[baseformat] Fix handling of scalar value if it's considered zero by python

In this case, don't do the check as numpy.array will guess the
value as int64 whatever the original dtype might be. In this case
return 0 as the dtype requested.

beat/backend/python/baseformat.py

@@ -95,10 +95,11 @@ def setup_scalar(formatname, attrname, dtype, value, casting, add_defaults):
         if value is None:  # use the default for the type
             return dtype.type()
         else:
-            if not numpy.can_cast(numpy.array(value).dtype, dtype, casting=casting):
-                raise TypeError("cannot safely cast attribute `%s' on dataformat " \
-                        "`%s' with type `%s' to `%s' without precision loss" % \
-                        (attrname, formatname, numpy.array(value).dtype, dtype))
+            if value: #  zero is classified as int64 which can't be safely casted to uint64
+                if not numpy.can_cast(numpy.array(value).dtype, dtype, casting=casting):
+                    raise TypeError("cannot safely cast attribute `%s' on dataformat " \
+                            "`%s' with type `%s' to `%s' without precision loss" % \
+                            (attrname, formatname, numpy.array(value).dtype, dtype))
             return dtype.type(value)
 
     elif issubclass(dtype, str):  # it is a string

beat/backend/python/test/test_format_valid_cast.py

@@ -111,6 +111,21 @@ def test_single_unsigned_integer64():
 #----------------------------------------------------------
 
 
+def test_zero():
+    value = 0
+    np_types = [numpy.uint64, numpy.uint32, numpy.uint16, numpy.uint8,
+                numpy.int64, numpy.int32, numpy.int16, numpy.int8]
+    dataformats = [('user/single_integer/1', numpy.int32),
+                   ('user/single_unsigned_integer/1', numpy.uint32),
+                   ('user/single_integer64/1', numpy.int64),
+                   ('user/single_unsigned_integer64/1', numpy.uint64)]
+
+    for dataformat, return_type in dataformats:
+        for np_type in np_types:
+            obj = doit(dataformat, 'value', np_type(value))
+            assert numpy.equal(obj.value, value)
+            assert isinstance(obj.value, return_type)
+
 
 #----------------------------------------------------------