C. Neidahl wrote:
> For C23, GCC has adjusted the value of LDBL_EPSILON when using the IBM
> long double format (PowerPCs) [1]. The value has changed from
> 4.94065645841246544176568792868221e-324L to
> 2.46519032881566189191165176650871e-32L.
Thanks for the report. Let's make sure that Gnulib provides the ISO C 23
compliant value on all PowerPC platforms, regardless of compiler version.
2025-07-07 Bruno Haible <br...@clisp.org>
float-h: Enforce the ISO C 23 compliant value for LDBL_EPSILON.
Reported by Cosima Neidahl <opna2...@protonmail.com> in
<https://lists.gnu.org/archive/html/bug-gnulib/2025-07/msg00021.html>.
* lib/float.in.h (LDBL_EPSILON): Override on all PowerPC platforms.
diff --git a/lib/float.in.h b/lib/float.in.h
index 86af55ad1b..4759e83478 100644
--- a/lib/float.in.h
+++ b/lib/float.in.h
@@ -181,6 +181,21 @@ extern const union gl_long_double_union gl_LDBL_MAX;
# endif
#endif
+/* On PowerPC platforms, 'long double' has a double-double representation.
+ Up to ISO C 17, this was outside the scope of ISO C because it can represent
+ numbers with mantissas of the form 1.<52 bits><many zeroes><52 bits>, such
as
+ 1.0L + 4.94065645841246544176568792868221e-324L = 1 + 2^-1074; see
+ ISO C 17 § 5.2.4.2.2.(3).
+ In ISO C 23, wording has been included that makes this 'long double'
+ representation compliant; see ISO C 23 § 5.2.5.3.3.(8)-(9). In this
setting,
+ numbers with mantissas of the form 1.<52 bits><many zeroes><52 bits> are
+ called "unnormalized". And since LDBL_EPSILON must be normalized (per
+ ISO C 23 § 5.2.5.3.3.(33)), it must be 2^-105. */
+#if defined __powerpc__ && LDBL_MANT_DIG == 106
+# undef LDBL_EPSILON
+# define LDBL_EPSILON 2.46519032881566189191165176650870696773e-32L /* 2^-105
*/
+#endif
+
/* ============================ ISO C11 support ============================ */
/* 'float' properties */
