treat time_delta as floating point for multiplication

diff --git a/src/lib/util/calc.c b/src/lib/util/calc.c
index 421342a27aed4658c4dd22b6fa8327807796ddfe..09d38f8ec61b48c579aae9952aad20e55c842a56 100644 (file)
--- a/src/lib/util/calc.c
+++ b/src/lib/util/calc.c
@@ -1813,12 +1813,30 @@ int fr_value_calc_binary_op(TALLOC_CTX *ctx, fr_value_box_t *dst, fr_type_t hint
        if (hint == FR_TYPE_NULL) do {
                /*
                 *      All kinds of special cases :(
+                *
+                *      date1 - date2 --> time_delta
+                *
+                *      time_delta * FOO --> float64, because time_delta is _printed_ as a floating point
+                *      number.  And this is the least surprising thing to do.
                 */
                if ((op == T_SUB) && (a->type == b->type) && (a->type == FR_TYPE_DATE)) {
                        hint = FR_TYPE_TIME_DELTA;
                        break;
                }
 
+               if (op == T_MUL) {
+                       if (a->type == FR_TYPE_TIME_DELTA) {
+                               hint = upcast_op[FR_TYPE_FLOAT64][b->type];
+                               if (hint == FR_TYPE_NULL) hint = upcast_op[b->type][FR_TYPE_FLOAT64];
+
+                       } else if (b->type == FR_TYPE_TIME_DELTA) {
+                               hint = upcast_op[a->type][FR_TYPE_FLOAT64];
+                               if (hint == FR_TYPE_NULL) hint = upcast_op[FR_TYPE_FLOAT64][a->type];
+                       }
+
+                       if (hint != FR_TYPE_NULL) break;
+               }
+
                switch (op) {
                case T_OP_CMP_EQ:
                case T_OP_NE: