diff options
| -rw-r--r-- | src/declarative/qml/v8/qjsvalue_impl_p.h | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/declarative/qml/v8/qjsvalue_impl_p.h b/src/declarative/qml/v8/qjsvalue_impl_p.h index 69ec70beed..cd33859c50 100644 --- a/src/declarative/qml/v8/qjsvalue_impl_p.h +++ b/src/declarative/qml/v8/qjsvalue_impl_p.h @@ -304,7 +304,11 @@ double QJSValuePrivate::toInteger() const return 0; if (qIsInf(result)) return result; - return (result > 0) ? qFloor(result) : -1 * qFloor(-result); + + // Must use floor explicitly rather than qFloor here. On some + // platforms qFloor will cast the value to a single precision float and use + // floorf() which results in test failures. + return (result > 0) ? floor(result) : -1 * floor(-result); } qint32 QJSValuePrivate::toInt32() const @@ -324,7 +328,11 @@ quint32 QJSValuePrivate::toUInt32() const // some of these operation are invoked in toInteger subcall. if (qIsInf(result)) return 0; - return result; + + // The explicit casts are required to avoid undefined behaviour. For example, casting + // a negative double directly to an unsigned int on ARM NEON FPU results in the value + // being set to zero. Casting to a signed int first ensures well defined behaviour. + return (quint32) (qint32) result; } quint16 QJSValuePrivate::toUInt16() const |