add nan-comparison checker for NaN comparisons

**After fix whatsnew to 2.7

6 files changed, 131 insertions, 2 deletions

diff --git a/ChangeLog b/ChangeLog
index 7a7d6f4e0..f3badb564 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -5,6 +5,8 @@ Pylint's ChangeLog
 What's New in Pylint 2.7.0?
 ===========================
 
+* Add `nan-comparison` check for NaN comparisons
+
 * Python 3.6+ is now required.
 
 * Bug fix for empty-comment message line number.
diff --git a/doc/whatsnew/2.7.rst b/doc/whatsnew/2.7.rst
index 0ee5531f5..1f7bab85a 100644
--- a/doc/whatsnew/2.7.rst
+++ b/doc/whatsnew/2.7.rst
@@ -15,6 +15,8 @@ Summary -- Release highlights
 New checkers
 ============
 
+* Add `nan-comparison` check for comparison of NaN values
+
 * Add support to ``ignored-argument-names`` in DocstringParameterChecker and
   adds `useless-param-doc` and `useless-type-doc` messages.
 
diff --git a/pylint/checkers/base.py b/pylint/checkers/base.py
index aeaa85379..cc7e90237 100644
--- a/pylint/checkers/base.py
+++ b/pylint/checkers/base.py
@@ -1661,7 +1661,6 @@ KNOWN_NAME_TYPES = {
     "inlinevar",
 }
 
-
 HUMAN_READABLE_TYPES = {
     "module": "module",
     "const": "constant",
@@ -1724,7 +1723,6 @@ def _create_naming_options():
 
 
 class NameChecker(_BasicChecker):
-
     msgs = {
         "C0102": (
             'Black listed name "%s"',
@@ -2342,6 +2340,12 @@ class ComparisonChecker(_BasicChecker):
             "callable was made, which might suggest that some parenthesis were omitted, "
             "resulting in potential unwanted behaviour.",
         ),
+        "W0177": (
+            "Comparison %s should be %s",
+            "nan-comparison",
+            "Used when an expression is compared to NaN"
+            "values like numpy.NaN and float('nan')",
+        ),
     }
 
     def _check_singleton_comparison(
@@ -2402,6 +2406,52 @@ class ComparisonChecker(_BasicChecker):
             args=(f"'{root_node.as_string()}'", suggestion),
         )
 
+    def _check_nan_comparison(
+        self, left_value, right_value, root_node, checking_for_absence: bool = False
+    ):
+        def _is_float_nan(node):
+            try:
+                if isinstance(node, astroid.Call) and len(node.args) == 1:
+                    if (
+                        node.args[0].value.lower() == "nan"
+                        and node.inferred()[0].pytype() == "builtins.float"
+                    ):
+                        return True
+                return False
+            except AttributeError:
+                return False
+
+        def _is_numpy_nan(node):
+            if isinstance(node, astroid.Attribute) and node.attrname == "NaN":
+                if isinstance(node.expr, astroid.Name):
+                    return node.expr.name in ("numpy", "nmp", "np")
+            return False
+
+        def _is_nan(node) -> bool:
+            return _is_float_nan(node) or _is_numpy_nan(node)
+
+        nan_left = _is_nan(left_value)
+        if not nan_left and not _is_nan(right_value):
+            return
+
+        absence_text = ""
+        if checking_for_absence:
+            absence_text = "not "
+        if nan_left:
+            suggestion = "'{}math.isnan({})'".format(
+                absence_text, right_value.as_string()
+            )
+        else:
+            suggestion = "'{}math.isnan({})'".format(
+                absence_text, left_value.as_string()
+            )
+
+        self.add_message(
+            "nan-comparison",
+            node=root_node,
+            args=("'{}'".format(root_node.as_string()), suggestion),
+        )
+
     def _check_literal_comparison(self, literal, node):
         """Check if we compare to a literal, which is usually what we do not want to do."""
         nodes = (astroid.List, astroid.Tuple, astroid.Dict, astroid.Set)
@@ -2495,6 +2545,11 @@ class ComparisonChecker(_BasicChecker):
             self._check_singleton_comparison(
                 left, right, node, checking_for_absence=operator == "!="
             )
+
+        if operator in ("==", "!=", "is", "is not"):
+            self._check_nan_comparison(
+                left, right, node, checking_for_absence=operator in ("!=", "is not")
+            )
         if operator in ("is", "is not"):
             self._check_literal_comparison(right, node)
 
diff --git a/tests/checkers/unittest_base.py b/tests/checkers/unittest_base.py
index a106e8c2b..3c2e443dc 100644
--- a/tests/checkers/unittest_base.py
+++ b/tests/checkers/unittest_base.py
@@ -445,6 +445,43 @@ class TestComparison(CheckerTestCase):
         with self.assertAddsMessages(message):
             self.checker.visit_compare(node)
 
