Handle overflow in interval parsing

psycopg/typecast_datetime.c

@@ -237,7 +237,19 @@ typecast_PYINTERVAL_cast(const char *str, Py_ssize_t len, PyObject *curs)
 
         case '0': case '1': case '2': case '3': case '4':
         case '5': case '6': case '7': case '8': case '9':
-            v = v * 10 + (*str - '0');
+            {
+                long v1;
+                v1 = v * 10 + (*str - '0');
+                /* detect either a rollover, happening if v is really too short,
+                 * or too big value. On Win where long == int the 2nd check
+                 * is useless. */
+                if (v1 < v || v1 > (long)INT_MAX) {
+                    PyErr_SetString(
+                        PyExc_OverflowError, "interval component too big");
+                    return NULL;
+                }
+                v = v1;
+            }
             if (part == 6) {
                 denom *= 10;
             }
@@ -308,7 +320,7 @@ typecast_PYINTERVAL_cast(const char *str, Py_ssize_t len, PyObject *curs)
             } while (denom < 1000000L);
         }
         else if (denom > 1000000L) {
-            micros = (long)((double)micros * 1000000.0 / denom);
+            micros = (long)round((double)micros / denom * 1000000.0);
         }
     }
 

tests/test_dates.py

@@ -373,6 +373,24 @@ class DatetimeTests(ConnectingTestCase, CommonDatetimeTestsMixin):
         t = self.execute("select '0.0000006'::interval")
         self.assertEqual(total_seconds(t), 1e-6)
 
+    def test_interval_overflow(self):
+        cur = self.conn.cursor()
+        # hack a cursor to receive values too extreme to be represented
+        # but still I want an error, not a random number
+        psycopg2.extensions.register_type(
+            psycopg2.extensions.new_type(
+                psycopg2.STRING.values, 'WAT', psycopg2.extensions.INTERVAL),
+            cur)
+
+        def f(val):
+            cur.execute("select '%s'::text" % val)
+            return cur.fetchone()[0]
+
+        self.assertRaises(OverflowError, f, '100000000000000000:00:00')
+        self.assertRaises(OverflowError, f, '00:100000000000000000:00:00')
+        self.assertRaises(OverflowError, f, '00:00:100000000000000000:00')
+        self.assertRaises(OverflowError, f, '00:00:00.100000000000000000')
+
 
 # Only run the datetime tests if psycopg was compiled with support.
 if not hasattr(psycopg2.extensions, 'PYDATETIME'):