python: Require a unit in anyToFrequency and anyToLatency

The anytToFrequency and anyToLatency conversion functions are
currently ambiguous when called without a unit. Fix this by always
requiring a unit.

Change-Id: I5ea94e655f7ca82c0efe70b9f9f7f734fbf711c1
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39435
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>

src/python/m5/util/convert.py

@@ -199,32 +199,40 @@ def toLatency(value):
     return toMetricFloat(value, 'latency', 's')
 
 def anyToLatency(value):
-    """result is a clock period"""
-    try:
-        return 1 / toFrequency(value)
-    except (ValueError, ZeroDivisionError):
-        pass
+    """Convert a magnitude and unit to a clock period."""
 
-    try:
-        return toLatency(value)
-    except ValueError:
-        pass
-
-    raise ValueError("cannot convert '%s' to clock period" % value)
+    magnitude, unit = toNum(value,
+                            target_type='latency',
+                            units=('Hz', 's'),
+                            prefixes=metric_prefixes,
+                            converter=float)
+    if unit == 's':
+        return magnitude
+    elif unit == 'Hz':
+        try:
+            return 1.0 / magnitude
+        except ZeroDivisionError:
+            raise ValueError(f"cannot convert '{value}' to clock period")
+    else:
+        raise ValueError(f"'{value}' needs a valid unit to be unambiguous.")
 
 def anyToFrequency(value):
-    """result is a clock period"""
-    try:
-        return toFrequency(value)
-    except ValueError:
-        pass
+    """Convert a magnitude and unit to a clock frequency."""
 
-    try:
-        return 1 / toLatency(value)
-    except ValueError as ZeroDivisionError:
-        pass
-
-    raise ValueError("cannot convert '%s' to clock period" % value)
+    magnitude, unit = toNum(value,
+                            target_type='frequency',
+                            units=('Hz', 's'),
+                            prefixes=metric_prefixes,
+                            converter=float)
+    if unit == 'Hz':
+        return magnitude
+    elif unit == 's':
+        try:
+            return 1.0 / magnitude
+        except ZeroDivisionError:
+            raise ValueError(f"cannot convert '{value}' to frequency")
+    else:
+        raise ValueError(f"'{value}' needs a valid unit to be unambiguous.")
 
 def toNetworkBandwidth(value):
     return toMetricFloat(value, 'network bandwidth', 'bps')

tests/pyunit/util/test_convert.py

@@ -163,28 +163,18 @@ class ConvertTestSuite(unittest.TestCase):
         self.assertEqual(conv('1kHz'), 1e-3)
 
         self.assertRaises(ValueError, conv, '42k')
-
-    @unittest.expectedFailure
-    def test_anyToLatency_ambiguous(self):
-        # This the behavior of anyToFrequency is currently ambiguous
-        # (and surprising) for unitless quantities. The following
-        # should be true to be consistent with the other conversion
-        # functions, but that isn't currently the case.
-        self.assertEqual(convert.anyToLatency('42'), 42.0)
-
+        self.assertRaises(ValueError, conv, '42')
 
     def test_anyToFrequency(self):
         conv = convert.anyToFrequency
 
-        # This is ambiguous and should probably not be allowed.
-        self.assertEqual(conv('42'), 42.0)
-
         self.assertEqual(conv('42kHz'), 42e3)
 
         self.assertEqual(conv('0.1s'), 10.0)
         self.assertEqual(conv('1ms'), 1000.0)
 
         self.assertRaises(ValueError, conv, '42k')
+        self.assertRaises(ValueError, conv, '42')
 
     def test_toNetworkBandwidth(self):
         conv = convert.toNetworkBandwidth