+        node = astroid.extract_node("foo is float('nan')")
+        message = Message(
+            "nan-comparison",
+            node=node,
+            args=("'foo is float('nan')'", "'math.isnan(foo)'"),
+        )
+        with self.assertAddsMessages(message):
+            self.checker.visit_compare(node)
+
+        node = astroid.extract_node(
+            """
+                                import numpy
+                                foo != numpy.NaN
+                                """
+        )
+        message = Message(
+            "nan-comparison",
+            node=node,
+            args=("'foo != numpy.NaN'", "'not math.isnan(foo)'"),
+        )
+        with self.assertAddsMessages(message):
+            self.checker.visit_compare(node)
+
+        node = astroid.extract_node(
+            """
+                                import numpy as nmp
+                                foo is not nmp.NaN
+                                """
+        )
+        message = Message(
+            "nan-comparison",
+            node=node,
+            args=("'foo is not nmp.NaN'", "'not math.isnan(foo)'"),
+        )
+        with self.assertAddsMessages(message):
+            self.checker.visit_compare(node)
+
         node = astroid.extract_node("True == foo")
         messages = (
             Message("misplaced-comparison-constant", node=node, args=("foo == True",)),
diff --git a/tests/functional/n/nan_comparison_check.py b/tests/functional/n/nan_comparison_check.py
new file mode 100644
index 000000000..db539a11e
--- /dev/null
+++ b/tests/functional/n/nan_comparison_check.py
@@ -0,0 +1,21 @@
+# pylint: disable=missing-docstring, invalid-name, misplaced-comparison-constant,literal-comparison,comparison-with-itself, import-error
+"""Test detection of NaN value comparison."""
+
+import numpy
+x = 42
+a = x is numpy.NaN  # [nan-comparison]
+b = x == numpy.NaN  # [nan-comparison]
+c = x == float('nan')  # [nan-comparison]
+e = numpy.NaN == numpy.NaN  # [nan-comparison]
+f = x is 1
+g = 123 is "123"
+h = numpy.NaN is not x  # [nan-comparison]
+i = numpy.NaN != x  # [nan-comparison]
+
+j = x != numpy.NaN  # [nan-comparison]
+j1 = x != float('nan')  # [nan-comparison]
+assert x == numpy.NaN  # [nan-comparison]
+assert x is not float('nan')  # [nan-comparison]
+if x == numpy.NaN:  # [nan-comparison]
+    pass
+z = bool(x is numpy.NaN)  # [nan-comparison]
diff --git a/tests/functional/n/nan_comparison_check.txt b/tests/functional/n/nan_comparison_check.txt
new file mode 100644
index 000000000..2351ea403
--- /dev/null
+++ b/tests/functional/n/nan_comparison_check.txt
@@ -0,0 +1,12 @@
+nan-comparison:6:4::Comparison 'x is numpy.NaN' should be 'math.isnan(x)'
+nan-comparison:7:4::Comparison 'x == numpy.NaN' should be 'math.isnan(x)'
+nan-comparison:8:4::Comparison 'x == float('nan')' should be 'math.isnan(x)'
+nan-comparison:9:4::Comparison 'numpy.NaN == numpy.NaN' should be 'math.isnan(numpy.NaN)'
+nan-comparison:12:4::Comparison 'numpy.NaN is not x' should be 'not math.isnan(x)'
+nan-comparison:13:4::Comparison 'numpy.NaN != x' should be 'not math.isnan(x)'
+nan-comparison:15:4::Comparison 'x != numpy.NaN' should be 'not math.isnan(x)'
+nan-comparison:16:5::Comparison 'x != float('nan')' should be 'not math.isnan(x)'
+nan-comparison:17:7::Comparison 'x == numpy.NaN' should be 'math.isnan(x)'
+nan-comparison:18:7::Comparison 'x is not float('nan')' should be 'not math.isnan(x)'
+nan-comparison:19:3::Comparison 'x == numpy.NaN' should be 'math.isnan(x)'
+nan-comparison:21:9::Comparison 'x is numpy.NaN' should be 'math.isnan(x)'