, end, roundtrip, chars_format::hex); > > +VERIFY( result.ec == errc{} ); > > +VERIFY( result.ptr == end ); > > +VERIFY( roundtrip == value ); > > +#endif > >
Re: [PATCH 1/2] libstdc++: Robustify long double std::to_chars testcase [PR98384]
On 23/02/21 11:30 -0500, Patrick Palka via Libstdc++ wrote: On Mon, 22 Feb 2021, Patrick Palka wrote: This makes the hexadecimal section of the long double std::to_chars testcase more robust by avoiding false-negative FAILs due to printf using a different leading hex digit than us, and by additionally verifying the correctness of the hexadecimal form via round-tripping through std::from_chars. Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this look OK for trunk? The commit message could explain the issue better, so here's v2 with a more detailed commit message. -- >8 -- Subject: [PATCH] libstdc++: Robustify long double std::to_chars testcase [PR98384] The long double std::to_chars testcase currently verifies the correctness of its output by comparing it to that of printf, so if there's a mismatch between to_chars and printf, the test FAILs. This works well for the scientific, fixed and general formatting modes, because the corresponding printf conversion specifiers (%e, %f and %g) are rigidly specified. But this doesn't work so well for the hex formatting mode because the corresponding printf conversion specifier %a is more flexibly specified. For instance, the hexadecimal forms 0x1p+0, 0x2p-1, 0x4p-2 and 0x8p-3 are all equivalent and valid outputs of the %a specifier for the number 1. The apparent freedom here is the choice of leading hex digit -- the standard just requires that the leading hex digit is nonzero for normalized numbers. Currently, our hexadecimal formatting implementation uses 0/1/2 as the leading hex digit for floating point types that have an implicit leading mantissa bit which in practice means all supported floating point types except x86 long double. The latter type has a 64 bit mantissa with an explicit leading mantissa bit, and for this type our implementation uses the most significant four bits of the mantissa as leading hex digit. This seems to be consistent with most printf implementations, but not all, as PR98384 illustrates. In order to avoid false-positive FAILs due to arbitrary disagreement between to_chars and printf about the choice of leading hex digit, this patch makes the testcase's verification via printf conditional on the leading hex digits first agreeing. An additional verification step is also added: round-tripping the output of to_chars through from_chars should yield the original value. Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this look OK for trunk? @@ -50,6 +51,38 @@ namespace detail void test01() { + // Verifies correctness of the hexadecimal form [BEGIN,END) for VALUE by + // round-tripping it through from_chars (if available). + auto verify_via_from_chars = [] (char *begin, char *end, long double value) { +#if __cpp_lib_to_chars >= 201611L || _GLIBCXX_HAVE_USELOCALE This is currently going to fail, because we don't actually define __cpp_lib_to_chars yet (we should fix that!) Is checking the feature test macro here useful? We know that floating-point from_chars was committed before to_chars, so if this test is running, we should have from_chars (modulo uselocale being available, so that check is right). Is this to make the test usable for other C++ std::lib implementations? +long double roundtrip; +auto result = from_chars(begin, end, roundtrip, chars_format::hex); +VERIFY( result.ec == errc{} ); +VERIFY( result.ptr == end ); +VERIFY( roundtrip == value ); +#endif
Re: [PATCH 1/2] libstdc++: Robustify long double std::to_chars testcase [PR98384]
