3 files changed, 89 insertions, 10 deletions
diff --git a/absl/strings/cord.cc b/absl/strings/cord.cc
index 8ecffc4b..70c399c1 100644
--- a/absl/strings/cord.cc
+++ b/absl/strings/cord.cc
@@ -493,10 +493,7 @@ static CordRep* NewSubstring(CordRep* child, size_t offset, size_t length) {
 // --------------------------------------------------------------------
 // Cord::InlineRep functions
 
-// This will trigger LNK2005 in MSVC.
-#ifndef COMPILER_MSVC
 constexpr unsigned char Cord::InlineRep::kMaxInline;
-#endif  // COMPILER_MSVC
 
 inline void Cord::InlineRep::set_data(const char* data, size_t n,
                                       bool nullify_tail) {
diff --git a/absl/strings/internal/str_format/convert_test.cc b/absl/strings/internal/str_format/convert_test.cc
index 488d4cd4..5ee5fbc9 100644
--- a/absl/strings/internal/str_format/convert_test.cc
+++ b/absl/strings/internal/str_format/convert_test.cc
@@ -947,6 +947,88 @@ TEST_F(FormatConvertTest, DoubleRoundA) {
   EXPECT_EQ(format("%.21a", hex_value2), "0x1.081828384858600000000p+3");
 }
 
