Reject an empty label name under UTF-8 validation

With legacy validation off, which is the default, an empty string passes
_validate_labelname. A metric declared with labelnames=[''] then exposes
g{="x"} 1.0, which both bundled parsers reject as invalid labels. Legacy
validation already refuses it, and _validate_metric_name refuses an empty
metric name the same way.

Both parsers call the same function, so a quoted empty label name such as
a{""="x"} 1 is now refused on input as well.

Signed-off-by: Christopher Krah <48338417+chrikrah@users.noreply.github.com>

Author: chrikrah

prometheus_client/validation.py

@@ -78,6 +78,8 @@ def _validate_labelname(l):
     
     This check uses the global legacy validation setting to determine the validation scheme.
     """
+    if not l:
+        raise ValueError('label name cannot be empty')
     if get_legacy_validation():
         if not METRIC_LABEL_NAME_RE.match(l):
             raise ValueError('Invalid label metric name: ' + l)

tests/openmetrics/test_parser.py

@@ -774,6 +774,7 @@ def test_invalid_input(self):
             ('a{a=" # "}1\n# EOF\n'),
             ('a{a=" # ",b=}1\n# EOF\n'),
             ('a{a=" # "b}1\n# EOF\n'),
+            ('a{""="1"} 1\n# EOF\n'),
             # Missing value.
             ('a\n# EOF\n'),
             ('a \n# EOF\n'),

tests/test_core.py

@@ -818,6 +818,21 @@ def test_duplicate_labelnames_error_lists_only_duplicates(self):
                 self.assertEqual(str(error.exception), f'Duplicate label names: {expected!r}')
                 self.assertEqual(list(registry.collect()), [])
 
+    def test_empty_labelname_raises(self):
+        # The exposition would contain {="value"}, which no parser accepts.
+        for metric_type in (Counter, Gauge, Summary, Histogram, Info, Enum):
+            with self.subTest(metric_type=metric_type):
+                registry = CollectorRegistry()
+                kwargs = {'states': ['ready']} if metric_type is Enum else {}
+                with self.assertRaisesRegex(ValueError, '^label name cannot be empty$'):
+                    metric_type('empty_label', 'help', labelnames=[''], registry=registry, **kwargs)
+                self.assertEqual(list(registry.collect()), [])
+
+    def test_empty_exemplar_labelname_raises(self):
+        counter = Counter('empty_exemplar', 'help', registry=self.registry)
+        with self.assertRaisesRegex(ValueError, '^label name cannot be empty$'):
+            counter.inc(exemplar={'': 'abc'})
+
     def test_unique_labelnames_iterator(self):
         # Valid one-shot label iterables still preserve both names.
         counter = Counter('iter_labels', 'help', iter(['a', 'b']), registry=self.registry)

tests/test_parser.py

@@ -217,6 +217,14 @@ def test_multiple_trailing_commas(self):
         text = """# TYPE a counter
 # HELP a help
 a{foo="bar",, } 1
+"""
+        self.assertRaises(ValueError,
+                          lambda: list(text_string_to_metric_families(text)))
+
+    def test_quoted_empty_label_name(self):
+        text = """# TYPE a counter
+# HELP a help
+a{""="bar"} 1
 """
         self.assertRaises(ValueError,
                           lambda: list(text_string_to_metric_families(text)))