Refine underflow check in QLocaleData::convertDoubleToFloat()

A string can parse as a non-zero double that's smaller than the
smallest float yet be a faithful representation of the smallest float.
So rather than testing for non-zero doubles less than the smallest
float, test for non-zero doubles that cast to float zero; these
underflow.  This means small values close below the smallest float
shall round up to it, rather than down to zero, requiring a tweak to
an existing test.  Added a test for the boundary case (and tidied the
test data).

Fixes: QTBUG-74833
Change-Id: I4cb30b3c0e54683574b98253505607caaf88fbfb
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

1 files changed, 2 insertions, 4 deletions

diff --git a/src/corelib/tools/qlocale_p.h b/src/corelib/tools/qlocale_p.h
index a96ecf1c1c..7487c9128c 100644
--- a/src/corelib/tools/qlocale_p.h
+++ b/src/corelib/tools/qlocale_p.h
@@ -252,10 +252,8 @@ public:
             const float huge = std::numeric_limits<float>::infinity();
             return d < 0 ? -huge : huge;
         }
-        if (std::fabs(d) >= std::numeric_limits<double>::min() // i.e. d != 0
-            && std::fabs(d) < std::numeric_limits<float>::min()) {
-            // Values smaller than std::numeric_limits<double>::min() have
-            // failed already; match them.
+        if (d != 0 && float(d) == 0) {
+            // Values that underflow double already failed. Match them:
             if (ok != 0)
                 *ok = false;
             return 0;