+TEST_F(FormatConvertTest, LongDoubleRoundA) {
+  if (std::numeric_limits<long double>::digits % 4 != 0) {
+    // This test doesn't really make sense to run on platforms where a long
+    // double has a different mantissa size (mod 4) than Prod, since then the
+    // leading digit will be formatted differently.
+    return;
+  }
+  const NativePrintfTraits &native_traits = VerifyNativeImplementation();
+  std::string s;
+  const auto format = [&](const char *fmt, long double d) -> std::string & {
+    s.clear();
+    FormatArgImpl args[1] = {FormatArgImpl(d)};
+    AppendPack(&s, UntypedFormatSpecImpl(fmt), absl::MakeSpan(args));
+    if (native_traits.hex_float_has_glibc_rounding &&
+        native_traits.hex_float_optimizes_leading_digit_bit_count) {
+      EXPECT_EQ(StrPrint(fmt, d), s);
+    }
+    return s;
+  };
+
+  // 0x8.8p+4
+  const long double on_boundary_even = 136.0;
+  EXPECT_EQ(format("%.0La", on_boundary_even), "0x8p+4");
+  EXPECT_EQ(format("%.1La", on_boundary_even), "0x8.8p+4");
+  EXPECT_EQ(format("%.2La", on_boundary_even), "0x8.80p+4");
+  EXPECT_EQ(format("%.3La", on_boundary_even), "0x8.800p+4");
+  EXPECT_EQ(format("%.4La", on_boundary_even), "0x8.8000p+4");
+  EXPECT_EQ(format("%.5La", on_boundary_even), "0x8.80000p+4");
+  EXPECT_EQ(format("%.6La", on_boundary_even), "0x8.800000p+4");
+
+  // 0x9.8p+4
+  const long double on_boundary_odd = 152.0;
+  EXPECT_EQ(format("%.0La", on_boundary_odd), "0xap+4");
+  EXPECT_EQ(format("%.1La", on_boundary_odd), "0x9.8p+4");
+  EXPECT_EQ(format("%.2La", on_boundary_odd), "0x9.80p+4");
+  EXPECT_EQ(format("%.3La", on_boundary_odd), "0x9.800p+4");
+  EXPECT_EQ(format("%.4La", on_boundary_odd), "0x9.8000p+4");
+  EXPECT_EQ(format("%.5La", on_boundary_odd), "0x9.80000p+4");
+  EXPECT_EQ(format("%.6La", on_boundary_odd), "0x9.800000p+4");
+
+  // 0x8.80001p+24
+  const long double slightly_over = 142606352.0;
+  EXPECT_EQ(format("%.0La", slightly_over), "0x9p+24");
+  EXPECT_EQ(format("%.1La", slightly_over), "0x8.8p+24");
+  EXPECT_EQ(format("%.2La", slightly_over), "0x8.80p+24");
+  EXPECT_EQ(format("%.3La", slightly_over), "0x8.800p+24");
+  EXPECT_EQ(format("%.4La", slightly_over), "0x8.8000p+24");
+  EXPECT_EQ(format("%.5La", slightly_over), "0x8.80001p+24");
+  EXPECT_EQ(format("%.6La", slightly_over), "0x8.800010p+24");
+
+  // 0x8.7ffffp+24
+  const long double slightly_under = 142606320.0;
+  EXPECT_EQ(format("%.0La", slightly_under), "0x8p+24");
+  EXPECT_EQ(format("%.1La", slightly_under), "0x8.8p+24");
+  EXPECT_EQ(format("%.2La", slightly_under), "0x8.80p+24");
+  EXPECT_EQ(format("%.3La", slightly_under), "0x8.800p+24");
+  EXPECT_EQ(format("%.4La", slightly_under), "0x8.8000p+24");
+  EXPECT_EQ(format("%.5La", slightly_under), "0x8.7ffffp+24");
+  EXPECT_EQ(format("%.6La", slightly_under), "0x8.7ffff0p+24");
+  EXPECT_EQ(format("%.7La", slightly_under), "0x8.7ffff00p+24");
+
+  // 0xc.0828384858688000p+128
+  const long double eights = 4094231060438608800781871108094404067328.0;
+  EXPECT_EQ(format("%.0La", eights), "0xcp+128");
+  EXPECT_EQ(format("%.1La", eights), "0xc.1p+128");
+  EXPECT_EQ(format("%.2La", eights), "0xc.08p+128");
+  EXPECT_EQ(format("%.3La", eights), "0xc.083p+128");
+  EXPECT_EQ(format("%.4La", eights), "0xc.0828p+128");
+  EXPECT_EQ(format("%.5La", eights), "0xc.08284p+128");
+  EXPECT_EQ(format("%.6La", eights), "0xc.082838p+128");
+  EXPECT_EQ(format("%.7La", eights), "0xc.0828385p+128");
+  EXPECT_EQ(format("%.8La", eights), "0xc.08283848p+128");
+  EXPECT_EQ(format("%.9La", eights), "0xc.082838486p+128");
+  EXPECT_EQ(format("%.10La", eights), "0xc.0828384858p+128");
+  EXPECT_EQ(format("%.11La", eights), "0xc.08283848587p+128");
+  EXPECT_EQ(format("%.12La", eights), "0xc.082838485868p+128");
+  EXPECT_EQ(format("%.13La", eights), "0xc.0828384858688p+128");
+  EXPECT_EQ(format("%.14La", eights), "0xc.08283848586880p+128");
+  EXPECT_EQ(format("%.15La", eights), "0xc.082838485868800p+128");
+  EXPECT_EQ(format("%.16La", eights), "0xc.0828384858688000p+128");
+}
+
 // We don't actually store the results. This is just to exercise the rest of the
 // machinery.
 struct NullSink {
diff --git a/absl/strings/internal/str_format/float_conversion.cc b/absl/strings/internal/str_format/float_conversion.cc
index 6eb7b9fc..cafa479b 100644
--- a/absl/strings/internal/str_format/float_conversion.cc
+++ b/absl/strings/internal/str_format/float_conversion.cc
@@ -738,7 +738,8 @@ constexpr int HexFloatLeadingDigitSizeInBits() {
 // point that is not followed by 800000..., it disregards the parity and rounds
 // up if > 8 and rounds down if < 8.
 template <typename Int>
-bool HexFloatNeedsRoundUp(Int mantissa, int final_nibble_displayed) {
+bool HexFloatNeedsRoundUp(Int mantissa, int final_nibble_displayed,
+                          uint8_t leading) {
   // If the last nibble (hex digit) to be displayed is the lowest on in the
   // mantissa then that means that we don't have any further nibbles to inform
   // rounding, so don't round.
@@ -755,11 +756,10 @@ bool HexFloatNeedsRoundUp(Int mantissa, int final_nibble_displayed) {
     return mantissa_up_to_rounding_nibble_inclusive > eight;
   }
   // Nibble in question == 8.
-  uint8_t should_round_at_8 =
-      (final_nibble_displayed >= kTotalNibbles)
-          ? true
-          : (GetNibble(mantissa, final_nibble_displayed) % 2 == 1);
-  return should_round_at_8;
+  uint8_t round_if_odd = (final_nibble_displayed == kTotalNibbles)
+                             ? leading
+                             : GetNibble(mantissa, final_nibble_displayed);
+  return round_if_odd % 2 == 1;
 }
 
 // Stores values associated with a Float type needed by the FormatA
@@ -788,7 +788,7 @@ void FormatARound(bool precision_specified, const FormatState &state,
   // Index of the last nibble that we could display given precision.
   int final_nibble_displayed =
       precision_specified ? std::max(0, (kTotalNibbles - state.precision)) : 0;
-  if (HexFloatNeedsRoundUp(*mantissa, final_nibble_displayed)) {
+  if (HexFloatNeedsRoundUp(*mantissa, final_nibble_displayed, *leading)) {
     // Need to round up.
     bool overflow = IncrementNibble(final_nibble_displayed, mantissa);
     *leading += (overflow ? 1 : 0);