On Mon, 22 Feb 2021, Patrick Palka wrote: > This makes the hexadecimal section of the long double std::to_chars > testcase more robust by avoiding false-negative FAILs due to printf > using a different leading hex digit than us, and by additionally > verifying the correctness of the hexadecimal form via round-tripping > through std::from_chars. > > Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this > look OK for trunk? The commit message could explain the issue better, so here's v2 with a more detailed commit message. -- >8 -- Subject: [PATCH] libstdc++: Robustify long double std::to_chars testcase [PR98384] The long double std::to_chars testcase currently verifies the correctness of its output by comparing it to that of printf, so if there's a mismatch between to_chars and printf, the test FAILs. This works well for the scientific, fixed and general formatting modes, because the corresponding printf conversion specifiers (%e, %f and %g) are rigidly specified. But this doesn't work so well for the hex formatting mode because the corresponding printf conversion specifier %a is more flexibly specified. For instance, the hexadecimal forms 0x1p+0, 0x2p-1, 0x4p-2 and 0x8p-3 are all equivalent and valid outputs of the %a specifier for the number 1. The apparent freedom here is the choice of leading hex digit -- the standard just requires that the leading hex digit is nonzero for normalized numbers. Currently, our hexadecimal formatting implementation uses 0/1/2 as the leading hex digit for floating point types that have an implicit leading mantissa bit which in practice means all supported floating point types except x86 long double. The latter type has a 64 bit mantissa with an explicit leading mantissa bit, and for this type our implementation uses the most significant four bits of the mantissa as leading hex digit. This seems to be consistent with most printf implementations, but not all, as PR98384 illustrates. In order to avoid false-positive FAILs due to arbitrary disagreement between to_chars and printf about the choice of leading hex digit, this patch makes the testcase's verification via printf conditional on the leading hex digits first agreeing. An additional verification step is also added: round-tripping the output of to_chars through from_chars should yield the original value. Tested on x86, x86_64, powerpc64be, powerpc64le and aarch64. Does this look OK for trunk? libstdc++-v3/ChangeLog: PR libstdc++/98384 * testsuite/20_util/to_chars/long_double.cc: Include . (test01): Simplify verifying the nearby values by using a 2-iteration loop and a dedicated output buffer to check that the nearby values are different. Factor out the printf-based verification into a local function, and check that the leading hex digits agree before comparing with the output of printf. Also verify the output by round-tripping it through from_chars. --- .../testsuite/20_util/to_chars/long_double.cc | 73 --- 1 file changed, 47 insertions(+), 26 deletions(-) diff --git a/libstdc++-v3/testsuite/20_util/to_chars/long_double.cc b/libstdc++-v3/testsuite/20_util/to_chars/long_double.cc index 4f72cb65400..da847ae5401 100644 --- a/libstdc++-v3/testsuite/20_util/to_chars/long_double.cc +++ b/libstdc++-v3/testsuite/20_util/to_chars/long_double.cc @@ -26,6 +26,7 @@ #include #include #include +#include #include #include @@ -50,6 +51,38 @@ namespace detail void test01() { + // Verifies correctness of the hexadecimal form [BEGIN,END) for VALUE by + // round-tripping it through from_chars (if available). + auto verify_via_from_chars = [] (char *begin, char *end, long double value) { +#if __cpp_lib_to_chars >= 201611L || _GLIBCXX_HAVE_USELOCALE +long double roundtrip; +auto result = from_chars(begin, end, roundtrip, chars_format::hex); +VERIFY( result.ec == errc{} ); +VERIFY( result.ptr == end ); +VERIFY( roundtrip == value ); +#endif + }; + + // Verifies correctness of the null-terminated hexadecimal form at BEGIN + // for VALUE and PRECISION by comparing it with the output of printf's %La + // conversion specifier. + auto verify_via_printf = [] (char *begin, long double value, + optional precision = nullopt) { +char printf_buffer[1024] = {}; +if (precision.has_value()) + sprintf(printf_buffer, "%.*La", precision.value(), value); +else + sprintf(printf_buffer, "%La", value); + +// Only compare with the output of printf if the leading hex digits agree. +// If the leading hex digit of our form doesn't agree with that of printf, +// then the two forms may still be equivalent (e.g. 1.1p+0 vs 8.8p-3). But +// if the leading hex digits do agree, then we do expect the two forms to be +// the same. +if (printf_buffer[strlen("0x")] == begin[0]) + VERIFY( !strcmp(begin, printf_buffer+strlen("0x")) ); + };