Python: Distinguish between Python 2 and 3

Also moves the filtering on `name` to before the big disjunction in
`MkModuleImport`.

python/ql/src/semmle/python/ApiGraphs.qll

@@ -303,15 +303,15 @@ module API {
       MkRoot() or
       /** An abstract representative for imports of the module called `name`. */
       MkModuleImport(string name) {
+        // Ignore the following module name for Python 2, as we alias `__builtin__` to `builtins` elsewhere
+        (name != "__builtin__" or major_version() = 3) and
         (
           imports(_, name)
           or
           // When we `import foo.bar.baz` we want to create API graph nodes also for the prefixes
           // `foo` and `foo.bar`:
           name = any(ImportExpr e | not e.isRelative()).getAnImportedModuleName()
-        ) and
-        // Ignore the following module name, as we alias `__builtin__` to `builtins` elsewhere
-        name != "__builtin__"
+        )
         or
         // The `builtins` module should always be implicitly available
         name = "builtins"
@@ -414,6 +414,7 @@ module API {
       )
       or
       // Ensure the Python 2 `__builtin__` module gets the name of the Python 3 `builtins` module.
+      major_version() = 2 and
       nd = MkModuleImport("builtins") and
       imports(ref, "__builtin__")
       or

python/ql/test/experimental/dataflow/ApiGraphs-py2/options

@@ -0,0 +1 @@
+semmle-extractor-options: --lang=2 --max-import-depth=1

python/ql/test/experimental/dataflow/ApiGraphs-py2/test.py

@@ -0,0 +1,3 @@
+def python2_style():
+    from __builtin__ import open #$ use=moduleImport("builtins").getMember("open")
+    open("hello.txt") #$ use=moduleImport("builtins").getMember("open").getReturn()

python/ql/test/experimental/dataflow/ApiGraphs-py2/use.ql

@@ -0,0 +1,30 @@
+import python
+import semmle.python.dataflow.new.DataFlow
+import TestUtilities.InlineExpectationsTest
+import semmle.python.ApiGraphs
+
+class ApiUseTest extends InlineExpectationsTest {
+  ApiUseTest() { this = "ApiUseTest" }
+
+  override string getARelevantTag() { result = "use" }
+
+  private predicate relevant_node(API::Node a, DataFlow::Node n, Location l) {
+    n = a.getAUse() and l = n.getLocation()
+  }
+
+  override predicate hasActualResult(Location location, string element, string tag, string value) {
+    exists(API::Node a, DataFlow::Node n | relevant_node(a, n, location) |
+      tag = "use" and
+      // Only report the longest path on this line:
+      value =
+        max(API::Node a2, Location l2 |
+          relevant_node(a2, _, l2) and
+          l2.getFile() = location.getFile() and
+          l2.getStartLine() = location.getStartLine()
+        |
+          a2.getPath()
+        ) and
+      element = n.toString()
+    )
+  }
+}

python/ql/test/experimental/dataflow/ApiGraphs/test.py

@@ -136,5 +136,6 @@ def obscured_print():
     p("Can you see me?") #$ use=moduleImport("builtins").getMember("print").getReturn()
 
 def python2_style():
-    from __builtin__ import open #$ use=moduleImport("builtins").getMember("open")
-    open("hello.txt") #$ use=moduleImport("builtins").getMember("open").getReturn()
+    # In Python 3, `__builtin__` has no special meaning.
+    from __builtin__ import open #$ use=moduleImport("__builtin__").getMember("open")
+    open("hello.txt") #$ use=moduleImport("__builtin__").getMember("open").getReturn()